Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * linux/kernel/printk.c
4 : *
5 : * Copyright (C) 1991, 1992 Linus Torvalds
6 : *
7 : * Modified to make sys_syslog() more flexible: added commands to
8 : * return the last 4k of kernel messages, regardless of whether
9 : * they've been read or not. Added option to suppress kernel printk's
10 : * to the console. Added hook for sending the console messages
11 : * elsewhere, in preparation for a serial line console (someday).
12 : * Ted Ts'o, 2/11/93.
13 : * Modified for sysctl support, 1/8/97, Chris Horn.
14 : * Fixed SMP synchronization, 08/08/99, Manfred Spraul
15 : * manfred@colorfullife.com
16 : * Rewrote bits to get rid of console_lock
17 : * 01Mar01 Andrew Morton
18 : */
19 :
20 : #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 :
22 : #include <linux/kernel.h>
23 : #include <linux/mm.h>
24 : #include <linux/tty.h>
25 : #include <linux/tty_driver.h>
26 : #include <linux/console.h>
27 : #include <linux/init.h>
28 : #include <linux/jiffies.h>
29 : #include <linux/nmi.h>
30 : #include <linux/module.h>
31 : #include <linux/moduleparam.h>
32 : #include <linux/delay.h>
33 : #include <linux/smp.h>
34 : #include <linux/security.h>
35 : #include <linux/memblock.h>
36 : #include <linux/syscalls.h>
37 : #include <linux/crash_core.h>
38 : #include <linux/ratelimit.h>
39 : #include <linux/kmsg_dump.h>
40 : #include <linux/syslog.h>
41 : #include <linux/cpu.h>
42 : #include <linux/rculist.h>
43 : #include <linux/poll.h>
44 : #include <linux/irq_work.h>
45 : #include <linux/ctype.h>
46 : #include <linux/uio.h>
47 : #include <linux/sched/clock.h>
48 : #include <linux/sched/debug.h>
49 : #include <linux/sched/task_stack.h>
50 :
51 : #include <linux/uaccess.h>
52 : #include <asm/sections.h>
53 :
54 : #include <trace/events/initcall.h>
55 : #define CREATE_TRACE_POINTS
56 : #include <trace/events/printk.h>
57 :
58 : #include "printk_ringbuffer.h"
59 : #include "console_cmdline.h"
60 : #include "braille.h"
61 : #include "internal.h"
62 :
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 : /*
75 : * Low level drivers may need that to know if they can schedule in
76 : * their unblank() callback or not. So let's export it.
77 : */
78 : int oops_in_progress;
79 : EXPORT_SYMBOL(oops_in_progress);
80 :
81 : /*
82 : * console_mutex protects console_list updates and console->flags updates.
83 : * The flags are synchronized only for consoles that are registered, i.e.
84 : * accessible via the console list.
85 : */
86 : static DEFINE_MUTEX(console_mutex);
87 :
88 : /*
89 : * console_sem protects updates to console->seq and console_suspended,
90 : * and also provides serialization for console printing.
91 : */
92 : static DEFINE_SEMAPHORE(console_sem);
93 : HLIST_HEAD(console_list);
94 : EXPORT_SYMBOL_GPL(console_list);
95 : DEFINE_STATIC_SRCU(console_srcu);
96 :
97 : /*
98 : * System may need to suppress printk message under certain
99 : * circumstances, like after kernel panic happens.
100 : */
101 : int __read_mostly suppress_printk;
102 :
103 : /*
104 : * During panic, heavy printk by other CPUs can delay the
105 : * panic and risk deadlock on console resources.
106 : */
107 : static int __read_mostly suppress_panic_printk;
108 :
109 : #ifdef CONFIG_LOCKDEP
110 : static struct lockdep_map console_lock_dep_map = {
111 : .name = "console_lock"
112 : };
113 :
114 : void lockdep_assert_console_list_lock_held(void)
115 : {
116 : lockdep_assert_held(&console_mutex);
117 : }
118 : EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
119 : #endif
120 :
121 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 : bool console_srcu_read_lock_is_held(void)
123 : {
124 : return srcu_read_lock_held(&console_srcu);
125 : }
126 : EXPORT_SYMBOL(console_srcu_read_lock_is_held);
127 : #endif
128 :
129 : enum devkmsg_log_bits {
130 : __DEVKMSG_LOG_BIT_ON = 0,
131 : __DEVKMSG_LOG_BIT_OFF,
132 : __DEVKMSG_LOG_BIT_LOCK,
133 : };
134 :
135 : enum devkmsg_log_masks {
136 : DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
137 : DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
138 : DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
139 : };
140 :
141 : /* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
142 : #define DEVKMSG_LOG_MASK_DEFAULT 0
143 :
144 : static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
145 :
146 0 : static int __control_devkmsg(char *str)
147 : {
148 : size_t len;
149 :
150 0 : if (!str)
151 : return -EINVAL;
152 :
153 0 : len = str_has_prefix(str, "on");
154 0 : if (len) {
155 0 : devkmsg_log = DEVKMSG_LOG_MASK_ON;
156 0 : return len;
157 : }
158 :
159 0 : len = str_has_prefix(str, "off");
160 0 : if (len) {
161 0 : devkmsg_log = DEVKMSG_LOG_MASK_OFF;
162 0 : return len;
163 : }
164 :
165 0 : len = str_has_prefix(str, "ratelimit");
166 0 : if (len) {
167 0 : devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
168 0 : return len;
169 : }
170 :
171 : return -EINVAL;
172 : }
173 :
174 0 : static int __init control_devkmsg(char *str)
175 : {
176 0 : if (__control_devkmsg(str) < 0) {
177 0 : pr_warn("printk.devkmsg: bad option string '%s'\n", str);
178 0 : return 1;
179 : }
180 :
181 : /*
182 : * Set sysctl string accordingly:
183 : */
184 0 : if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
185 0 : strcpy(devkmsg_log_str, "on");
186 0 : else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
187 0 : strcpy(devkmsg_log_str, "off");
188 : /* else "ratelimit" which is set by default. */
189 :
190 : /*
191 : * Sysctl cannot change it anymore. The kernel command line setting of
192 : * this parameter is to force the setting to be permanent throughout the
193 : * runtime of the system. This is a precation measure against userspace
194 : * trying to be a smarta** and attempting to change it up on us.
195 : */
196 0 : devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
197 :
198 0 : return 1;
199 : }
200 : __setup("printk.devkmsg=", control_devkmsg);
201 :
202 : char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
203 : #if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
204 0 : int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
205 : void *buffer, size_t *lenp, loff_t *ppos)
206 : {
207 : char old_str[DEVKMSG_STR_MAX_SIZE];
208 : unsigned int old;
209 : int err;
210 :
211 0 : if (write) {
212 0 : if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
213 : return -EINVAL;
214 :
215 0 : old = devkmsg_log;
216 0 : strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
217 : }
218 :
219 0 : err = proc_dostring(table, write, buffer, lenp, ppos);
220 0 : if (err)
221 : return err;
222 :
223 0 : if (write) {
224 0 : err = __control_devkmsg(devkmsg_log_str);
225 :
226 : /*
227 : * Do not accept an unknown string OR a known string with
228 : * trailing crap...
229 : */
230 0 : if (err < 0 || (err + 1 != *lenp)) {
231 :
232 : /* ... and restore old setting. */
233 0 : devkmsg_log = old;
234 0 : strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
235 :
236 0 : return -EINVAL;
237 : }
238 : }
239 :
240 : return 0;
241 : }
242 : #endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
243 :
244 : /**
245 : * console_list_lock - Lock the console list
246 : *
247 : * For console list or console->flags updates
248 : */
249 0 : void console_list_lock(void)
250 : {
251 : /*
252 : * In unregister_console() and console_force_preferred_locked(),
253 : * synchronize_srcu() is called with the console_list_lock held.
254 : * Therefore it is not allowed that the console_list_lock is taken
255 : * with the srcu_lock held.
256 : *
257 : * Detecting if this context is really in the read-side critical
258 : * section is only possible if the appropriate debug options are
259 : * enabled.
260 : */
261 4 : WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
262 : srcu_read_lock_held(&console_srcu));
263 :
264 4 : mutex_lock(&console_mutex);
265 0 : }
266 : EXPORT_SYMBOL(console_list_lock);
267 :
268 : /**
269 : * console_list_unlock - Unlock the console list
270 : *
271 : * Counterpart to console_list_lock()
272 : */
273 0 : void console_list_unlock(void)
274 : {
275 4 : mutex_unlock(&console_mutex);
276 0 : }
277 : EXPORT_SYMBOL(console_list_unlock);
278 :
279 : /**
280 : * console_srcu_read_lock - Register a new reader for the
281 : * SRCU-protected console list
282 : *
283 : * Use for_each_console_srcu() to iterate the console list
284 : *
285 : * Context: Any context.
286 : * Return: A cookie to pass to console_srcu_read_unlock().
287 : */
288 0 : int console_srcu_read_lock(void)
289 : {
290 1559 : return srcu_read_lock_nmisafe(&console_srcu);
291 : }
292 : EXPORT_SYMBOL(console_srcu_read_lock);
293 :
294 : /**
295 : * console_srcu_read_unlock - Unregister an old reader from
296 : * the SRCU-protected console list
297 : * @cookie: cookie returned from console_srcu_read_lock()
298 : *
299 : * Counterpart to console_srcu_read_lock()
300 : */
301 0 : void console_srcu_read_unlock(int cookie)
302 : {
303 1559 : srcu_read_unlock_nmisafe(&console_srcu, cookie);
304 0 : }
305 : EXPORT_SYMBOL(console_srcu_read_unlock);
306 :
307 : /*
308 : * Helper macros to handle lockdep when locking/unlocking console_sem. We use
309 : * macros instead of functions so that _RET_IP_ contains useful information.
310 : */
311 : #define down_console_sem() do { \
312 : down(&console_sem);\
313 : mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
314 : } while (0)
315 :
316 811 : static int __down_trylock_console_sem(unsigned long ip)
317 : {
318 : int lock_failed;
319 : unsigned long flags;
320 :
321 : /*
322 : * Here and in __up_console_sem() we need to be in safe mode,
323 : * because spindump/WARN/etc from under console ->lock will
324 : * deadlock in printk()->down_trylock_console_sem() otherwise.
325 : */
326 811 : printk_safe_enter_irqsave(flags);
327 811 : lock_failed = down_trylock(&console_sem);
328 1622 : printk_safe_exit_irqrestore(flags);
329 :
330 811 : if (lock_failed)
331 : return 1;
332 : mutex_acquire(&console_lock_dep_map, 0, 1, ip);
333 : return 0;
334 : }
335 : #define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
336 :
337 811 : static void __up_console_sem(unsigned long ip)
338 : {
339 : unsigned long flags;
340 :
341 : mutex_release(&console_lock_dep_map, ip);
342 :
343 811 : printk_safe_enter_irqsave(flags);
344 811 : up(&console_sem);
345 1622 : printk_safe_exit_irqrestore(flags);
346 811 : }
347 : #define up_console_sem() __up_console_sem(_RET_IP_)
348 :
349 : static bool panic_in_progress(void)
350 : {
351 1388 : return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
352 : }
353 :
354 : /*
355 : * This is used for debugging the mess that is the VT code by
356 : * keeping track if we have the console semaphore held. It's
357 : * definitely not the perfect debug tool (we don't know if _WE_
358 : * hold it and are racing, but it helps tracking those weird code
359 : * paths in the console code where we end up in places I want
360 : * locked without the console semaphore held).
361 : */
362 : static int console_locked, console_suspended;
363 :
364 : /*
365 : * Array of consoles built from command line options (console=)
366 : */
367 :
368 : #define MAX_CMDLINECONSOLES 8
369 :
370 : static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
371 :
372 : static int preferred_console = -1;
373 : int console_set_on_cmdline;
374 : EXPORT_SYMBOL(console_set_on_cmdline);
375 :
376 : /* Flag: console code may call schedule() */
377 : static int console_may_schedule;
378 :
379 : enum con_msg_format_flags {
380 : MSG_FORMAT_DEFAULT = 0,
381 : MSG_FORMAT_SYSLOG = (1 << 0),
382 : };
383 :
384 : static int console_msg_format = MSG_FORMAT_DEFAULT;
385 :
386 : /*
387 : * The printk log buffer consists of a sequenced collection of records, each
388 : * containing variable length message text. Every record also contains its
389 : * own meta-data (@info).
390 : *
391 : * Every record meta-data carries the timestamp in microseconds, as well as
392 : * the standard userspace syslog level and syslog facility. The usual kernel
393 : * messages use LOG_KERN; userspace-injected messages always carry a matching
394 : * syslog facility, by default LOG_USER. The origin of every message can be
395 : * reliably determined that way.
396 : *
397 : * The human readable log message of a record is available in @text, the
398 : * length of the message text in @text_len. The stored message is not
399 : * terminated.
400 : *
401 : * Optionally, a record can carry a dictionary of properties (key/value
402 : * pairs), to provide userspace with a machine-readable message context.
403 : *
404 : * Examples for well-defined, commonly used property names are:
405 : * DEVICE=b12:8 device identifier
406 : * b12:8 block dev_t
407 : * c127:3 char dev_t
408 : * n8 netdev ifindex
409 : * +sound:card0 subsystem:devname
410 : * SUBSYSTEM=pci driver-core subsystem name
411 : *
412 : * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
413 : * and values are terminated by a '\0' character.
414 : *
415 : * Example of record values:
416 : * record.text_buf = "it's a line" (unterminated)
417 : * record.info.seq = 56
418 : * record.info.ts_nsec = 36863
419 : * record.info.text_len = 11
420 : * record.info.facility = 0 (LOG_KERN)
421 : * record.info.flags = 0
422 : * record.info.level = 3 (LOG_ERR)
423 : * record.info.caller_id = 299 (task 299)
424 : * record.info.dev_info.subsystem = "pci" (terminated)
425 : * record.info.dev_info.device = "+pci:0000:00:01.0" (terminated)
426 : *
427 : * The 'struct printk_info' buffer must never be directly exported to
428 : * userspace, it is a kernel-private implementation detail that might
429 : * need to be changed in the future, when the requirements change.
430 : *
431 : * /dev/kmsg exports the structured data in the following line format:
432 : * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
433 : *
434 : * Users of the export format should ignore possible additional values
435 : * separated by ',', and find the message after the ';' character.
436 : *
437 : * The optional key/value pairs are attached as continuation lines starting
438 : * with a space character and terminated by a newline. All possible
439 : * non-prinatable characters are escaped in the "\xff" notation.
440 : */
441 :
442 : /* syslog_lock protects syslog_* variables and write access to clear_seq. */
443 : static DEFINE_MUTEX(syslog_lock);
444 :
445 : #ifdef CONFIG_PRINTK
446 : DECLARE_WAIT_QUEUE_HEAD(log_wait);
447 : /* All 3 protected by @syslog_lock. */
448 : /* the next printk record to read by syslog(READ) or /proc/kmsg */
449 : static u64 syslog_seq;
450 : static size_t syslog_partial;
451 : static bool syslog_time;
452 :
453 : struct latched_seq {
454 : seqcount_latch_t latch;
455 : u64 val[2];
456 : };
457 :
458 : /*
459 : * The next printk record to read after the last 'clear' command. There are
460 : * two copies (updated with seqcount_latch) so that reads can locklessly
461 : * access a valid value. Writers are synchronized by @syslog_lock.
462 : */
463 : static struct latched_seq clear_seq = {
464 : .latch = SEQCNT_LATCH_ZERO(clear_seq.latch),
465 : .val[0] = 0,
466 : .val[1] = 0,
467 : };
468 :
469 : #define LOG_LEVEL(v) ((v) & 0x07)
470 : #define LOG_FACILITY(v) ((v) >> 3 & 0xff)
471 :
472 : /* record buffer */
473 : #define LOG_ALIGN __alignof__(unsigned long)
474 : #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
475 : #define LOG_BUF_LEN_MAX (u32)(1 << 31)
476 : static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
477 : static char *log_buf = __log_buf;
478 : static u32 log_buf_len = __LOG_BUF_LEN;
479 :
480 : /*
481 : * Define the average message size. This only affects the number of
482 : * descriptors that will be available. Underestimating is better than
483 : * overestimating (too many available descriptors is better than not enough).
484 : */
485 : #define PRB_AVGBITS 5 /* 32 character average length */
486 :
487 : #if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
488 : #error CONFIG_LOG_BUF_SHIFT value too small.
489 : #endif
490 : _DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
491 : PRB_AVGBITS, &__log_buf[0]);
492 :
493 : static struct printk_ringbuffer printk_rb_dynamic;
494 :
495 : static struct printk_ringbuffer *prb = &printk_rb_static;
496 :
497 : /*
498 : * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
499 : * per_cpu_areas are initialised. This variable is set to true when
500 : * it's safe to access per-CPU data.
501 : */
502 : static bool __printk_percpu_data_ready __ro_after_init;
503 :
504 0 : bool printk_percpu_data_ready(void)
505 : {
506 1622 : return __printk_percpu_data_ready;
507 : }
508 :
509 : /* Must be called under syslog_lock. */
510 : static void latched_seq_write(struct latched_seq *ls, u64 val)
511 : {
512 0 : raw_write_seqcount_latch(&ls->latch);
513 0 : ls->val[0] = val;
514 0 : raw_write_seqcount_latch(&ls->latch);
515 0 : ls->val[1] = val;
516 : }
517 :
518 : /* Can be called from any context. */
519 : static u64 latched_seq_read_nolock(struct latched_seq *ls)
520 : {
521 : unsigned int seq;
522 : unsigned int idx;
523 : u64 val;
524 :
525 : do {
526 0 : seq = raw_read_seqcount_latch(&ls->latch);
527 0 : idx = seq & 0x1;
528 0 : val = ls->val[idx];
529 0 : } while (read_seqcount_latch_retry(&ls->latch, seq));
530 :
531 : return val;
532 : }
533 :
534 : /* Return log buffer address */
535 0 : char *log_buf_addr_get(void)
536 : {
537 0 : return log_buf;
538 : }
539 :
540 : /* Return log buffer size */
541 0 : u32 log_buf_len_get(void)
542 : {
543 0 : return log_buf_len;
544 : }
545 :
546 : /*
547 : * Define how much of the log buffer we could take at maximum. The value
548 : * must be greater than two. Note that only half of the buffer is available
549 : * when the index points to the middle.
550 : */
551 : #define MAX_LOG_TAKE_PART 4
552 : static const char trunc_msg[] = "<truncated>";
553 :
554 0 : static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
555 : {
556 : /*
557 : * The message should not take the whole buffer. Otherwise, it might
558 : * get removed too soon.
559 : */
560 0 : u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
561 :
562 0 : if (*text_len > max_text_len)
563 0 : *text_len = max_text_len;
564 :
565 : /* enable the warning message (if there is room) */
566 0 : *trunc_msg_len = strlen(trunc_msg);
567 0 : if (*text_len >= *trunc_msg_len)
568 0 : *text_len -= *trunc_msg_len;
569 : else
570 0 : *trunc_msg_len = 0;
571 0 : }
572 :
573 : int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
574 :
575 : static int syslog_action_restricted(int type)
576 : {
577 0 : if (dmesg_restrict)
578 : return 1;
579 : /*
580 : * Unless restricted, we allow "read all" and "get buffer size"
581 : * for everybody.
582 : */
583 0 : return type != SYSLOG_ACTION_READ_ALL &&
584 0 : type != SYSLOG_ACTION_SIZE_BUFFER;
585 : }
586 :
587 0 : static int check_syslog_permissions(int type, int source)
588 : {
589 : /*
590 : * If this is from /proc/kmsg and we've already opened it, then we've
591 : * already done the capabilities checks at open time.
592 : */
593 0 : if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
594 : goto ok;
595 :
596 0 : if (syslog_action_restricted(type)) {
597 0 : if (capable(CAP_SYSLOG))
598 : goto ok;
599 : /*
600 : * For historical reasons, accept CAP_SYS_ADMIN too, with
601 : * a warning.
602 : */
603 0 : if (capable(CAP_SYS_ADMIN)) {
604 0 : pr_warn_once("%s (%d): Attempt to access syslog with "
605 : "CAP_SYS_ADMIN but no CAP_SYSLOG "
606 : "(deprecated).\n",
607 : current->comm, task_pid_nr(current));
608 : goto ok;
609 : }
610 : return -EPERM;
611 : }
612 : ok:
613 : return security_syslog(type);
614 : }
615 :
616 : static void append_char(char **pp, char *e, char c)
617 : {
618 0 : if (*pp < e)
619 0 : *(*pp)++ = c;
620 : }
621 :
622 0 : static ssize_t info_print_ext_header(char *buf, size_t size,
623 : struct printk_info *info)
624 : {
625 0 : u64 ts_usec = info->ts_nsec;
626 : char caller[20];
627 : #ifdef CONFIG_PRINTK_CALLER
628 : u32 id = info->caller_id;
629 :
630 : snprintf(caller, sizeof(caller), ",caller=%c%u",
631 : id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
632 : #else
633 0 : caller[0] = '\0';
634 : #endif
635 :
636 0 : do_div(ts_usec, 1000);
637 :
638 0 : return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
639 0 : (info->facility << 3) | info->level, info->seq,
640 0 : ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
641 : }
642 :
643 0 : static ssize_t msg_add_ext_text(char *buf, size_t size,
644 : const char *text, size_t text_len,
645 : unsigned char endc)
646 : {
647 0 : char *p = buf, *e = buf + size;
648 : size_t i;
649 :
650 : /* escape non-printable characters */
651 0 : for (i = 0; i < text_len; i++) {
652 0 : unsigned char c = text[i];
653 :
654 0 : if (c < ' ' || c >= 127 || c == '\\')
655 0 : p += scnprintf(p, e - p, "\\x%02x", c);
656 : else
657 0 : append_char(&p, e, c);
658 : }
659 0 : append_char(&p, e, endc);
660 :
661 0 : return p - buf;
662 : }
663 :
664 0 : static ssize_t msg_add_dict_text(char *buf, size_t size,
665 : const char *key, const char *val)
666 : {
667 0 : size_t val_len = strlen(val);
668 : ssize_t len;
669 :
670 0 : if (!val_len)
671 : return 0;
672 :
673 0 : len = msg_add_ext_text(buf, size, "", 0, ' '); /* dict prefix */
674 0 : len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
675 0 : len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
676 :
677 0 : return len;
678 : }
679 :
680 0 : static ssize_t msg_print_ext_body(char *buf, size_t size,
681 : char *text, size_t text_len,
682 : struct dev_printk_info *dev_info)
683 : {
684 : ssize_t len;
685 :
686 0 : len = msg_add_ext_text(buf, size, text, text_len, '\n');
687 :
688 0 : if (!dev_info)
689 : goto out;
690 :
691 0 : len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
692 0 : dev_info->subsystem);
693 0 : len += msg_add_dict_text(buf + len, size - len, "DEVICE",
694 0 : dev_info->device);
695 : out:
696 0 : return len;
697 : }
698 :
699 : static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
700 : bool is_extended, bool may_supress);
701 :
702 : /* /dev/kmsg - userspace message inject/listen interface */
703 : struct devkmsg_user {
704 : atomic64_t seq;
705 : struct ratelimit_state rs;
706 : struct mutex lock;
707 : struct printk_buffers pbufs;
708 : };
709 :
710 : static __printf(3, 4) __cold
711 0 : int devkmsg_emit(int facility, int level, const char *fmt, ...)
712 : {
713 : va_list args;
714 : int r;
715 :
716 0 : va_start(args, fmt);
717 0 : r = vprintk_emit(facility, level, NULL, fmt, args);
718 0 : va_end(args);
719 :
720 0 : return r;
721 : }
722 :
723 0 : static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
724 : {
725 : char *buf, *line;
726 0 : int level = default_message_loglevel;
727 0 : int facility = 1; /* LOG_USER */
728 0 : struct file *file = iocb->ki_filp;
729 0 : struct devkmsg_user *user = file->private_data;
730 0 : size_t len = iov_iter_count(from);
731 0 : ssize_t ret = len;
732 :
733 0 : if (!user || len > PRINTKRB_RECORD_MAX)
734 : return -EINVAL;
735 :
736 : /* Ignore when user logging is disabled. */
737 0 : if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
738 : return len;
739 :
740 : /* Ratelimit when not explicitly enabled. */
741 0 : if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
742 0 : if (!___ratelimit(&user->rs, current->comm))
743 : return ret;
744 : }
745 :
746 0 : buf = kmalloc(len+1, GFP_KERNEL);
747 0 : if (buf == NULL)
748 : return -ENOMEM;
749 :
750 0 : buf[len] = '\0';
751 0 : if (!copy_from_iter_full(buf, len, from)) {
752 0 : kfree(buf);
753 0 : return -EFAULT;
754 : }
755 :
756 : /*
757 : * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
758 : * the decimal value represents 32bit, the lower 3 bit are the log
759 : * level, the rest are the log facility.
760 : *
761 : * If no prefix or no userspace facility is specified, we
762 : * enforce LOG_USER, to be able to reliably distinguish
763 : * kernel-generated messages from userspace-injected ones.
764 : */
765 0 : line = buf;
766 0 : if (line[0] == '<') {
767 0 : char *endp = NULL;
768 : unsigned int u;
769 :
770 0 : u = simple_strtoul(line + 1, &endp, 10);
771 0 : if (endp && endp[0] == '>') {
772 0 : level = LOG_LEVEL(u);
773 0 : if (LOG_FACILITY(u) != 0)
774 0 : facility = LOG_FACILITY(u);
775 0 : endp++;
776 0 : line = endp;
777 : }
778 : }
779 :
780 0 : devkmsg_emit(facility, level, "%s", line);
781 0 : kfree(buf);
782 0 : return ret;
783 : }
784 :
785 0 : static ssize_t devkmsg_read(struct file *file, char __user *buf,
786 : size_t count, loff_t *ppos)
787 : {
788 0 : struct devkmsg_user *user = file->private_data;
789 0 : char *outbuf = &user->pbufs.outbuf[0];
790 0 : struct printk_message pmsg = {
791 0 : .pbufs = &user->pbufs,
792 : };
793 : ssize_t ret;
794 :
795 0 : if (!user)
796 : return -EBADF;
797 :
798 0 : ret = mutex_lock_interruptible(&user->lock);
799 0 : if (ret)
800 : return ret;
801 :
802 0 : if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
803 0 : if (file->f_flags & O_NONBLOCK) {
804 : ret = -EAGAIN;
805 : goto out;
806 : }
807 :
808 : /*
809 : * Guarantee this task is visible on the waitqueue before
810 : * checking the wake condition.
811 : *
812 : * The full memory barrier within set_current_state() of
813 : * prepare_to_wait_event() pairs with the full memory barrier
814 : * within wq_has_sleeper().
815 : *
816 : * This pairs with __wake_up_klogd:A.
817 : */
818 0 : ret = wait_event_interruptible(log_wait,
819 : printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
820 : false)); /* LMM(devkmsg_read:A) */
821 0 : if (ret)
822 : goto out;
823 : }
824 :
825 0 : if (pmsg.dropped) {
826 : /* our last seen message is gone, return error and reset */
827 0 : atomic64_set(&user->seq, pmsg.seq);
828 0 : ret = -EPIPE;
829 0 : goto out;
830 : }
831 :
832 0 : atomic64_set(&user->seq, pmsg.seq + 1);
833 :
834 0 : if (pmsg.outbuf_len > count) {
835 : ret = -EINVAL;
836 : goto out;
837 : }
838 :
839 0 : if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
840 : ret = -EFAULT;
841 : goto out;
842 : }
843 0 : ret = pmsg.outbuf_len;
844 : out:
845 0 : mutex_unlock(&user->lock);
846 0 : return ret;
847 : }
848 :
849 : /*
850 : * Be careful when modifying this function!!!
851 : *
852 : * Only few operations are supported because the device works only with the
853 : * entire variable length messages (records). Non-standard values are
854 : * returned in the other cases and has been this way for quite some time.
855 : * User space applications might depend on this behavior.
856 : */
857 0 : static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
858 : {
859 0 : struct devkmsg_user *user = file->private_data;
860 0 : loff_t ret = 0;
861 :
862 0 : if (!user)
863 : return -EBADF;
864 0 : if (offset)
865 : return -ESPIPE;
866 :
867 0 : switch (whence) {
868 : case SEEK_SET:
869 : /* the first record */
870 0 : atomic64_set(&user->seq, prb_first_valid_seq(prb));
871 : break;
872 : case SEEK_DATA:
873 : /*
874 : * The first record after the last SYSLOG_ACTION_CLEAR,
875 : * like issued by 'dmesg -c'. Reading /dev/kmsg itself
876 : * changes no global state, and does not clear anything.
877 : */
878 0 : atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
879 : break;
880 : case SEEK_END:
881 : /* after the last record */
882 0 : atomic64_set(&user->seq, prb_next_seq(prb));
883 : break;
884 : default:
885 : ret = -EINVAL;
886 : }
887 : return ret;
888 : }
889 :
890 0 : static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
891 : {
892 0 : struct devkmsg_user *user = file->private_data;
893 : struct printk_info info;
894 0 : __poll_t ret = 0;
895 :
896 0 : if (!user)
897 : return EPOLLERR|EPOLLNVAL;
898 :
899 0 : poll_wait(file, &log_wait, wait);
900 :
901 0 : if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
902 : /* return error when data has vanished underneath us */
903 0 : if (info.seq != atomic64_read(&user->seq))
904 : ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
905 : else
906 0 : ret = EPOLLIN|EPOLLRDNORM;
907 : }
908 :
909 : return ret;
910 : }
911 :
912 0 : static int devkmsg_open(struct inode *inode, struct file *file)
913 : {
914 : struct devkmsg_user *user;
915 : int err;
916 :
917 0 : if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
918 : return -EPERM;
919 :
920 : /* write-only does not need any file context */
921 0 : if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
922 0 : err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
923 : SYSLOG_FROM_READER);
924 0 : if (err)
925 : return err;
926 : }
927 :
928 0 : user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
929 0 : if (!user)
930 : return -ENOMEM;
931 :
932 0 : ratelimit_default_init(&user->rs);
933 0 : ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
934 :
935 0 : mutex_init(&user->lock);
936 :
937 0 : atomic64_set(&user->seq, prb_first_valid_seq(prb));
938 :
939 0 : file->private_data = user;
940 0 : return 0;
941 : }
942 :
943 0 : static int devkmsg_release(struct inode *inode, struct file *file)
944 : {
945 0 : struct devkmsg_user *user = file->private_data;
946 :
947 0 : if (!user)
948 : return 0;
949 :
950 0 : ratelimit_state_exit(&user->rs);
951 :
952 0 : mutex_destroy(&user->lock);
953 0 : kvfree(user);
954 0 : return 0;
955 : }
956 :
957 : const struct file_operations kmsg_fops = {
958 : .open = devkmsg_open,
959 : .read = devkmsg_read,
960 : .write_iter = devkmsg_write,
961 : .llseek = devkmsg_llseek,
962 : .poll = devkmsg_poll,
963 : .release = devkmsg_release,
964 : };
965 :
966 : #ifdef CONFIG_CRASH_CORE
967 : /*
968 : * This appends the listed symbols to /proc/vmcore
969 : *
970 : * /proc/vmcore is used by various utilities, like crash and makedumpfile to
971 : * obtain access to symbols that are otherwise very difficult to locate. These
972 : * symbols are specifically used so that utilities can access and extract the
973 : * dmesg log from a vmcore file after a crash.
974 : */
975 : void log_buf_vmcoreinfo_setup(void)
976 : {
977 : struct dev_printk_info *dev_info = NULL;
978 :
979 : VMCOREINFO_SYMBOL(prb);
980 : VMCOREINFO_SYMBOL(printk_rb_static);
981 : VMCOREINFO_SYMBOL(clear_seq);
982 :
983 : /*
984 : * Export struct size and field offsets. User space tools can
985 : * parse it and detect any changes to structure down the line.
986 : */
987 :
988 : VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
989 : VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
990 : VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
991 : VMCOREINFO_OFFSET(printk_ringbuffer, fail);
992 :
993 : VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
994 : VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
995 : VMCOREINFO_OFFSET(prb_desc_ring, descs);
996 : VMCOREINFO_OFFSET(prb_desc_ring, infos);
997 : VMCOREINFO_OFFSET(prb_desc_ring, head_id);
998 : VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
999 :
1000 : VMCOREINFO_STRUCT_SIZE(prb_desc);
1001 : VMCOREINFO_OFFSET(prb_desc, state_var);
1002 : VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1003 :
1004 : VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1005 : VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1006 : VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1007 :
1008 : VMCOREINFO_STRUCT_SIZE(printk_info);
1009 : VMCOREINFO_OFFSET(printk_info, seq);
1010 : VMCOREINFO_OFFSET(printk_info, ts_nsec);
1011 : VMCOREINFO_OFFSET(printk_info, text_len);
1012 : VMCOREINFO_OFFSET(printk_info, caller_id);
1013 : VMCOREINFO_OFFSET(printk_info, dev_info);
1014 :
1015 : VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1016 : VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1017 : VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1018 : VMCOREINFO_OFFSET(dev_printk_info, device);
1019 : VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1020 :
1021 : VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1022 : VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1023 : VMCOREINFO_OFFSET(prb_data_ring, data);
1024 : VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1025 : VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1026 :
1027 : VMCOREINFO_SIZE(atomic_long_t);
1028 : VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1029 :
1030 : VMCOREINFO_STRUCT_SIZE(latched_seq);
1031 : VMCOREINFO_OFFSET(latched_seq, val);
1032 : }
1033 : #endif
1034 :
1035 : /* requested log_buf_len from kernel cmdline */
1036 : static unsigned long __initdata new_log_buf_len;
1037 :
1038 : /* we practice scaling the ring buffer by powers of 2 */
1039 0 : static void __init log_buf_len_update(u64 size)
1040 : {
1041 0 : if (size > (u64)LOG_BUF_LEN_MAX) {
1042 0 : size = (u64)LOG_BUF_LEN_MAX;
1043 0 : pr_err("log_buf over 2G is not supported.\n");
1044 : }
1045 :
1046 0 : if (size)
1047 0 : size = roundup_pow_of_two(size);
1048 0 : if (size > log_buf_len)
1049 0 : new_log_buf_len = (unsigned long)size;
1050 0 : }
1051 :
1052 : /* save requested log_buf_len since it's too early to process it */
1053 0 : static int __init log_buf_len_setup(char *str)
1054 : {
1055 : u64 size;
1056 :
1057 0 : if (!str)
1058 : return -EINVAL;
1059 :
1060 0 : size = memparse(str, &str);
1061 :
1062 0 : log_buf_len_update(size);
1063 :
1064 0 : return 0;
1065 : }
1066 : early_param("log_buf_len", log_buf_len_setup);
1067 :
1068 : #ifdef CONFIG_SMP
1069 : #define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1070 :
1071 : static void __init log_buf_add_cpu(void)
1072 : {
1073 : unsigned int cpu_extra;
1074 :
1075 : /*
1076 : * archs should set up cpu_possible_bits properly with
1077 : * set_cpu_possible() after setup_arch() but just in
1078 : * case lets ensure this is valid.
1079 : */
1080 : if (num_possible_cpus() == 1)
1081 : return;
1082 :
1083 : cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1084 :
1085 : /* by default this will only continue through for large > 64 CPUs */
1086 : if (cpu_extra <= __LOG_BUF_LEN / 2)
1087 : return;
1088 :
1089 : pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1090 : __LOG_CPU_MAX_BUF_LEN);
1091 : pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1092 : cpu_extra);
1093 : pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1094 :
1095 : log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1096 : }
1097 : #else /* !CONFIG_SMP */
1098 : static inline void log_buf_add_cpu(void) {}
1099 : #endif /* CONFIG_SMP */
1100 :
1101 : static void __init set_percpu_data_ready(void)
1102 : {
1103 1 : __printk_percpu_data_ready = true;
1104 : }
1105 :
1106 0 : static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1107 : struct printk_record *r)
1108 : {
1109 : struct prb_reserved_entry e;
1110 : struct printk_record dest_r;
1111 :
1112 0 : prb_rec_init_wr(&dest_r, r->info->text_len);
1113 :
1114 0 : if (!prb_reserve(&e, rb, &dest_r))
1115 : return 0;
1116 :
1117 0 : memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1118 0 : dest_r.info->text_len = r->info->text_len;
1119 0 : dest_r.info->facility = r->info->facility;
1120 0 : dest_r.info->level = r->info->level;
1121 0 : dest_r.info->flags = r->info->flags;
1122 0 : dest_r.info->ts_nsec = r->info->ts_nsec;
1123 0 : dest_r.info->caller_id = r->info->caller_id;
1124 0 : memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1125 :
1126 0 : prb_final_commit(&e);
1127 :
1128 0 : return prb_record_text_space(&e);
1129 : }
1130 :
1131 : static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1132 :
1133 1 : void __init setup_log_buf(int early)
1134 : {
1135 : struct printk_info *new_infos;
1136 : unsigned int new_descs_count;
1137 : struct prb_desc *new_descs;
1138 : struct printk_info info;
1139 : struct printk_record r;
1140 : unsigned int text_size;
1141 : size_t new_descs_size;
1142 : size_t new_infos_size;
1143 : unsigned long flags;
1144 : char *new_log_buf;
1145 : unsigned int free;
1146 : u64 seq;
1147 :
1148 : /*
1149 : * Some archs call setup_log_buf() multiple times - first is very
1150 : * early, e.g. from setup_arch(), and second - when percpu_areas
1151 : * are initialised.
1152 : */
1153 1 : if (!early)
1154 : set_percpu_data_ready();
1155 :
1156 1 : if (log_buf != __log_buf)
1157 1 : return;
1158 :
1159 : if (!early && !new_log_buf_len)
1160 : log_buf_add_cpu();
1161 :
1162 1 : if (!new_log_buf_len)
1163 : return;
1164 :
1165 0 : new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1166 0 : if (new_descs_count == 0) {
1167 0 : pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1168 0 : return;
1169 : }
1170 :
1171 0 : new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1172 0 : if (unlikely(!new_log_buf)) {
1173 0 : pr_err("log_buf_len: %lu text bytes not available\n",
1174 : new_log_buf_len);
1175 0 : return;
1176 : }
1177 :
1178 0 : new_descs_size = new_descs_count * sizeof(struct prb_desc);
1179 0 : new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1180 0 : if (unlikely(!new_descs)) {
1181 0 : pr_err("log_buf_len: %zu desc bytes not available\n",
1182 : new_descs_size);
1183 0 : goto err_free_log_buf;
1184 : }
1185 :
1186 0 : new_infos_size = new_descs_count * sizeof(struct printk_info);
1187 0 : new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1188 0 : if (unlikely(!new_infos)) {
1189 0 : pr_err("log_buf_len: %zu info bytes not available\n",
1190 : new_infos_size);
1191 : goto err_free_descs;
1192 : }
1193 :
1194 0 : prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1195 :
1196 0 : prb_init(&printk_rb_dynamic,
1197 0 : new_log_buf, ilog2(new_log_buf_len),
1198 0 : new_descs, ilog2(new_descs_count),
1199 : new_infos);
1200 :
1201 0 : local_irq_save(flags);
1202 :
1203 0 : log_buf_len = new_log_buf_len;
1204 0 : log_buf = new_log_buf;
1205 0 : new_log_buf_len = 0;
1206 :
1207 0 : free = __LOG_BUF_LEN;
1208 0 : prb_for_each_record(0, &printk_rb_static, seq, &r) {
1209 0 : text_size = add_to_rb(&printk_rb_dynamic, &r);
1210 0 : if (text_size > free)
1211 : free = 0;
1212 : else
1213 0 : free -= text_size;
1214 : }
1215 :
1216 0 : prb = &printk_rb_dynamic;
1217 :
1218 0 : local_irq_restore(flags);
1219 :
1220 : /*
1221 : * Copy any remaining messages that might have appeared from
1222 : * NMI context after copying but before switching to the
1223 : * dynamic buffer.
1224 : */
1225 0 : prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1226 0 : text_size = add_to_rb(&printk_rb_dynamic, &r);
1227 0 : if (text_size > free)
1228 : free = 0;
1229 : else
1230 0 : free -= text_size;
1231 : }
1232 :
1233 0 : if (seq != prb_next_seq(&printk_rb_static)) {
1234 0 : pr_err("dropped %llu messages\n",
1235 : prb_next_seq(&printk_rb_static) - seq);
1236 : }
1237 :
1238 0 : pr_info("log_buf_len: %u bytes\n", log_buf_len);
1239 0 : pr_info("early log buf free: %u(%u%%)\n",
1240 : free, (free * 100) / __LOG_BUF_LEN);
1241 0 : return;
1242 :
1243 : err_free_descs:
1244 0 : memblock_free(new_descs, new_descs_size);
1245 : err_free_log_buf:
1246 0 : memblock_free(new_log_buf, new_log_buf_len);
1247 : }
1248 :
1249 : static bool __read_mostly ignore_loglevel;
1250 :
1251 0 : static int __init ignore_loglevel_setup(char *str)
1252 : {
1253 0 : ignore_loglevel = true;
1254 0 : pr_info("debug: ignoring loglevel setting.\n");
1255 :
1256 0 : return 0;
1257 : }
1258 :
1259 : early_param("ignore_loglevel", ignore_loglevel_setup);
1260 : module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1261 : MODULE_PARM_DESC(ignore_loglevel,
1262 : "ignore loglevel setting (prints all kernel messages to the console)");
1263 :
1264 : static bool suppress_message_printing(int level)
1265 : {
1266 1388 : return (level >= console_loglevel && !ignore_loglevel);
1267 : }
1268 :
1269 : #ifdef CONFIG_BOOT_PRINTK_DELAY
1270 :
1271 : static int boot_delay; /* msecs delay after each printk during bootup */
1272 : static unsigned long long loops_per_msec; /* based on boot_delay */
1273 :
1274 : static int __init boot_delay_setup(char *str)
1275 : {
1276 : unsigned long lpj;
1277 :
1278 : lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1279 : loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1280 :
1281 : get_option(&str, &boot_delay);
1282 : if (boot_delay > 10 * 1000)
1283 : boot_delay = 0;
1284 :
1285 : pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1286 : "HZ: %d, loops_per_msec: %llu\n",
1287 : boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1288 : return 0;
1289 : }
1290 : early_param("boot_delay", boot_delay_setup);
1291 :
1292 : static void boot_delay_msec(int level)
1293 : {
1294 : unsigned long long k;
1295 : unsigned long timeout;
1296 :
1297 : if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1298 : || suppress_message_printing(level)) {
1299 : return;
1300 : }
1301 :
1302 : k = (unsigned long long)loops_per_msec * boot_delay;
1303 :
1304 : timeout = jiffies + msecs_to_jiffies(boot_delay);
1305 : while (k) {
1306 : k--;
1307 : cpu_relax();
1308 : /*
1309 : * use (volatile) jiffies to prevent
1310 : * compiler reduction; loop termination via jiffies
1311 : * is secondary and may or may not happen.
1312 : */
1313 : if (time_after(jiffies, timeout))
1314 : break;
1315 : touch_nmi_watchdog();
1316 : }
1317 : }
1318 : #else
1319 : static inline void boot_delay_msec(int level)
1320 : {
1321 : }
1322 : #endif
1323 :
1324 : static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1325 : module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1326 :
1327 : static size_t print_syslog(unsigned int level, char *buf)
1328 : {
1329 0 : return sprintf(buf, "<%u>", level);
1330 : }
1331 :
1332 : static size_t print_time(u64 ts, char *buf)
1333 : {
1334 0 : unsigned long rem_nsec = do_div(ts, 1000000000);
1335 :
1336 0 : return sprintf(buf, "[%5lu.%06lu]",
1337 : (unsigned long)ts, rem_nsec / 1000);
1338 : }
1339 :
1340 : #ifdef CONFIG_PRINTK_CALLER
1341 : static size_t print_caller(u32 id, char *buf)
1342 : {
1343 : char caller[12];
1344 :
1345 : snprintf(caller, sizeof(caller), "%c%u",
1346 : id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1347 : return sprintf(buf, "[%6s]", caller);
1348 : }
1349 : #else
1350 : #define print_caller(id, buf) 0
1351 : #endif
1352 :
1353 1386 : static size_t info_print_prefix(const struct printk_info *info, bool syslog,
1354 : bool time, char *buf)
1355 : {
1356 1386 : size_t len = 0;
1357 :
1358 1386 : if (syslog)
1359 0 : len = print_syslog((info->facility << 3) | info->level, buf);
1360 :
1361 1386 : if (time)
1362 0 : len += print_time(info->ts_nsec, buf + len);
1363 :
1364 1386 : len += print_caller(info->caller_id, buf + len);
1365 :
1366 1386 : if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1367 0 : buf[len++] = ' ';
1368 0 : buf[len] = '\0';
1369 : }
1370 :
1371 1386 : return len;
1372 : }
1373 :
1374 : /*
1375 : * Prepare the record for printing. The text is shifted within the given
1376 : * buffer to avoid a need for another one. The following operations are
1377 : * done:
1378 : *
1379 : * - Add prefix for each line.
1380 : * - Drop truncated lines that no longer fit into the buffer.
1381 : * - Add the trailing newline that has been removed in vprintk_store().
1382 : * - Add a string terminator.
1383 : *
1384 : * Since the produced string is always terminated, the maximum possible
1385 : * return value is @r->text_buf_size - 1;
1386 : *
1387 : * Return: The length of the updated/prepared text, including the added
1388 : * prefixes and the newline. The terminator is not counted. The dropped
1389 : * line(s) are not counted.
1390 : */
1391 1386 : static size_t record_print_text(struct printk_record *r, bool syslog,
1392 : bool time)
1393 : {
1394 1386 : size_t text_len = r->info->text_len;
1395 1386 : size_t buf_size = r->text_buf_size;
1396 1386 : char *text = r->text_buf;
1397 : char prefix[PRINTK_PREFIX_MAX];
1398 1386 : bool truncated = false;
1399 : size_t prefix_len;
1400 : size_t line_len;
1401 1386 : size_t len = 0;
1402 : char *next;
1403 :
1404 : /*
1405 : * If the message was truncated because the buffer was not large
1406 : * enough, treat the available text as if it were the full text.
1407 : */
1408 1386 : if (text_len > buf_size)
1409 0 : text_len = buf_size;
1410 :
1411 1386 : prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1412 :
1413 : /*
1414 : * @text_len: bytes of unprocessed text
1415 : * @line_len: bytes of current line _without_ newline
1416 : * @text: pointer to beginning of current line
1417 : * @len: number of bytes prepared in r->text_buf
1418 : */
1419 : for (;;) {
1420 1386 : next = memchr(text, '\n', text_len);
1421 1386 : if (next) {
1422 0 : line_len = next - text;
1423 : } else {
1424 : /* Drop truncated line(s). */
1425 1386 : if (truncated)
1426 : break;
1427 : line_len = text_len;
1428 : }
1429 :
1430 : /*
1431 : * Truncate the text if there is not enough space to add the
1432 : * prefix and a trailing newline and a terminator.
1433 : */
1434 1386 : if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1435 : /* Drop even the current line if no space. */
1436 0 : if (len + prefix_len + line_len + 1 + 1 > buf_size)
1437 : break;
1438 :
1439 0 : text_len = buf_size - len - prefix_len - 1 - 1;
1440 0 : truncated = true;
1441 : }
1442 :
1443 1386 : memmove(text + prefix_len, text, text_len);
1444 1386 : memcpy(text, prefix, prefix_len);
1445 :
1446 : /*
1447 : * Increment the prepared length to include the text and
1448 : * prefix that were just moved+copied. Also increment for the
1449 : * newline at the end of this line. If this is the last line,
1450 : * there is no newline, but it will be added immediately below.
1451 : */
1452 1386 : len += prefix_len + line_len + 1;
1453 1386 : if (text_len == line_len) {
1454 : /*
1455 : * This is the last line. Add the trailing newline
1456 : * removed in vprintk_store().
1457 : */
1458 1386 : text[prefix_len + line_len] = '\n';
1459 1386 : break;
1460 : }
1461 :
1462 : /*
1463 : * Advance beyond the added prefix and the related line with
1464 : * its newline.
1465 : */
1466 0 : text += prefix_len + line_len + 1;
1467 :
1468 : /*
1469 : * The remaining text has only decreased by the line with its
1470 : * newline.
1471 : *
1472 : * Note that @text_len can become zero. It happens when @text
1473 : * ended with a newline (either due to truncation or the
1474 : * original string ending with "\n\n"). The loop is correctly
1475 : * repeated and (if not truncated) an empty line with a prefix
1476 : * will be prepared.
1477 : */
1478 0 : text_len -= line_len + 1;
1479 : }
1480 :
1481 : /*
1482 : * If a buffer was provided, it will be terminated. Space for the
1483 : * string terminator is guaranteed to be available. The terminator is
1484 : * not counted in the return value.
1485 : */
1486 1386 : if (buf_size > 0)
1487 1386 : r->text_buf[len] = 0;
1488 :
1489 1386 : return len;
1490 : }
1491 :
1492 : static size_t get_record_print_text_size(struct printk_info *info,
1493 : unsigned int line_count,
1494 : bool syslog, bool time)
1495 : {
1496 : char prefix[PRINTK_PREFIX_MAX];
1497 : size_t prefix_len;
1498 :
1499 0 : prefix_len = info_print_prefix(info, syslog, time, prefix);
1500 :
1501 : /*
1502 : * Each line will be preceded with a prefix. The intermediate
1503 : * newlines are already within the text, but a final trailing
1504 : * newline will be added.
1505 : */
1506 0 : return ((prefix_len * line_count) + info->text_len + 1);
1507 : }
1508 :
1509 : /*
1510 : * Beginning with @start_seq, find the first record where it and all following
1511 : * records up to (but not including) @max_seq fit into @size.
1512 : *
1513 : * @max_seq is simply an upper bound and does not need to exist. If the caller
1514 : * does not require an upper bound, -1 can be used for @max_seq.
1515 : */
1516 0 : static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1517 : bool syslog, bool time)
1518 : {
1519 : struct printk_info info;
1520 : unsigned int line_count;
1521 0 : size_t len = 0;
1522 : u64 seq;
1523 :
1524 : /* Determine the size of the records up to @max_seq. */
1525 0 : prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1526 0 : if (info.seq >= max_seq)
1527 : break;
1528 0 : len += get_record_print_text_size(&info, line_count, syslog, time);
1529 : }
1530 :
1531 : /*
1532 : * Adjust the upper bound for the next loop to avoid subtracting
1533 : * lengths that were never added.
1534 : */
1535 0 : if (seq < max_seq)
1536 0 : max_seq = seq;
1537 :
1538 : /*
1539 : * Move first record forward until length fits into the buffer. Ignore
1540 : * newest messages that were not counted in the above cycle. Messages
1541 : * might appear and get lost in the meantime. This is a best effort
1542 : * that prevents an infinite loop that could occur with a retry.
1543 : */
1544 0 : prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1545 0 : if (len <= size || info.seq >= max_seq)
1546 : break;
1547 0 : len -= get_record_print_text_size(&info, line_count, syslog, time);
1548 : }
1549 :
1550 0 : return seq;
1551 : }
1552 :
1553 : /* The caller is responsible for making sure @size is greater than 0. */
1554 0 : static int syslog_print(char __user *buf, int size)
1555 : {
1556 : struct printk_info info;
1557 : struct printk_record r;
1558 : char *text;
1559 0 : int len = 0;
1560 : u64 seq;
1561 :
1562 0 : text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1563 0 : if (!text)
1564 : return -ENOMEM;
1565 :
1566 0 : prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1567 :
1568 0 : mutex_lock(&syslog_lock);
1569 :
1570 : /*
1571 : * Wait for the @syslog_seq record to be available. @syslog_seq may
1572 : * change while waiting.
1573 : */
1574 : do {
1575 0 : seq = syslog_seq;
1576 :
1577 0 : mutex_unlock(&syslog_lock);
1578 : /*
1579 : * Guarantee this task is visible on the waitqueue before
1580 : * checking the wake condition.
1581 : *
1582 : * The full memory barrier within set_current_state() of
1583 : * prepare_to_wait_event() pairs with the full memory barrier
1584 : * within wq_has_sleeper().
1585 : *
1586 : * This pairs with __wake_up_klogd:A.
1587 : */
1588 0 : len = wait_event_interruptible(log_wait,
1589 : prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1590 0 : mutex_lock(&syslog_lock);
1591 :
1592 0 : if (len)
1593 : goto out;
1594 0 : } while (syslog_seq != seq);
1595 :
1596 : /*
1597 : * Copy records that fit into the buffer. The above cycle makes sure
1598 : * that the first record is always available.
1599 : */
1600 : do {
1601 : size_t n;
1602 : size_t skip;
1603 : int err;
1604 :
1605 0 : if (!prb_read_valid(prb, syslog_seq, &r))
1606 : break;
1607 :
1608 0 : if (r.info->seq != syslog_seq) {
1609 : /* message is gone, move to next valid one */
1610 0 : syslog_seq = r.info->seq;
1611 0 : syslog_partial = 0;
1612 : }
1613 :
1614 : /*
1615 : * To keep reading/counting partial line consistent,
1616 : * use printk_time value as of the beginning of a line.
1617 : */
1618 0 : if (!syslog_partial)
1619 0 : syslog_time = printk_time;
1620 :
1621 0 : skip = syslog_partial;
1622 0 : n = record_print_text(&r, true, syslog_time);
1623 0 : if (n - syslog_partial <= size) {
1624 : /* message fits into buffer, move forward */
1625 0 : syslog_seq = r.info->seq + 1;
1626 0 : n -= syslog_partial;
1627 0 : syslog_partial = 0;
1628 0 : } else if (!len){
1629 : /* partial read(), remember position */
1630 0 : n = size;
1631 0 : syslog_partial += n;
1632 : } else
1633 : n = 0;
1634 :
1635 0 : if (!n)
1636 : break;
1637 :
1638 0 : mutex_unlock(&syslog_lock);
1639 0 : err = copy_to_user(buf, text + skip, n);
1640 0 : mutex_lock(&syslog_lock);
1641 :
1642 0 : if (err) {
1643 0 : if (!len)
1644 0 : len = -EFAULT;
1645 : break;
1646 : }
1647 :
1648 0 : len += n;
1649 0 : size -= n;
1650 0 : buf += n;
1651 0 : } while (size);
1652 : out:
1653 0 : mutex_unlock(&syslog_lock);
1654 0 : kfree(text);
1655 0 : return len;
1656 : }
1657 :
1658 0 : static int syslog_print_all(char __user *buf, int size, bool clear)
1659 : {
1660 : struct printk_info info;
1661 : struct printk_record r;
1662 : char *text;
1663 0 : int len = 0;
1664 : u64 seq;
1665 : bool time;
1666 :
1667 0 : text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1668 0 : if (!text)
1669 : return -ENOMEM;
1670 :
1671 0 : time = printk_time;
1672 : /*
1673 : * Find first record that fits, including all following records,
1674 : * into the user-provided buffer for this dump.
1675 : */
1676 0 : seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1677 : size, true, time);
1678 :
1679 0 : prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1680 :
1681 0 : len = 0;
1682 0 : prb_for_each_record(seq, prb, seq, &r) {
1683 : int textlen;
1684 :
1685 0 : textlen = record_print_text(&r, true, time);
1686 :
1687 0 : if (len + textlen > size) {
1688 0 : seq--;
1689 0 : break;
1690 : }
1691 :
1692 0 : if (copy_to_user(buf + len, text, textlen))
1693 : len = -EFAULT;
1694 : else
1695 0 : len += textlen;
1696 :
1697 0 : if (len < 0)
1698 : break;
1699 : }
1700 :
1701 0 : if (clear) {
1702 0 : mutex_lock(&syslog_lock);
1703 0 : latched_seq_write(&clear_seq, seq);
1704 0 : mutex_unlock(&syslog_lock);
1705 : }
1706 :
1707 0 : kfree(text);
1708 0 : return len;
1709 : }
1710 :
1711 0 : static void syslog_clear(void)
1712 : {
1713 0 : mutex_lock(&syslog_lock);
1714 0 : latched_seq_write(&clear_seq, prb_next_seq(prb));
1715 0 : mutex_unlock(&syslog_lock);
1716 0 : }
1717 :
1718 0 : int do_syslog(int type, char __user *buf, int len, int source)
1719 : {
1720 : struct printk_info info;
1721 0 : bool clear = false;
1722 : static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1723 : int error;
1724 :
1725 0 : error = check_syslog_permissions(type, source);
1726 0 : if (error)
1727 : return error;
1728 :
1729 0 : switch (type) {
1730 : case SYSLOG_ACTION_CLOSE: /* Close log */
1731 : break;
1732 : case SYSLOG_ACTION_OPEN: /* Open log */
1733 : break;
1734 : case SYSLOG_ACTION_READ: /* Read from log */
1735 0 : if (!buf || len < 0)
1736 : return -EINVAL;
1737 0 : if (!len)
1738 : return 0;
1739 0 : if (!access_ok(buf, len))
1740 : return -EFAULT;
1741 0 : error = syslog_print(buf, len);
1742 0 : break;
1743 : /* Read/clear last kernel messages */
1744 : case SYSLOG_ACTION_READ_CLEAR:
1745 0 : clear = true;
1746 : fallthrough;
1747 : /* Read last kernel messages */
1748 : case SYSLOG_ACTION_READ_ALL:
1749 0 : if (!buf || len < 0)
1750 : return -EINVAL;
1751 0 : if (!len)
1752 : return 0;
1753 0 : if (!access_ok(buf, len))
1754 : return -EFAULT;
1755 0 : error = syslog_print_all(buf, len, clear);
1756 0 : break;
1757 : /* Clear ring buffer */
1758 : case SYSLOG_ACTION_CLEAR:
1759 0 : syslog_clear();
1760 0 : break;
1761 : /* Disable logging to console */
1762 : case SYSLOG_ACTION_CONSOLE_OFF:
1763 0 : if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1764 0 : saved_console_loglevel = console_loglevel;
1765 0 : console_loglevel = minimum_console_loglevel;
1766 0 : break;
1767 : /* Enable logging to console */
1768 : case SYSLOG_ACTION_CONSOLE_ON:
1769 0 : if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1770 0 : console_loglevel = saved_console_loglevel;
1771 0 : saved_console_loglevel = LOGLEVEL_DEFAULT;
1772 : }
1773 : break;
1774 : /* Set level of messages printed to console */
1775 : case SYSLOG_ACTION_CONSOLE_LEVEL:
1776 0 : if (len < 1 || len > 8)
1777 : return -EINVAL;
1778 0 : if (len < minimum_console_loglevel)
1779 0 : len = minimum_console_loglevel;
1780 0 : console_loglevel = len;
1781 : /* Implicitly re-enable logging to console */
1782 0 : saved_console_loglevel = LOGLEVEL_DEFAULT;
1783 0 : break;
1784 : /* Number of chars in the log buffer */
1785 : case SYSLOG_ACTION_SIZE_UNREAD:
1786 0 : mutex_lock(&syslog_lock);
1787 0 : if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1788 : /* No unread messages. */
1789 0 : mutex_unlock(&syslog_lock);
1790 0 : return 0;
1791 : }
1792 0 : if (info.seq != syslog_seq) {
1793 : /* messages are gone, move to first one */
1794 0 : syslog_seq = info.seq;
1795 0 : syslog_partial = 0;
1796 : }
1797 0 : if (source == SYSLOG_FROM_PROC) {
1798 : /*
1799 : * Short-cut for poll(/"proc/kmsg") which simply checks
1800 : * for pending data, not the size; return the count of
1801 : * records, not the length.
1802 : */
1803 0 : error = prb_next_seq(prb) - syslog_seq;
1804 : } else {
1805 0 : bool time = syslog_partial ? syslog_time : printk_time;
1806 : unsigned int line_count;
1807 : u64 seq;
1808 :
1809 0 : prb_for_each_info(syslog_seq, prb, seq, &info,
1810 : &line_count) {
1811 0 : error += get_record_print_text_size(&info, line_count,
1812 : true, time);
1813 0 : time = printk_time;
1814 : }
1815 0 : error -= syslog_partial;
1816 : }
1817 0 : mutex_unlock(&syslog_lock);
1818 0 : break;
1819 : /* Size of the log buffer */
1820 : case SYSLOG_ACTION_SIZE_BUFFER:
1821 0 : error = log_buf_len;
1822 0 : break;
1823 : default:
1824 0 : error = -EINVAL;
1825 0 : break;
1826 : }
1827 :
1828 : return error;
1829 : }
1830 :
1831 0 : SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1832 : {
1833 0 : return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1834 : }
1835 :
1836 : /*
1837 : * Special console_lock variants that help to reduce the risk of soft-lockups.
1838 : * They allow to pass console_lock to another printk() call using a busy wait.
1839 : */
1840 :
1841 : #ifdef CONFIG_LOCKDEP
1842 : static struct lockdep_map console_owner_dep_map = {
1843 : .name = "console_owner"
1844 : };
1845 : #endif
1846 :
1847 : static DEFINE_RAW_SPINLOCK(console_owner_lock);
1848 : static struct task_struct *console_owner;
1849 : static bool console_waiter;
1850 :
1851 : /**
1852 : * console_lock_spinning_enable - mark beginning of code where another
1853 : * thread might safely busy wait
1854 : *
1855 : * This basically converts console_lock into a spinlock. This marks
1856 : * the section where the console_lock owner can not sleep, because
1857 : * there may be a waiter spinning (like a spinlock). Also it must be
1858 : * ready to hand over the lock at the end of the section.
1859 : */
1860 : static void console_lock_spinning_enable(void)
1861 : {
1862 1386 : raw_spin_lock(&console_owner_lock);
1863 1386 : console_owner = current;
1864 1386 : raw_spin_unlock(&console_owner_lock);
1865 :
1866 : /* The waiter may spin on us after setting console_owner */
1867 : spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1868 : }
1869 :
1870 : /**
1871 : * console_lock_spinning_disable_and_check - mark end of code where another
1872 : * thread was able to busy wait and check if there is a waiter
1873 : * @cookie: cookie returned from console_srcu_read_lock()
1874 : *
1875 : * This is called at the end of the section where spinning is allowed.
1876 : * It has two functions. First, it is a signal that it is no longer
1877 : * safe to start busy waiting for the lock. Second, it checks if
1878 : * there is a busy waiter and passes the lock rights to her.
1879 : *
1880 : * Important: Callers lose both the console_lock and the SRCU read lock if
1881 : * there was a busy waiter. They must not touch items synchronized by
1882 : * console_lock or SRCU read lock in this case.
1883 : *
1884 : * Return: 1 if the lock rights were passed, 0 otherwise.
1885 : */
1886 : static int console_lock_spinning_disable_and_check(int cookie)
1887 : {
1888 : int waiter;
1889 :
1890 1386 : raw_spin_lock(&console_owner_lock);
1891 1386 : waiter = READ_ONCE(console_waiter);
1892 1386 : console_owner = NULL;
1893 1386 : raw_spin_unlock(&console_owner_lock);
1894 :
1895 1386 : if (!waiter) {
1896 : spin_release(&console_owner_dep_map, _THIS_IP_);
1897 : return 0;
1898 : }
1899 :
1900 : /* The waiter is now free to continue */
1901 0 : WRITE_ONCE(console_waiter, false);
1902 :
1903 : spin_release(&console_owner_dep_map, _THIS_IP_);
1904 :
1905 : /*
1906 : * Preserve lockdep lock ordering. Release the SRCU read lock before
1907 : * releasing the console_lock.
1908 : */
1909 0 : console_srcu_read_unlock(cookie);
1910 :
1911 : /*
1912 : * Hand off console_lock to waiter. The waiter will perform
1913 : * the up(). After this, the waiter is the console_lock owner.
1914 : */
1915 : mutex_release(&console_lock_dep_map, _THIS_IP_);
1916 : return 1;
1917 : }
1918 :
1919 : /**
1920 : * console_trylock_spinning - try to get console_lock by busy waiting
1921 : *
1922 : * This allows to busy wait for the console_lock when the current
1923 : * owner is running in specially marked sections. It means that
1924 : * the current owner is running and cannot reschedule until it
1925 : * is ready to lose the lock.
1926 : *
1927 : * Return: 1 if we got the lock, 0 othrewise
1928 : */
1929 811 : static int console_trylock_spinning(void)
1930 : {
1931 811 : struct task_struct *owner = NULL;
1932 : bool waiter;
1933 811 : bool spin = false;
1934 : unsigned long flags;
1935 :
1936 811 : if (console_trylock())
1937 : return 1;
1938 :
1939 : /*
1940 : * It's unsafe to spin once a panic has begun. If we are the
1941 : * panic CPU, we may have already halted the owner of the
1942 : * console_sem. If we are not the panic CPU, then we should
1943 : * avoid taking console_sem, so the panic CPU has a better
1944 : * chance of cleanly acquiring it later.
1945 : */
1946 0 : if (panic_in_progress())
1947 : return 0;
1948 :
1949 0 : printk_safe_enter_irqsave(flags);
1950 :
1951 0 : raw_spin_lock(&console_owner_lock);
1952 0 : owner = READ_ONCE(console_owner);
1953 0 : waiter = READ_ONCE(console_waiter);
1954 0 : if (!waiter && owner && owner != current) {
1955 0 : WRITE_ONCE(console_waiter, true);
1956 0 : spin = true;
1957 : }
1958 0 : raw_spin_unlock(&console_owner_lock);
1959 :
1960 : /*
1961 : * If there is an active printk() writing to the
1962 : * consoles, instead of having it write our data too,
1963 : * see if we can offload that load from the active
1964 : * printer, and do some printing ourselves.
1965 : * Go into a spin only if there isn't already a waiter
1966 : * spinning, and there is an active printer, and
1967 : * that active printer isn't us (recursive printk?).
1968 : */
1969 0 : if (!spin) {
1970 0 : printk_safe_exit_irqrestore(flags);
1971 0 : return 0;
1972 : }
1973 :
1974 : /* We spin waiting for the owner to release us */
1975 : spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1976 : /* Owner will clear console_waiter on hand off */
1977 0 : while (READ_ONCE(console_waiter))
1978 : cpu_relax();
1979 : spin_release(&console_owner_dep_map, _THIS_IP_);
1980 :
1981 0 : printk_safe_exit_irqrestore(flags);
1982 : /*
1983 : * The owner passed the console lock to us.
1984 : * Since we did not spin on console lock, annotate
1985 : * this as a trylock. Otherwise lockdep will
1986 : * complain.
1987 : */
1988 : mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1989 :
1990 0 : return 1;
1991 : }
1992 :
1993 : /*
1994 : * Recursion is tracked separately on each CPU. If NMIs are supported, an
1995 : * additional NMI context per CPU is also separately tracked. Until per-CPU
1996 : * is available, a separate "early tracking" is performed.
1997 : */
1998 : static DEFINE_PER_CPU(u8, printk_count);
1999 : static u8 printk_count_early;
2000 : #ifdef CONFIG_HAVE_NMI
2001 : static DEFINE_PER_CPU(u8, printk_count_nmi);
2002 : static u8 printk_count_nmi_early;
2003 : #endif
2004 :
2005 : /*
2006 : * Recursion is limited to keep the output sane. printk() should not require
2007 : * more than 1 level of recursion (allowing, for example, printk() to trigger
2008 : * a WARN), but a higher value is used in case some printk-internal errors
2009 : * exist, such as the ringbuffer validation checks failing.
2010 : */
2011 : #define PRINTK_MAX_RECURSION 3
2012 :
2013 : /*
2014 : * Return a pointer to the dedicated counter for the CPU+context of the
2015 : * caller.
2016 : */
2017 : static u8 *__printk_recursion_counter(void)
2018 : {
2019 : #ifdef CONFIG_HAVE_NMI
2020 : if (in_nmi()) {
2021 : if (printk_percpu_data_ready())
2022 : return this_cpu_ptr(&printk_count_nmi);
2023 : return &printk_count_nmi_early;
2024 : }
2025 : #endif
2026 811 : if (printk_percpu_data_ready())
2027 : return this_cpu_ptr(&printk_count);
2028 : return &printk_count_early;
2029 : }
2030 :
2031 : /*
2032 : * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2033 : * The caller must check the boolean return value to see if the recursion is
2034 : * allowed. On failure, interrupts are not disabled.
2035 : *
2036 : * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2037 : * that is passed to printk_exit_irqrestore().
2038 : */
2039 : #define printk_enter_irqsave(recursion_ptr, flags) \
2040 : ({ \
2041 : bool success = true; \
2042 : \
2043 : typecheck(u8 *, recursion_ptr); \
2044 : local_irq_save(flags); \
2045 : (recursion_ptr) = __printk_recursion_counter(); \
2046 : if (*(recursion_ptr) > PRINTK_MAX_RECURSION) { \
2047 : local_irq_restore(flags); \
2048 : success = false; \
2049 : } else { \
2050 : (*(recursion_ptr))++; \
2051 : } \
2052 : success; \
2053 : })
2054 :
2055 : /* Exit recursion tracking, restoring interrupts. */
2056 : #define printk_exit_irqrestore(recursion_ptr, flags) \
2057 : do { \
2058 : typecheck(u8 *, recursion_ptr); \
2059 : (*(recursion_ptr))--; \
2060 : local_irq_restore(flags); \
2061 : } while (0)
2062 :
2063 : int printk_delay_msec __read_mostly;
2064 :
2065 : static inline void printk_delay(int level)
2066 : {
2067 811 : boot_delay_msec(level);
2068 :
2069 811 : if (unlikely(printk_delay_msec)) {
2070 : int m = printk_delay_msec;
2071 :
2072 0 : while (m--) {
2073 : mdelay(1);
2074 : touch_nmi_watchdog();
2075 : }
2076 : }
2077 : }
2078 :
2079 : static inline u32 printk_caller_id(void)
2080 : {
2081 1622 : return in_task() ? task_pid_nr(current) :
2082 : 0x80000000 + smp_processor_id();
2083 : }
2084 :
2085 : /**
2086 : * printk_parse_prefix - Parse level and control flags.
2087 : *
2088 : * @text: The terminated text message.
2089 : * @level: A pointer to the current level value, will be updated.
2090 : * @flags: A pointer to the current printk_info flags, will be updated.
2091 : *
2092 : * @level may be NULL if the caller is not interested in the parsed value.
2093 : * Otherwise the variable pointed to by @level must be set to
2094 : * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2095 : *
2096 : * @flags may be NULL if the caller is not interested in the parsed value.
2097 : * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2098 : * value.
2099 : *
2100 : * Return: The length of the parsed level and control flags.
2101 : */
2102 1622 : u16 printk_parse_prefix(const char *text, int *level,
2103 : enum printk_info_flags *flags)
2104 : {
2105 1622 : u16 prefix_len = 0;
2106 : int kern_level;
2107 :
2108 4602 : while (*text) {
2109 2980 : kern_level = printk_get_level(text);
2110 2980 : if (!kern_level)
2111 : break;
2112 :
2113 1358 : switch (kern_level) {
2114 : case '0' ... '7':
2115 1202 : if (level && *level == LOGLEVEL_DEFAULT)
2116 601 : *level = kern_level - '0';
2117 : break;
2118 : case 'c': /* KERN_CONT */
2119 156 : if (flags)
2120 78 : *flags |= LOG_CONT;
2121 : }
2122 :
2123 1358 : prefix_len += 2;
2124 1358 : text += 2;
2125 : }
2126 :
2127 1622 : return prefix_len;
2128 : }
2129 :
2130 : __printf(5, 0)
2131 811 : static u16 printk_sprint(char *text, u16 size, int facility,
2132 : enum printk_info_flags *flags, const char *fmt,
2133 : va_list args)
2134 : {
2135 : u16 text_len;
2136 :
2137 811 : text_len = vscnprintf(text, size, fmt, args);
2138 :
2139 : /* Mark and strip a trailing newline. */
2140 811 : if (text_len && text[text_len - 1] == '\n') {
2141 290 : text_len--;
2142 290 : *flags |= LOG_NEWLINE;
2143 : }
2144 :
2145 : /* Strip log level and control flags. */
2146 811 : if (facility == 0) {
2147 : u16 prefix_len;
2148 :
2149 811 : prefix_len = printk_parse_prefix(text, NULL, NULL);
2150 811 : if (prefix_len) {
2151 679 : text_len -= prefix_len;
2152 679 : memmove(text, text + prefix_len, text_len);
2153 : }
2154 : }
2155 :
2156 811 : trace_console(text, text_len);
2157 :
2158 811 : return text_len;
2159 : }
2160 :
2161 : __printf(4, 0)
2162 811 : int vprintk_store(int facility, int level,
2163 : const struct dev_printk_info *dev_info,
2164 : const char *fmt, va_list args)
2165 : {
2166 : struct prb_reserved_entry e;
2167 811 : enum printk_info_flags flags = 0;
2168 : struct printk_record r;
2169 : unsigned long irqflags;
2170 811 : u16 trunc_msg_len = 0;
2171 : char prefix_buf[8];
2172 : u8 *recursion_ptr;
2173 : u16 reserve_size;
2174 : va_list args2;
2175 : u32 caller_id;
2176 : u16 text_len;
2177 811 : int ret = 0;
2178 : u64 ts_nsec;
2179 :
2180 2433 : if (!printk_enter_irqsave(recursion_ptr, irqflags))
2181 : return 0;
2182 :
2183 : /*
2184 : * Since the duration of printk() can vary depending on the message
2185 : * and state of the ringbuffer, grab the timestamp now so that it is
2186 : * close to the call of printk(). This provides a more deterministic
2187 : * timestamp with respect to the caller.
2188 : */
2189 811 : ts_nsec = local_clock();
2190 :
2191 811 : caller_id = printk_caller_id();
2192 :
2193 : /*
2194 : * The sprintf needs to come first since the syslog prefix might be
2195 : * passed in as a parameter. An extra byte must be reserved so that
2196 : * later the vscnprintf() into the reserved buffer has room for the
2197 : * terminating '\0', which is not counted by vsnprintf().
2198 : */
2199 811 : va_copy(args2, args);
2200 811 : reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2201 811 : va_end(args2);
2202 :
2203 811 : if (reserve_size > PRINTKRB_RECORD_MAX)
2204 0 : reserve_size = PRINTKRB_RECORD_MAX;
2205 :
2206 : /* Extract log level or control flags. */
2207 811 : if (facility == 0)
2208 811 : printk_parse_prefix(&prefix_buf[0], &level, &flags);
2209 :
2210 811 : if (level == LOGLEVEL_DEFAULT)
2211 210 : level = default_message_loglevel;
2212 :
2213 811 : if (dev_info)
2214 0 : flags |= LOG_NEWLINE;
2215 :
2216 811 : if (flags & LOG_CONT) {
2217 156 : prb_rec_init_wr(&r, reserve_size);
2218 78 : if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2219 63 : text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2220 : facility, &flags, fmt, args);
2221 63 : r.info->text_len += text_len;
2222 :
2223 63 : if (flags & LOG_NEWLINE) {
2224 16 : r.info->flags |= LOG_NEWLINE;
2225 16 : prb_final_commit(&e);
2226 : } else {
2227 47 : prb_commit(&e);
2228 : }
2229 :
2230 63 : ret = text_len;
2231 63 : goto out;
2232 : }
2233 : }
2234 :
2235 : /*
2236 : * Explicitly initialize the record before every prb_reserve() call.
2237 : * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2238 : * structure when they fail.
2239 : */
2240 1496 : prb_rec_init_wr(&r, reserve_size);
2241 748 : if (!prb_reserve(&e, prb, &r)) {
2242 : /* truncate the message if it is too long for empty buffer */
2243 0 : truncate_msg(&reserve_size, &trunc_msg_len);
2244 :
2245 0 : prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2246 0 : if (!prb_reserve(&e, prb, &r))
2247 : goto out;
2248 : }
2249 :
2250 : /* fill message */
2251 748 : text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2252 748 : if (trunc_msg_len)
2253 0 : memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2254 748 : r.info->text_len = text_len + trunc_msg_len;
2255 748 : r.info->facility = facility;
2256 748 : r.info->level = level & 7;
2257 748 : r.info->flags = flags & 0x1f;
2258 748 : r.info->ts_nsec = ts_nsec;
2259 748 : r.info->caller_id = caller_id;
2260 748 : if (dev_info)
2261 0 : memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2262 :
2263 : /* A message without a trailing newline can be continued. */
2264 748 : if (!(flags & LOG_NEWLINE))
2265 474 : prb_commit(&e);
2266 : else
2267 274 : prb_final_commit(&e);
2268 :
2269 748 : ret = text_len + trunc_msg_len;
2270 : out:
2271 1622 : printk_exit_irqrestore(recursion_ptr, irqflags);
2272 811 : return ret;
2273 : }
2274 :
2275 811 : asmlinkage int vprintk_emit(int facility, int level,
2276 : const struct dev_printk_info *dev_info,
2277 : const char *fmt, va_list args)
2278 : {
2279 : int printed_len;
2280 811 : bool in_sched = false;
2281 :
2282 : /* Suppress unimportant messages after panic happens */
2283 811 : if (unlikely(suppress_printk))
2284 : return 0;
2285 :
2286 811 : if (unlikely(suppress_panic_printk) &&
2287 0 : atomic_read(&panic_cpu) != raw_smp_processor_id())
2288 : return 0;
2289 :
2290 811 : if (level == LOGLEVEL_SCHED) {
2291 0 : level = LOGLEVEL_DEFAULT;
2292 0 : in_sched = true;
2293 : }
2294 :
2295 1622 : printk_delay(level);
2296 :
2297 811 : printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2298 :
2299 : /* If called from the scheduler, we can not call up(). */
2300 811 : if (!in_sched) {
2301 : /*
2302 : * The caller may be holding system-critical or
2303 : * timing-sensitive locks. Disable preemption during
2304 : * printing of all remaining records to all consoles so that
2305 : * this context can return as soon as possible. Hopefully
2306 : * another printk() caller will take over the printing.
2307 : */
2308 811 : preempt_disable();
2309 : /*
2310 : * Try to acquire and then immediately release the console
2311 : * semaphore. The release will print out buffers. With the
2312 : * spinning variant, this context tries to take over the
2313 : * printing from another printing context.
2314 : */
2315 811 : if (console_trylock_spinning())
2316 811 : console_unlock();
2317 811 : preempt_enable();
2318 : }
2319 :
2320 : wake_up_klogd();
2321 811 : return printed_len;
2322 : }
2323 : EXPORT_SYMBOL(vprintk_emit);
2324 :
2325 811 : int vprintk_default(const char *fmt, va_list args)
2326 : {
2327 811 : return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2328 : }
2329 : EXPORT_SYMBOL_GPL(vprintk_default);
2330 :
2331 810 : asmlinkage __visible int _printk(const char *fmt, ...)
2332 : {
2333 : va_list args;
2334 : int r;
2335 :
2336 810 : va_start(args, fmt);
2337 810 : r = vprintk(fmt, args);
2338 810 : va_end(args);
2339 :
2340 810 : return r;
2341 : }
2342 : EXPORT_SYMBOL(_printk);
2343 :
2344 : static bool pr_flush(int timeout_ms, bool reset_on_progress);
2345 : static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2346 :
2347 : #else /* CONFIG_PRINTK */
2348 :
2349 : #define printk_time false
2350 :
2351 : #define prb_read_valid(rb, seq, r) false
2352 : #define prb_first_valid_seq(rb) 0
2353 : #define prb_next_seq(rb) 0
2354 :
2355 : static u64 syslog_seq;
2356 :
2357 : static size_t record_print_text(const struct printk_record *r,
2358 : bool syslog, bool time)
2359 : {
2360 : return 0;
2361 : }
2362 : static ssize_t info_print_ext_header(char *buf, size_t size,
2363 : struct printk_info *info)
2364 : {
2365 : return 0;
2366 : }
2367 : static ssize_t msg_print_ext_body(char *buf, size_t size,
2368 : char *text, size_t text_len,
2369 : struct dev_printk_info *dev_info) { return 0; }
2370 : static void console_lock_spinning_enable(void) { }
2371 : static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2372 : static bool suppress_message_printing(int level) { return false; }
2373 : static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2374 : static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2375 :
2376 : #endif /* CONFIG_PRINTK */
2377 :
2378 : #ifdef CONFIG_EARLY_PRINTK
2379 : struct console *early_console;
2380 :
2381 0 : asmlinkage __visible void early_printk(const char *fmt, ...)
2382 : {
2383 : va_list ap;
2384 : char buf[512];
2385 : int n;
2386 :
2387 0 : if (!early_console)
2388 0 : return;
2389 :
2390 0 : va_start(ap, fmt);
2391 0 : n = vscnprintf(buf, sizeof(buf), fmt, ap);
2392 0 : va_end(ap);
2393 :
2394 0 : early_console->write(early_console, buf, n);
2395 : }
2396 : #endif
2397 :
2398 : static void set_user_specified(struct console_cmdline *c, bool user_specified)
2399 : {
2400 1 : if (!user_specified)
2401 : return;
2402 :
2403 : /*
2404 : * @c console was defined by the user on the command line.
2405 : * Do not clear when added twice also by SPCR or the device tree.
2406 : */
2407 1 : c->user_specified = true;
2408 : /* At least one console defined by the user on the command line. */
2409 1 : console_set_on_cmdline = 1;
2410 : }
2411 :
2412 1 : static int __add_preferred_console(char *name, int idx, char *options,
2413 : char *brl_options, bool user_specified)
2414 : {
2415 : struct console_cmdline *c;
2416 : int i;
2417 :
2418 : /*
2419 : * See if this tty is not yet registered, and
2420 : * if we have a slot free.
2421 : */
2422 2 : for (i = 0, c = console_cmdline;
2423 1 : i < MAX_CMDLINECONSOLES && c->name[0];
2424 0 : i++, c++) {
2425 0 : if (strcmp(c->name, name) == 0 && c->index == idx) {
2426 0 : if (!brl_options)
2427 0 : preferred_console = i;
2428 0 : set_user_specified(c, user_specified);
2429 : return 0;
2430 : }
2431 : }
2432 1 : if (i == MAX_CMDLINECONSOLES)
2433 : return -E2BIG;
2434 1 : if (!brl_options)
2435 1 : preferred_console = i;
2436 1 : strscpy(c->name, name, sizeof(c->name));
2437 1 : c->options = options;
2438 2 : set_user_specified(c, user_specified);
2439 1 : braille_set_options(c, brl_options);
2440 :
2441 1 : c->index = idx;
2442 1 : return 0;
2443 : }
2444 :
2445 0 : static int __init console_msg_format_setup(char *str)
2446 : {
2447 0 : if (!strcmp(str, "syslog"))
2448 0 : console_msg_format = MSG_FORMAT_SYSLOG;
2449 0 : if (!strcmp(str, "default"))
2450 0 : console_msg_format = MSG_FORMAT_DEFAULT;
2451 0 : return 1;
2452 : }
2453 : __setup("console_msg_format=", console_msg_format_setup);
2454 :
2455 : /*
2456 : * Set up a console. Called via do_early_param() in init/main.c
2457 : * for each "console=" parameter in the boot command line.
2458 : */
2459 1 : static int __init console_setup(char *str)
2460 : {
2461 : char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2462 1 : char *s, *options, *brl_options = NULL;
2463 : int idx;
2464 :
2465 : /*
2466 : * console="" or console=null have been suggested as a way to
2467 : * disable console output. Use ttynull that has been created
2468 : * for exactly this purpose.
2469 : */
2470 1 : if (str[0] == 0 || strcmp(str, "null") == 0) {
2471 0 : __add_preferred_console("ttynull", 0, NULL, NULL, true);
2472 0 : return 1;
2473 : }
2474 :
2475 1 : if (_braille_console_setup(&str, &brl_options))
2476 : return 1;
2477 :
2478 : /*
2479 : * Decode str into name, index, options.
2480 : */
2481 1 : if (str[0] >= '0' && str[0] <= '9') {
2482 0 : strcpy(buf, "ttyS");
2483 0 : strncpy(buf + 4, str, sizeof(buf) - 5);
2484 : } else {
2485 1 : strncpy(buf, str, sizeof(buf) - 1);
2486 : }
2487 1 : buf[sizeof(buf) - 1] = 0;
2488 1 : options = strchr(str, ',');
2489 1 : if (options)
2490 0 : *(options++) = 0;
2491 : #ifdef __sparc__
2492 : if (!strcmp(str, "ttya"))
2493 : strcpy(buf, "ttyS0");
2494 : if (!strcmp(str, "ttyb"))
2495 : strcpy(buf, "ttyS1");
2496 : #endif
2497 4 : for (s = buf; *s; s++)
2498 6 : if (isdigit(*s) || *s == ',')
2499 : break;
2500 1 : idx = simple_strtoul(s, NULL, 10);
2501 1 : *s = 0;
2502 :
2503 1 : __add_preferred_console(buf, idx, options, brl_options, true);
2504 1 : return 1;
2505 : }
2506 : __setup("console=", console_setup);
2507 :
2508 : /**
2509 : * add_preferred_console - add a device to the list of preferred consoles.
2510 : * @name: device name
2511 : * @idx: device index
2512 : * @options: options for this console
2513 : *
2514 : * The last preferred console added will be used for kernel messages
2515 : * and stdin/out/err for init. Normally this is used by console_setup
2516 : * above to handle user-supplied console arguments; however it can also
2517 : * be used by arch-specific code either to override the user or more
2518 : * commonly to provide a default console (ie from PROM variables) when
2519 : * the user has not supplied one.
2520 : */
2521 0 : int add_preferred_console(char *name, int idx, char *options)
2522 : {
2523 0 : return __add_preferred_console(name, idx, options, NULL, false);
2524 : }
2525 :
2526 : bool console_suspend_enabled = true;
2527 : EXPORT_SYMBOL(console_suspend_enabled);
2528 :
2529 0 : static int __init console_suspend_disable(char *str)
2530 : {
2531 0 : console_suspend_enabled = false;
2532 0 : return 1;
2533 : }
2534 : __setup("no_console_suspend", console_suspend_disable);
2535 : module_param_named(console_suspend, console_suspend_enabled,
2536 : bool, S_IRUGO | S_IWUSR);
2537 : MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2538 : " and hibernate operations");
2539 :
2540 : static bool printk_console_no_auto_verbose;
2541 :
2542 0 : void console_verbose(void)
2543 : {
2544 0 : if (console_loglevel && !printk_console_no_auto_verbose)
2545 0 : console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2546 0 : }
2547 : EXPORT_SYMBOL_GPL(console_verbose);
2548 :
2549 : module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2550 : MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2551 :
2552 : /**
2553 : * suspend_console - suspend the console subsystem
2554 : *
2555 : * This disables printk() while we go into suspend states
2556 : */
2557 0 : void suspend_console(void)
2558 : {
2559 0 : if (!console_suspend_enabled)
2560 : return;
2561 0 : pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2562 0 : pr_flush(1000, true);
2563 : console_lock();
2564 0 : console_suspended = 1;
2565 0 : up_console_sem();
2566 : }
2567 :
2568 0 : void resume_console(void)
2569 : {
2570 0 : if (!console_suspend_enabled)
2571 : return;
2572 0 : down_console_sem();
2573 0 : console_suspended = 0;
2574 0 : console_unlock();
2575 : pr_flush(1000, true);
2576 : }
2577 :
2578 : /**
2579 : * console_cpu_notify - print deferred console messages after CPU hotplug
2580 : * @cpu: unused
2581 : *
2582 : * If printk() is called from a CPU that is not online yet, the messages
2583 : * will be printed on the console only if there are CON_ANYTIME consoles.
2584 : * This function is called when a new CPU comes online (or fails to come
2585 : * up) or goes offline.
2586 : */
2587 0 : static int console_cpu_notify(unsigned int cpu)
2588 : {
2589 : if (!cpuhp_tasks_frozen) {
2590 : /* If trylock fails, someone else is doing the printing */
2591 0 : if (console_trylock())
2592 0 : console_unlock();
2593 : }
2594 0 : return 0;
2595 : }
2596 :
2597 : /**
2598 : * console_lock - block the console subsystem from printing
2599 : *
2600 : * Acquires a lock which guarantees that no consoles will
2601 : * be in or enter their write() callback.
2602 : *
2603 : * Can sleep, returns nothing.
2604 : */
2605 0 : void console_lock(void)
2606 : {
2607 : might_sleep();
2608 :
2609 0 : down_console_sem();
2610 0 : if (console_suspended)
2611 : return;
2612 0 : console_locked = 1;
2613 0 : console_may_schedule = 1;
2614 : }
2615 : EXPORT_SYMBOL(console_lock);
2616 :
2617 : /**
2618 : * console_trylock - try to block the console subsystem from printing
2619 : *
2620 : * Try to acquire a lock which guarantees that no consoles will
2621 : * be in or enter their write() callback.
2622 : *
2623 : * returns 1 on success, and 0 on failure to acquire the lock.
2624 : */
2625 811 : int console_trylock(void)
2626 : {
2627 811 : if (down_trylock_console_sem())
2628 : return 0;
2629 811 : if (console_suspended) {
2630 0 : up_console_sem();
2631 0 : return 0;
2632 : }
2633 811 : console_locked = 1;
2634 811 : console_may_schedule = 0;
2635 811 : return 1;
2636 : }
2637 : EXPORT_SYMBOL(console_trylock);
2638 :
2639 0 : int is_console_locked(void)
2640 : {
2641 0 : return console_locked;
2642 : }
2643 : EXPORT_SYMBOL(is_console_locked);
2644 :
2645 : /*
2646 : * Return true when this CPU should unlock console_sem without pushing all
2647 : * messages to the console. This reduces the chance that the console is
2648 : * locked when the panic CPU tries to use it.
2649 : */
2650 : static bool abandon_console_lock_in_panic(void)
2651 : {
2652 1388 : if (!panic_in_progress())
2653 : return false;
2654 :
2655 : /*
2656 : * We can use raw_smp_processor_id() here because it is impossible for
2657 : * the task to be migrated to the panic_cpu, or away from it. If
2658 : * panic_cpu has already been set, and we're not currently executing on
2659 : * that CPU, then we never will be.
2660 : */
2661 0 : return atomic_read(&panic_cpu) != raw_smp_processor_id();
2662 : }
2663 :
2664 : /*
2665 : * Check if the given console is currently capable and allowed to print
2666 : * records.
2667 : *
2668 : * Requires the console_srcu_read_lock.
2669 : */
2670 : static inline bool console_is_usable(struct console *con)
2671 : {
2672 2757 : short flags = console_srcu_read_flags(con);
2673 :
2674 2757 : if (!(flags & CON_ENABLED))
2675 : return false;
2676 :
2677 2757 : if (!con->write)
2678 : return false;
2679 :
2680 : /*
2681 : * Console drivers may assume that per-cpu resources have been
2682 : * allocated. So unless they're explicitly marked as being able to
2683 : * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2684 : */
2685 2757 : if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2686 : return false;
2687 :
2688 : return true;
2689 : }
2690 :
2691 : static void __console_unlock(void)
2692 : {
2693 811 : console_locked = 0;
2694 811 : up_console_sem();
2695 : }
2696 :
2697 : /*
2698 : * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2699 : * is achieved by shifting the existing message over and inserting the dropped
2700 : * message.
2701 : *
2702 : * @pmsg is the printk message to prepend.
2703 : *
2704 : * @dropped is the dropped count to report in the dropped message.
2705 : *
2706 : * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2707 : * the dropped message, the message text will be sufficiently truncated.
2708 : *
2709 : * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2710 : */
2711 : #ifdef CONFIG_PRINTK
2712 0 : static void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2713 : {
2714 0 : struct printk_buffers *pbufs = pmsg->pbufs;
2715 0 : const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2716 0 : const size_t outbuf_sz = sizeof(pbufs->outbuf);
2717 0 : char *scratchbuf = &pbufs->scratchbuf[0];
2718 0 : char *outbuf = &pbufs->outbuf[0];
2719 : size_t len;
2720 :
2721 0 : len = scnprintf(scratchbuf, scratchbuf_sz,
2722 : "** %lu printk messages dropped **\n", dropped);
2723 :
2724 : /*
2725 : * Make sure outbuf is sufficiently large before prepending.
2726 : * Keep at least the prefix when the message must be truncated.
2727 : * It is a rather theoretical problem when someone tries to
2728 : * use a minimalist buffer.
2729 : */
2730 0 : if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2731 : return;
2732 :
2733 0 : if (pmsg->outbuf_len + len >= outbuf_sz) {
2734 : /* Truncate the message, but keep it terminated. */
2735 0 : pmsg->outbuf_len = outbuf_sz - (len + 1);
2736 0 : outbuf[pmsg->outbuf_len] = 0;
2737 : }
2738 :
2739 0 : memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2740 0 : memcpy(outbuf, scratchbuf, len);
2741 0 : pmsg->outbuf_len += len;
2742 : }
2743 : #else
2744 : #define console_prepend_dropped(pmsg, dropped)
2745 : #endif /* CONFIG_PRINTK */
2746 :
2747 : /*
2748 : * Read and format the specified record (or a later record if the specified
2749 : * record is not available).
2750 : *
2751 : * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2752 : * struct printk_buffers.
2753 : *
2754 : * @seq is the record to read and format. If it is not available, the next
2755 : * valid record is read.
2756 : *
2757 : * @is_extended specifies if the message should be formatted for extended
2758 : * console output.
2759 : *
2760 : * @may_supress specifies if records may be skipped based on loglevel.
2761 : *
2762 : * Returns false if no record is available. Otherwise true and all fields
2763 : * of @pmsg are valid. (See the documentation of struct printk_message
2764 : * for information about the @pmsg fields.)
2765 : */
2766 2757 : static bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2767 : bool is_extended, bool may_suppress)
2768 : {
2769 : static int panic_console_dropped;
2770 :
2771 2757 : struct printk_buffers *pbufs = pmsg->pbufs;
2772 2757 : const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2773 2757 : const size_t outbuf_sz = sizeof(pbufs->outbuf);
2774 2757 : char *scratchbuf = &pbufs->scratchbuf[0];
2775 2757 : char *outbuf = &pbufs->outbuf[0];
2776 : struct printk_info info;
2777 : struct printk_record r;
2778 2757 : size_t len = 0;
2779 :
2780 : /*
2781 : * Formatting extended messages requires a separate buffer, so use the
2782 : * scratch buffer to read in the ringbuffer text.
2783 : *
2784 : * Formatting normal messages is done in-place, so read the ringbuffer
2785 : * text directly into the output buffer.
2786 : */
2787 2757 : if (is_extended)
2788 0 : prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2789 : else
2790 2757 : prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2791 :
2792 2757 : if (!prb_read_valid(prb, seq, &r))
2793 : return false;
2794 :
2795 1388 : pmsg->seq = r.info->seq;
2796 1388 : pmsg->dropped = r.info->seq - seq;
2797 :
2798 : /*
2799 : * Check for dropped messages in panic here so that printk
2800 : * suppression can occur as early as possible if necessary.
2801 : */
2802 1388 : if (pmsg->dropped &&
2803 0 : panic_in_progress() &&
2804 0 : panic_console_dropped++ > 10) {
2805 0 : suppress_panic_printk = 1;
2806 0 : pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2807 : }
2808 :
2809 : /* Skip record that has level above the console loglevel. */
2810 2776 : if (may_suppress && suppress_message_printing(r.info->level))
2811 : goto out;
2812 :
2813 1386 : if (is_extended) {
2814 0 : len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2815 0 : len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2816 0 : &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2817 : } else {
2818 1386 : len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2819 : }
2820 : out:
2821 1388 : pmsg->outbuf_len = len;
2822 1388 : return true;
2823 : }
2824 :
2825 : /*
2826 : * Print one record for the given console. The record printed is whatever
2827 : * record is the next available record for the given console.
2828 : *
2829 : * @handover will be set to true if a printk waiter has taken over the
2830 : * console_lock, in which case the caller is no longer holding both the
2831 : * console_lock and the SRCU read lock. Otherwise it is set to false.
2832 : *
2833 : * @cookie is the cookie from the SRCU read lock.
2834 : *
2835 : * Returns false if the given console has no next record to print, otherwise
2836 : * true.
2837 : *
2838 : * Requires the console_lock and the SRCU read lock.
2839 : */
2840 2757 : static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2841 : {
2842 : static struct printk_buffers pbufs;
2843 :
2844 2757 : bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2845 2757 : char *outbuf = &pbufs.outbuf[0];
2846 2757 : struct printk_message pmsg = {
2847 : .pbufs = &pbufs,
2848 : };
2849 : unsigned long flags;
2850 :
2851 2757 : *handover = false;
2852 :
2853 2757 : if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2854 : return false;
2855 :
2856 1388 : con->dropped += pmsg.dropped;
2857 :
2858 : /* Skip messages of formatted length 0. */
2859 1388 : if (pmsg.outbuf_len == 0) {
2860 2 : con->seq = pmsg.seq + 1;
2861 2 : goto skip;
2862 : }
2863 :
2864 1386 : if (con->dropped && !is_extended) {
2865 0 : console_prepend_dropped(&pmsg, con->dropped);
2866 0 : con->dropped = 0;
2867 : }
2868 :
2869 : /*
2870 : * While actively printing out messages, if another printk()
2871 : * were to occur on another CPU, it may wait for this one to
2872 : * finish. This task can not be preempted if there is a
2873 : * waiter waiting to take over.
2874 : *
2875 : * Interrupts are disabled because the hand over to a waiter
2876 : * must not be interrupted until the hand over is completed
2877 : * (@console_waiter is cleared).
2878 : */
2879 1386 : printk_safe_enter_irqsave(flags);
2880 1386 : console_lock_spinning_enable();
2881 :
2882 : /* Do not trace print latency. */
2883 : stop_critical_timings();
2884 :
2885 : /* Write everything out to the hardware. */
2886 1386 : con->write(con, outbuf, pmsg.outbuf_len);
2887 :
2888 : start_critical_timings();
2889 :
2890 1386 : con->seq = pmsg.seq + 1;
2891 :
2892 1386 : *handover = console_lock_spinning_disable_and_check(cookie);
2893 1386 : printk_safe_exit_irqrestore(flags);
2894 : skip:
2895 : return true;
2896 : }
2897 :
2898 : /*
2899 : * Print out all remaining records to all consoles.
2900 : *
2901 : * @do_cond_resched is set by the caller. It can be true only in schedulable
2902 : * context.
2903 : *
2904 : * @next_seq is set to the sequence number after the last available record.
2905 : * The value is valid only when this function returns true. It means that all
2906 : * usable consoles are completely flushed.
2907 : *
2908 : * @handover will be set to true if a printk waiter has taken over the
2909 : * console_lock, in which case the caller is no longer holding the
2910 : * console_lock. Otherwise it is set to false.
2911 : *
2912 : * Returns true when there was at least one usable console and all messages
2913 : * were flushed to all usable consoles. A returned false informs the caller
2914 : * that everything was not flushed (either there were no usable consoles or
2915 : * another context has taken over printing or it is a panic situation and this
2916 : * is not the panic CPU). Regardless the reason, the caller should assume it
2917 : * is not useful to immediately try again.
2918 : *
2919 : * Requires the console_lock.
2920 : */
2921 811 : static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2922 : {
2923 811 : bool any_usable = false;
2924 : struct console *con;
2925 : bool any_progress;
2926 : int cookie;
2927 :
2928 811 : *next_seq = 0;
2929 811 : *handover = false;
2930 :
2931 : do {
2932 1559 : any_progress = false;
2933 :
2934 1559 : cookie = console_srcu_read_lock();
2935 4316 : for_each_console_srcu(con) {
2936 : bool progress;
2937 :
2938 2757 : if (!console_is_usable(con))
2939 0 : continue;
2940 2757 : any_usable = true;
2941 :
2942 2757 : progress = console_emit_next_record(con, handover, cookie);
2943 :
2944 : /*
2945 : * If a handover has occurred, the SRCU read lock
2946 : * is already released.
2947 : */
2948 2757 : if (*handover)
2949 : return false;
2950 :
2951 : /* Track the next of the highest seq flushed. */
2952 2757 : if (con->seq > *next_seq)
2953 839 : *next_seq = con->seq;
2954 :
2955 2757 : if (!progress)
2956 1369 : continue;
2957 1388 : any_progress = true;
2958 :
2959 : /* Allow panic_cpu to take over the consoles safely. */
2960 1388 : if (abandon_console_lock_in_panic())
2961 : goto abandon;
2962 :
2963 1388 : if (do_cond_resched)
2964 0 : cond_resched();
2965 : }
2966 1559 : console_srcu_read_unlock(cookie);
2967 1559 : } while (any_progress);
2968 :
2969 : return any_usable;
2970 :
2971 : abandon:
2972 0 : console_srcu_read_unlock(cookie);
2973 0 : return false;
2974 : }
2975 :
2976 : /**
2977 : * console_unlock - unblock the console subsystem from printing
2978 : *
2979 : * Releases the console_lock which the caller holds to block printing of
2980 : * the console subsystem.
2981 : *
2982 : * While the console_lock was held, console output may have been buffered
2983 : * by printk(). If this is the case, console_unlock(); emits
2984 : * the output prior to releasing the lock.
2985 : *
2986 : * console_unlock(); may be called from any context.
2987 : */
2988 811 : void console_unlock(void)
2989 : {
2990 : bool do_cond_resched;
2991 : bool handover;
2992 : bool flushed;
2993 : u64 next_seq;
2994 :
2995 811 : if (console_suspended) {
2996 0 : up_console_sem();
2997 0 : return;
2998 : }
2999 :
3000 : /*
3001 : * Console drivers are called with interrupts disabled, so
3002 : * @console_may_schedule should be cleared before; however, we may
3003 : * end up dumping a lot of lines, for example, if called from
3004 : * console registration path, and should invoke cond_resched()
3005 : * between lines if allowable. Not doing so can cause a very long
3006 : * scheduling stall on a slow console leading to RCU stall and
3007 : * softlockup warnings which exacerbate the issue with more
3008 : * messages practically incapacitating the system. Therefore, create
3009 : * a local to use for the printing loop.
3010 : */
3011 811 : do_cond_resched = console_may_schedule;
3012 :
3013 : do {
3014 811 : console_may_schedule = 0;
3015 :
3016 811 : flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3017 811 : if (!handover)
3018 : __console_unlock();
3019 :
3020 : /*
3021 : * Abort if there was a failure to flush all messages to all
3022 : * usable consoles. Either it is not possible to flush (in
3023 : * which case it would be an infinite loop of retrying) or
3024 : * another context has taken over printing.
3025 : */
3026 811 : if (!flushed)
3027 : break;
3028 :
3029 : /*
3030 : * Some context may have added new records after
3031 : * console_flush_all() but before unlocking the console.
3032 : * Re-check if there is a new record to flush. If the trylock
3033 : * fails, another context is already handling the printing.
3034 : */
3035 685 : } while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3036 : }
3037 : EXPORT_SYMBOL(console_unlock);
3038 :
3039 : /**
3040 : * console_conditional_schedule - yield the CPU if required
3041 : *
3042 : * If the console code is currently allowed to sleep, and
3043 : * if this CPU should yield the CPU to another task, do
3044 : * so here.
3045 : *
3046 : * Must be called within console_lock();.
3047 : */
3048 0 : void __sched console_conditional_schedule(void)
3049 : {
3050 0 : if (console_may_schedule)
3051 0 : cond_resched();
3052 0 : }
3053 : EXPORT_SYMBOL(console_conditional_schedule);
3054 :
3055 0 : void console_unblank(void)
3056 : {
3057 : struct console *c;
3058 : int cookie;
3059 :
3060 : /*
3061 : * Stop console printing because the unblank() callback may
3062 : * assume the console is not within its write() callback.
3063 : *
3064 : * If @oops_in_progress is set, this may be an atomic context.
3065 : * In that case, attempt a trylock as best-effort.
3066 : */
3067 0 : if (oops_in_progress) {
3068 0 : if (down_trylock_console_sem() != 0)
3069 : return;
3070 : } else
3071 : console_lock();
3072 :
3073 0 : console_locked = 1;
3074 0 : console_may_schedule = 0;
3075 :
3076 0 : cookie = console_srcu_read_lock();
3077 0 : for_each_console_srcu(c) {
3078 0 : if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3079 0 : c->unblank();
3080 : }
3081 0 : console_srcu_read_unlock(cookie);
3082 :
3083 0 : console_unlock();
3084 :
3085 0 : if (!oops_in_progress)
3086 : pr_flush(1000, true);
3087 : }
3088 :
3089 : /**
3090 : * console_flush_on_panic - flush console content on panic
3091 : * @mode: flush all messages in buffer or just the pending ones
3092 : *
3093 : * Immediately output all pending messages no matter what.
3094 : */
3095 0 : void console_flush_on_panic(enum con_flush_mode mode)
3096 : {
3097 : /*
3098 : * If someone else is holding the console lock, trylock will fail
3099 : * and may_schedule may be set. Ignore and proceed to unlock so
3100 : * that messages are flushed out. As this can be called from any
3101 : * context and we don't want to get preempted while flushing,
3102 : * ensure may_schedule is cleared.
3103 : */
3104 0 : console_trylock();
3105 0 : console_may_schedule = 0;
3106 :
3107 0 : if (mode == CONSOLE_REPLAY_ALL) {
3108 : struct console *c;
3109 : int cookie;
3110 : u64 seq;
3111 :
3112 0 : seq = prb_first_valid_seq(prb);
3113 :
3114 0 : cookie = console_srcu_read_lock();
3115 0 : for_each_console_srcu(c) {
3116 : /*
3117 : * If the above console_trylock() failed, this is an
3118 : * unsynchronized assignment. But in that case, the
3119 : * kernel is in "hope and pray" mode anyway.
3120 : */
3121 0 : c->seq = seq;
3122 : }
3123 : console_srcu_read_unlock(cookie);
3124 : }
3125 0 : console_unlock();
3126 0 : }
3127 :
3128 : /*
3129 : * Return the console tty driver structure and its associated index
3130 : */
3131 0 : struct tty_driver *console_device(int *index)
3132 : {
3133 : struct console *c;
3134 0 : struct tty_driver *driver = NULL;
3135 : int cookie;
3136 :
3137 : /*
3138 : * Take console_lock to serialize device() callback with
3139 : * other console operations. For example, fg_console is
3140 : * modified under console_lock when switching vt.
3141 : */
3142 : console_lock();
3143 :
3144 0 : cookie = console_srcu_read_lock();
3145 0 : for_each_console_srcu(c) {
3146 0 : if (!c->device)
3147 0 : continue;
3148 0 : driver = c->device(c, index);
3149 0 : if (driver)
3150 : break;
3151 : }
3152 0 : console_srcu_read_unlock(cookie);
3153 :
3154 0 : console_unlock();
3155 0 : return driver;
3156 : }
3157 :
3158 : /*
3159 : * Prevent further output on the passed console device so that (for example)
3160 : * serial drivers can disable console output before suspending a port, and can
3161 : * re-enable output afterwards.
3162 : */
3163 0 : void console_stop(struct console *console)
3164 : {
3165 0 : __pr_flush(console, 1000, true);
3166 : console_list_lock();
3167 0 : console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3168 : console_list_unlock();
3169 :
3170 : /*
3171 : * Ensure that all SRCU list walks have completed. All contexts must
3172 : * be able to see that this console is disabled so that (for example)
3173 : * the caller can suspend the port without risk of another context
3174 : * using the port.
3175 : */
3176 0 : synchronize_srcu(&console_srcu);
3177 0 : }
3178 : EXPORT_SYMBOL(console_stop);
3179 :
3180 0 : void console_start(struct console *console)
3181 : {
3182 : console_list_lock();
3183 0 : console_srcu_write_flags(console, console->flags | CON_ENABLED);
3184 : console_list_unlock();
3185 0 : __pr_flush(console, 1000, true);
3186 0 : }
3187 : EXPORT_SYMBOL(console_start);
3188 :
3189 : static int __read_mostly keep_bootcon;
3190 :
3191 0 : static int __init keep_bootcon_setup(char *str)
3192 : {
3193 0 : keep_bootcon = 1;
3194 0 : pr_info("debug: skip boot console de-registration.\n");
3195 :
3196 0 : return 0;
3197 : }
3198 :
3199 : early_param("keep_bootcon", keep_bootcon_setup);
3200 :
3201 : /*
3202 : * This is called by register_console() to try to match
3203 : * the newly registered console with any of the ones selected
3204 : * by either the command line or add_preferred_console() and
3205 : * setup/enable it.
3206 : *
3207 : * Care need to be taken with consoles that are statically
3208 : * enabled such as netconsole
3209 : */
3210 3 : static int try_enable_preferred_console(struct console *newcon,
3211 : bool user_specified)
3212 : {
3213 : struct console_cmdline *c;
3214 : int i, err;
3215 :
3216 8 : for (i = 0, c = console_cmdline;
3217 5 : i < MAX_CMDLINECONSOLES && c->name[0];
3218 2 : i++, c++) {
3219 3 : if (c->user_specified != user_specified)
3220 1 : continue;
3221 2 : if (!newcon->match ||
3222 0 : newcon->match(newcon, c->name, c->index, c->options) != 0) {
3223 : /* default matching */
3224 : BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3225 2 : if (strcmp(c->name, newcon->name) != 0)
3226 1 : continue;
3227 1 : if (newcon->index >= 0 &&
3228 0 : newcon->index != c->index)
3229 0 : continue;
3230 1 : if (newcon->index < 0)
3231 1 : newcon->index = c->index;
3232 :
3233 1 : if (_braille_register_console(newcon, c))
3234 : return 0;
3235 :
3236 2 : if (newcon->setup &&
3237 1 : (err = newcon->setup(newcon, c->options)) != 0)
3238 : return err;
3239 : }
3240 1 : newcon->flags |= CON_ENABLED;
3241 1 : if (i == preferred_console)
3242 1 : newcon->flags |= CON_CONSDEV;
3243 : return 0;
3244 : }
3245 :
3246 : /*
3247 : * Some consoles, such as pstore and netconsole, can be enabled even
3248 : * without matching. Accept the pre-enabled consoles only when match()
3249 : * and setup() had a chance to be called.
3250 : */
3251 2 : if (newcon->flags & CON_ENABLED && c->user_specified == user_specified)
3252 : return 0;
3253 :
3254 1 : return -ENOENT;
3255 : }
3256 :
3257 : /* Try to enable the console unconditionally */
3258 0 : static void try_enable_default_console(struct console *newcon)
3259 : {
3260 0 : if (newcon->index < 0)
3261 0 : newcon->index = 0;
3262 :
3263 0 : if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3264 : return;
3265 :
3266 0 : newcon->flags |= CON_ENABLED;
3267 :
3268 0 : if (newcon->device)
3269 0 : newcon->flags |= CON_CONSDEV;
3270 : }
3271 :
3272 : #define con_printk(lvl, con, fmt, ...) \
3273 : printk(lvl pr_fmt("%sconsole [%s%d] " fmt), \
3274 : (con->flags & CON_BOOT) ? "boot" : "", \
3275 : con->name, con->index, ##__VA_ARGS__)
3276 :
3277 2 : static void console_init_seq(struct console *newcon, bool bootcon_registered)
3278 : {
3279 : struct console *con;
3280 : bool handover;
3281 :
3282 2 : if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3283 : /* Get a consistent copy of @syslog_seq. */
3284 1 : mutex_lock(&syslog_lock);
3285 1 : newcon->seq = syslog_seq;
3286 1 : mutex_unlock(&syslog_lock);
3287 : } else {
3288 : /* Begin with next message added to ringbuffer. */
3289 1 : newcon->seq = prb_next_seq(prb);
3290 :
3291 : /*
3292 : * If any enabled boot consoles are due to be unregistered
3293 : * shortly, some may not be caught up and may be the same
3294 : * device as @newcon. Since it is not known which boot console
3295 : * is the same device, flush all consoles and, if necessary,
3296 : * start with the message of the enabled boot console that is
3297 : * the furthest behind.
3298 : */
3299 1 : if (bootcon_registered && !keep_bootcon) {
3300 : /*
3301 : * Hold the console_lock to stop console printing and
3302 : * guarantee safe access to console->seq.
3303 : */
3304 : console_lock();
3305 :
3306 : /*
3307 : * Flush all consoles and set the console to start at
3308 : * the next unprinted sequence number.
3309 : */
3310 0 : if (!console_flush_all(true, &newcon->seq, &handover)) {
3311 : /*
3312 : * Flushing failed. Just choose the lowest
3313 : * sequence of the enabled boot consoles.
3314 : */
3315 :
3316 : /*
3317 : * If there was a handover, this context no
3318 : * longer holds the console_lock.
3319 : */
3320 0 : if (handover)
3321 : console_lock();
3322 :
3323 0 : newcon->seq = prb_next_seq(prb);
3324 0 : for_each_console(con) {
3325 0 : if ((con->flags & CON_BOOT) &&
3326 0 : (con->flags & CON_ENABLED) &&
3327 0 : con->seq < newcon->seq) {
3328 0 : newcon->seq = con->seq;
3329 : }
3330 : }
3331 : }
3332 :
3333 0 : console_unlock();
3334 : }
3335 : }
3336 2 : }
3337 :
3338 : #define console_first() \
3339 : hlist_entry(console_list.first, struct console, node)
3340 :
3341 : static int unregister_console_locked(struct console *console);
3342 :
3343 : /*
3344 : * The console driver calls this routine during kernel initialization
3345 : * to register the console printing procedure with printk() and to
3346 : * print any messages that were printed by the kernel before the
3347 : * console driver was initialized.
3348 : *
3349 : * This can happen pretty early during the boot process (because of
3350 : * early_printk) - sometimes before setup_arch() completes - be careful
3351 : * of what kernel features are used - they may not be initialised yet.
3352 : *
3353 : * There are two types of consoles - bootconsoles (early_printk) and
3354 : * "real" consoles (everything which is not a bootconsole) which are
3355 : * handled differently.
3356 : * - Any number of bootconsoles can be registered at any time.
3357 : * - As soon as a "real" console is registered, all bootconsoles
3358 : * will be unregistered automatically.
3359 : * - Once a "real" console is registered, any attempt to register a
3360 : * bootconsoles will be rejected
3361 : */
3362 2 : void register_console(struct console *newcon)
3363 : {
3364 : struct console *con;
3365 2 : bool bootcon_registered = false;
3366 2 : bool realcon_registered = false;
3367 : int err;
3368 :
3369 : console_list_lock();
3370 :
3371 3 : for_each_console(con) {
3372 1 : if (WARN(con == newcon, "console '%s%d' already registered\n",
3373 : con->name, con->index)) {
3374 : goto unlock;
3375 : }
3376 :
3377 1 : if (con->flags & CON_BOOT)
3378 : bootcon_registered = true;
3379 : else
3380 1 : realcon_registered = true;
3381 : }
3382 :
3383 : /* Do not register boot consoles when there already is a real one. */
3384 2 : if ((newcon->flags & CON_BOOT) && realcon_registered) {
3385 0 : pr_info("Too late to register bootconsole %s%d\n",
3386 : newcon->name, newcon->index);
3387 0 : goto unlock;
3388 : }
3389 :
3390 : /*
3391 : * See if we want to enable this console driver by default.
3392 : *
3393 : * Nope when a console is preferred by the command line, device
3394 : * tree, or SPCR.
3395 : *
3396 : * The first real console with tty binding (driver) wins. More
3397 : * consoles might get enabled before the right one is found.
3398 : *
3399 : * Note that a console with tty binding will have CON_CONSDEV
3400 : * flag set and will be first in the list.
3401 : */
3402 2 : if (preferred_console < 0) {
3403 0 : if (hlist_empty(&console_list) || !console_first()->device ||
3404 0 : console_first()->flags & CON_BOOT) {
3405 0 : try_enable_default_console(newcon);
3406 : }
3407 : }
3408 :
3409 : /* See if this console matches one we selected on the command line */
3410 2 : err = try_enable_preferred_console(newcon, true);
3411 :
3412 : /* If not, try to match against the platform default(s) */
3413 2 : if (err == -ENOENT)
3414 1 : err = try_enable_preferred_console(newcon, false);
3415 :
3416 : /* printk() messages are not printed to the Braille console. */
3417 2 : if (err || newcon->flags & CON_BRL)
3418 : goto unlock;
3419 :
3420 : /*
3421 : * If we have a bootconsole, and are switching to a real console,
3422 : * don't print everything out again, since when the boot console, and
3423 : * the real console are the same physical device, it's annoying to
3424 : * see the beginning boot messages twice
3425 : */
3426 2 : if (bootcon_registered &&
3427 : ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3428 0 : newcon->flags &= ~CON_PRINTBUFFER;
3429 : }
3430 :
3431 2 : newcon->dropped = 0;
3432 2 : console_init_seq(newcon, bootcon_registered);
3433 :
3434 : /*
3435 : * Put this console in the list - keep the
3436 : * preferred driver at the head of the list.
3437 : */
3438 2 : if (hlist_empty(&console_list)) {
3439 : /* Ensure CON_CONSDEV is always set for the head. */
3440 1 : newcon->flags |= CON_CONSDEV;
3441 1 : hlist_add_head_rcu(&newcon->node, &console_list);
3442 :
3443 1 : } else if (newcon->flags & CON_CONSDEV) {
3444 : /* Only the new head can have CON_CONSDEV set. */
3445 0 : console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3446 0 : hlist_add_head_rcu(&newcon->node, &console_list);
3447 :
3448 : } else {
3449 1 : hlist_add_behind_rcu(&newcon->node, console_list.first);
3450 : }
3451 :
3452 : /*
3453 : * No need to synchronize SRCU here! The caller does not rely
3454 : * on all contexts being able to see the new console before
3455 : * register_console() completes.
3456 : */
3457 :
3458 2 : console_sysfs_notify();
3459 :
3460 : /*
3461 : * By unregistering the bootconsoles after we enable the real console
3462 : * we get the "console xxx enabled" message on all the consoles -
3463 : * boot consoles, real consoles, etc - this is to ensure that end
3464 : * users know there might be something in the kernel's log buffer that
3465 : * went to the bootconsole (that they do not see on the real console)
3466 : */
3467 2 : con_printk(KERN_INFO, newcon, "enabled\n");
3468 2 : if (bootcon_registered &&
3469 0 : ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3470 0 : !keep_bootcon) {
3471 : struct hlist_node *tmp;
3472 :
3473 0 : hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3474 0 : if (con->flags & CON_BOOT)
3475 0 : unregister_console_locked(con);
3476 : }
3477 : }
3478 : unlock:
3479 : console_list_unlock();
3480 2 : }
3481 : EXPORT_SYMBOL(register_console);
3482 :
3483 : /* Must be called under console_list_lock(). */
3484 1 : static int unregister_console_locked(struct console *console)
3485 : {
3486 : int res;
3487 :
3488 : lockdep_assert_console_list_lock_held();
3489 :
3490 1 : con_printk(KERN_INFO, console, "disabled\n");
3491 :
3492 1 : res = _braille_unregister_console(console);
3493 : if (res < 0)
3494 : return res;
3495 : if (res > 0)
3496 : return 0;
3497 :
3498 : /* Disable it unconditionally */
3499 2 : console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3500 :
3501 1 : if (!console_is_registered_locked(console))
3502 : return -ENODEV;
3503 :
3504 0 : hlist_del_init_rcu(&console->node);
3505 :
3506 : /*
3507 : * <HISTORICAL>
3508 : * If this isn't the last console and it has CON_CONSDEV set, we
3509 : * need to set it on the next preferred console.
3510 : * </HISTORICAL>
3511 : *
3512 : * The above makes no sense as there is no guarantee that the next
3513 : * console has any device attached. Oh well....
3514 : */
3515 0 : if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3516 0 : console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3517 :
3518 : /*
3519 : * Ensure that all SRCU list walks have completed. All contexts
3520 : * must not be able to see this console in the list so that any
3521 : * exit/cleanup routines can be performed safely.
3522 : */
3523 0 : synchronize_srcu(&console_srcu);
3524 :
3525 0 : console_sysfs_notify();
3526 :
3527 0 : if (console->exit)
3528 0 : res = console->exit(console);
3529 :
3530 : return res;
3531 : }
3532 :
3533 1 : int unregister_console(struct console *console)
3534 : {
3535 : int res;
3536 :
3537 : console_list_lock();
3538 1 : res = unregister_console_locked(console);
3539 : console_list_unlock();
3540 1 : return res;
3541 : }
3542 : EXPORT_SYMBOL(unregister_console);
3543 :
3544 : /**
3545 : * console_force_preferred_locked - force a registered console preferred
3546 : * @con: The registered console to force preferred.
3547 : *
3548 : * Must be called under console_list_lock().
3549 : */
3550 0 : void console_force_preferred_locked(struct console *con)
3551 : {
3552 : struct console *cur_pref_con;
3553 :
3554 0 : if (!console_is_registered_locked(con))
3555 : return;
3556 :
3557 0 : cur_pref_con = console_first();
3558 :
3559 : /* Already preferred? */
3560 0 : if (cur_pref_con == con)
3561 : return;
3562 :
3563 : /*
3564 : * Delete, but do not re-initialize the entry. This allows the console
3565 : * to continue to appear registered (via any hlist_unhashed_lockless()
3566 : * checks), even though it was briefly removed from the console list.
3567 : */
3568 0 : hlist_del_rcu(&con->node);
3569 :
3570 : /*
3571 : * Ensure that all SRCU list walks have completed so that the console
3572 : * can be added to the beginning of the console list and its forward
3573 : * list pointer can be re-initialized.
3574 : */
3575 0 : synchronize_srcu(&console_srcu);
3576 :
3577 0 : con->flags |= CON_CONSDEV;
3578 0 : WARN_ON(!con->device);
3579 :
3580 : /* Only the new head can have CON_CONSDEV set. */
3581 0 : console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3582 0 : hlist_add_head_rcu(&con->node, &console_list);
3583 : }
3584 : EXPORT_SYMBOL(console_force_preferred_locked);
3585 :
3586 : /*
3587 : * Initialize the console device. This is called *early*, so
3588 : * we can't necessarily depend on lots of kernel help here.
3589 : * Just do some early initializations, and do the complex setup
3590 : * later.
3591 : */
3592 1 : void __init console_init(void)
3593 : {
3594 : int ret;
3595 : initcall_t call;
3596 : initcall_entry_t *ce;
3597 :
3598 : /* Setup the default TTY line discipline. */
3599 1 : n_tty_init();
3600 :
3601 : /*
3602 : * set up the console device so that later boot sequences can
3603 : * inform about problems etc..
3604 : */
3605 1 : ce = __con_initcall_start;
3606 1 : trace_initcall_level("console");
3607 2 : while (ce < __con_initcall_end) {
3608 1 : call = initcall_from_entry(ce);
3609 : trace_initcall_start(call);
3610 1 : ret = call();
3611 1 : trace_initcall_finish(call, ret);
3612 1 : ce++;
3613 : }
3614 1 : }
3615 :
3616 : /*
3617 : * Some boot consoles access data that is in the init section and which will
3618 : * be discarded after the initcalls have been run. To make sure that no code
3619 : * will access this data, unregister the boot consoles in a late initcall.
3620 : *
3621 : * If for some reason, such as deferred probe or the driver being a loadable
3622 : * module, the real console hasn't registered yet at this point, there will
3623 : * be a brief interval in which no messages are logged to the console, which
3624 : * makes it difficult to diagnose problems that occur during this time.
3625 : *
3626 : * To mitigate this problem somewhat, only unregister consoles whose memory
3627 : * intersects with the init section. Note that all other boot consoles will
3628 : * get unregistered when the real preferred console is registered.
3629 : */
3630 1 : static int __init printk_late_init(void)
3631 : {
3632 : struct hlist_node *tmp;
3633 : struct console *con;
3634 : int ret;
3635 :
3636 : console_list_lock();
3637 3 : hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3638 2 : if (!(con->flags & CON_BOOT))
3639 2 : continue;
3640 :
3641 : /* Check addresses that might be used for enabled consoles. */
3642 0 : if (init_section_intersects(con, sizeof(*con)) ||
3643 0 : init_section_contains(con->write, 0) ||
3644 0 : init_section_contains(con->read, 0) ||
3645 0 : init_section_contains(con->device, 0) ||
3646 0 : init_section_contains(con->unblank, 0) ||
3647 0 : init_section_contains(con->data, 0)) {
3648 : /*
3649 : * Please, consider moving the reported consoles out
3650 : * of the init section.
3651 : */
3652 0 : pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3653 : con->name, con->index);
3654 0 : unregister_console_locked(con);
3655 : }
3656 : }
3657 1 : console_list_unlock();
3658 :
3659 1 : ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3660 : console_cpu_notify);
3661 1 : WARN_ON(ret < 0);
3662 1 : ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3663 : console_cpu_notify, NULL);
3664 1 : WARN_ON(ret < 0);
3665 1 : printk_sysctl_init();
3666 1 : return 0;
3667 : }
3668 : late_initcall(printk_late_init);
3669 :
3670 : #if defined CONFIG_PRINTK
3671 : /* If @con is specified, only wait for that console. Otherwise wait for all. */
3672 0 : static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3673 : {
3674 0 : int remaining = timeout_ms;
3675 : struct console *c;
3676 0 : u64 last_diff = 0;
3677 : u64 printk_seq;
3678 : int cookie;
3679 : u64 diff;
3680 : u64 seq;
3681 :
3682 : might_sleep();
3683 :
3684 0 : seq = prb_next_seq(prb);
3685 :
3686 : for (;;) {
3687 0 : diff = 0;
3688 :
3689 : /*
3690 : * Hold the console_lock to guarantee safe access to
3691 : * console->seq and to prevent changes to @console_suspended
3692 : * until all consoles have been processed.
3693 : */
3694 : console_lock();
3695 :
3696 0 : cookie = console_srcu_read_lock();
3697 0 : for_each_console_srcu(c) {
3698 0 : if (con && con != c)
3699 0 : continue;
3700 0 : if (!console_is_usable(c))
3701 0 : continue;
3702 0 : printk_seq = c->seq;
3703 0 : if (printk_seq < seq)
3704 0 : diff += seq - printk_seq;
3705 : }
3706 0 : console_srcu_read_unlock(cookie);
3707 :
3708 : /*
3709 : * If consoles are suspended, it cannot be expected that they
3710 : * make forward progress, so timeout immediately. @diff is
3711 : * still used to return a valid flush status.
3712 : */
3713 0 : if (console_suspended)
3714 : remaining = 0;
3715 0 : else if (diff != last_diff && reset_on_progress)
3716 0 : remaining = timeout_ms;
3717 :
3718 0 : console_unlock();
3719 :
3720 0 : if (diff == 0 || remaining == 0)
3721 : break;
3722 :
3723 0 : if (remaining < 0) {
3724 : /* no timeout limit */
3725 0 : msleep(100);
3726 0 : } else if (remaining < 100) {
3727 0 : msleep(remaining);
3728 0 : remaining = 0;
3729 : } else {
3730 0 : msleep(100);
3731 0 : remaining -= 100;
3732 : }
3733 :
3734 : last_diff = diff;
3735 : }
3736 :
3737 0 : return (diff == 0);
3738 : }
3739 :
3740 : /**
3741 : * pr_flush() - Wait for printing threads to catch up.
3742 : *
3743 : * @timeout_ms: The maximum time (in ms) to wait.
3744 : * @reset_on_progress: Reset the timeout if forward progress is seen.
3745 : *
3746 : * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3747 : * represents infinite waiting.
3748 : *
3749 : * If @reset_on_progress is true, the timeout will be reset whenever any
3750 : * printer has been seen to make some forward progress.
3751 : *
3752 : * Context: Process context. May sleep while acquiring console lock.
3753 : * Return: true if all enabled printers are caught up.
3754 : */
3755 : static bool pr_flush(int timeout_ms, bool reset_on_progress)
3756 : {
3757 0 : return __pr_flush(NULL, timeout_ms, reset_on_progress);
3758 : }
3759 :
3760 : /*
3761 : * Delayed printk version, for scheduler-internal messages:
3762 : */
3763 : #define PRINTK_PENDING_WAKEUP 0x01
3764 : #define PRINTK_PENDING_OUTPUT 0x02
3765 :
3766 : static DEFINE_PER_CPU(int, printk_pending);
3767 :
3768 0 : static void wake_up_klogd_work_func(struct irq_work *irq_work)
3769 : {
3770 0 : int pending = this_cpu_xchg(printk_pending, 0);
3771 :
3772 0 : if (pending & PRINTK_PENDING_OUTPUT) {
3773 : /* If trylock fails, someone else is doing the printing */
3774 0 : if (console_trylock())
3775 0 : console_unlock();
3776 : }
3777 :
3778 0 : if (pending & PRINTK_PENDING_WAKEUP)
3779 0 : wake_up_interruptible(&log_wait);
3780 0 : }
3781 :
3782 : static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3783 : IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3784 :
3785 811 : static void __wake_up_klogd(int val)
3786 : {
3787 811 : if (!printk_percpu_data_ready())
3788 : return;
3789 :
3790 793 : preempt_disable();
3791 : /*
3792 : * Guarantee any new records can be seen by tasks preparing to wait
3793 : * before this context checks if the wait queue is empty.
3794 : *
3795 : * The full memory barrier within wq_has_sleeper() pairs with the full
3796 : * memory barrier within set_current_state() of
3797 : * prepare_to_wait_event(), which is called after ___wait_event() adds
3798 : * the waiter but before it has checked the wait condition.
3799 : *
3800 : * This pairs with devkmsg_read:A and syslog_print:A.
3801 : */
3802 1586 : if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3803 793 : (val & PRINTK_PENDING_OUTPUT)) {
3804 0 : this_cpu_or(printk_pending, val);
3805 0 : irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3806 : }
3807 793 : preempt_enable();
3808 : }
3809 :
3810 0 : void wake_up_klogd(void)
3811 : {
3812 811 : __wake_up_klogd(PRINTK_PENDING_WAKEUP);
3813 0 : }
3814 :
3815 0 : void defer_console_output(void)
3816 : {
3817 : /*
3818 : * New messages may have been added directly to the ringbuffer
3819 : * using vprintk_store(), so wake any waiters as well.
3820 : */
3821 0 : __wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3822 0 : }
3823 :
3824 0 : void printk_trigger_flush(void)
3825 : {
3826 : defer_console_output();
3827 0 : }
3828 :
3829 0 : int vprintk_deferred(const char *fmt, va_list args)
3830 : {
3831 : int r;
3832 :
3833 0 : r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3834 : defer_console_output();
3835 :
3836 0 : return r;
3837 : }
3838 :
3839 0 : int _printk_deferred(const char *fmt, ...)
3840 : {
3841 : va_list args;
3842 : int r;
3843 :
3844 0 : va_start(args, fmt);
3845 0 : r = vprintk_deferred(fmt, args);
3846 0 : va_end(args);
3847 :
3848 0 : return r;
3849 : }
3850 :
3851 : /*
3852 : * printk rate limiting, lifted from the networking subsystem.
3853 : *
3854 : * This enforces a rate limit: not more than 10 kernel messages
3855 : * every 5s to make a denial-of-service attack impossible.
3856 : */
3857 : DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3858 :
3859 0 : int __printk_ratelimit(const char *func)
3860 : {
3861 0 : return ___ratelimit(&printk_ratelimit_state, func);
3862 : }
3863 : EXPORT_SYMBOL(__printk_ratelimit);
3864 :
3865 : /**
3866 : * printk_timed_ratelimit - caller-controlled printk ratelimiting
3867 : * @caller_jiffies: pointer to caller's state
3868 : * @interval_msecs: minimum interval between prints
3869 : *
3870 : * printk_timed_ratelimit() returns true if more than @interval_msecs
3871 : * milliseconds have elapsed since the last time printk_timed_ratelimit()
3872 : * returned true.
3873 : */
3874 0 : bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3875 : unsigned int interval_msecs)
3876 : {
3877 0 : unsigned long elapsed = jiffies - *caller_jiffies;
3878 :
3879 0 : if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3880 : return false;
3881 :
3882 0 : *caller_jiffies = jiffies;
3883 0 : return true;
3884 : }
3885 : EXPORT_SYMBOL(printk_timed_ratelimit);
3886 :
3887 : static DEFINE_SPINLOCK(dump_list_lock);
3888 : static LIST_HEAD(dump_list);
3889 :
3890 : /**
3891 : * kmsg_dump_register - register a kernel log dumper.
3892 : * @dumper: pointer to the kmsg_dumper structure
3893 : *
3894 : * Adds a kernel log dumper to the system. The dump callback in the
3895 : * structure will be called when the kernel oopses or panics and must be
3896 : * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3897 : */
3898 1 : int kmsg_dump_register(struct kmsg_dumper *dumper)
3899 : {
3900 : unsigned long flags;
3901 1 : int err = -EBUSY;
3902 :
3903 : /* The dump callback needs to be set */
3904 1 : if (!dumper->dump)
3905 : return -EINVAL;
3906 :
3907 1 : spin_lock_irqsave(&dump_list_lock, flags);
3908 : /* Don't allow registering multiple times */
3909 1 : if (!dumper->registered) {
3910 1 : dumper->registered = 1;
3911 2 : list_add_tail_rcu(&dumper->list, &dump_list);
3912 1 : err = 0;
3913 : }
3914 1 : spin_unlock_irqrestore(&dump_list_lock, flags);
3915 :
3916 1 : return err;
3917 : }
3918 : EXPORT_SYMBOL_GPL(kmsg_dump_register);
3919 :
3920 : /**
3921 : * kmsg_dump_unregister - unregister a kmsg dumper.
3922 : * @dumper: pointer to the kmsg_dumper structure
3923 : *
3924 : * Removes a dump device from the system. Returns zero on success and
3925 : * %-EINVAL otherwise.
3926 : */
3927 0 : int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3928 : {
3929 : unsigned long flags;
3930 0 : int err = -EINVAL;
3931 :
3932 0 : spin_lock_irqsave(&dump_list_lock, flags);
3933 0 : if (dumper->registered) {
3934 0 : dumper->registered = 0;
3935 0 : list_del_rcu(&dumper->list);
3936 0 : err = 0;
3937 : }
3938 0 : spin_unlock_irqrestore(&dump_list_lock, flags);
3939 0 : synchronize_rcu();
3940 :
3941 0 : return err;
3942 : }
3943 : EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3944 :
3945 : static bool always_kmsg_dump;
3946 : module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3947 :
3948 0 : const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3949 : {
3950 : switch (reason) {
3951 : case KMSG_DUMP_PANIC:
3952 : return "Panic";
3953 : case KMSG_DUMP_OOPS:
3954 : return "Oops";
3955 : case KMSG_DUMP_EMERG:
3956 : return "Emergency";
3957 : case KMSG_DUMP_SHUTDOWN:
3958 : return "Shutdown";
3959 : default:
3960 : return "Unknown";
3961 : }
3962 : }
3963 : EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3964 :
3965 : /**
3966 : * kmsg_dump - dump kernel log to kernel message dumpers.
3967 : * @reason: the reason (oops, panic etc) for dumping
3968 : *
3969 : * Call each of the registered dumper's dump() callback, which can
3970 : * retrieve the kmsg records with kmsg_dump_get_line() or
3971 : * kmsg_dump_get_buffer().
3972 : */
3973 1 : void kmsg_dump(enum kmsg_dump_reason reason)
3974 : {
3975 : struct kmsg_dumper *dumper;
3976 :
3977 : rcu_read_lock();
3978 2 : list_for_each_entry_rcu(dumper, &dump_list, list) {
3979 1 : enum kmsg_dump_reason max_reason = dumper->max_reason;
3980 :
3981 : /*
3982 : * If client has not provided a specific max_reason, default
3983 : * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3984 : */
3985 1 : if (max_reason == KMSG_DUMP_UNDEF) {
3986 1 : max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3987 : KMSG_DUMP_OOPS;
3988 : }
3989 1 : if (reason > max_reason)
3990 1 : continue;
3991 :
3992 : /* invoke dumper which will iterate over records */
3993 0 : dumper->dump(dumper, reason);
3994 : }
3995 : rcu_read_unlock();
3996 1 : }
3997 :
3998 : /**
3999 : * kmsg_dump_get_line - retrieve one kmsg log line
4000 : * @iter: kmsg dump iterator
4001 : * @syslog: include the "<4>" prefixes
4002 : * @line: buffer to copy the line to
4003 : * @size: maximum size of the buffer
4004 : * @len: length of line placed into buffer
4005 : *
4006 : * Start at the beginning of the kmsg buffer, with the oldest kmsg
4007 : * record, and copy one record into the provided buffer.
4008 : *
4009 : * Consecutive calls will return the next available record moving
4010 : * towards the end of the buffer with the youngest messages.
4011 : *
4012 : * A return value of FALSE indicates that there are no more records to
4013 : * read.
4014 : */
4015 0 : bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4016 : char *line, size_t size, size_t *len)
4017 : {
4018 0 : u64 min_seq = latched_seq_read_nolock(&clear_seq);
4019 : struct printk_info info;
4020 : unsigned int line_count;
4021 : struct printk_record r;
4022 0 : size_t l = 0;
4023 0 : bool ret = false;
4024 :
4025 0 : if (iter->cur_seq < min_seq)
4026 0 : iter->cur_seq = min_seq;
4027 :
4028 0 : prb_rec_init_rd(&r, &info, line, size);
4029 :
4030 : /* Read text or count text lines? */
4031 0 : if (line) {
4032 0 : if (!prb_read_valid(prb, iter->cur_seq, &r))
4033 : goto out;
4034 0 : l = record_print_text(&r, syslog, printk_time);
4035 : } else {
4036 0 : if (!prb_read_valid_info(prb, iter->cur_seq,
4037 : &info, &line_count)) {
4038 : goto out;
4039 : }
4040 0 : l = get_record_print_text_size(&info, line_count, syslog,
4041 : printk_time);
4042 :
4043 : }
4044 :
4045 0 : iter->cur_seq = r.info->seq + 1;
4046 0 : ret = true;
4047 : out:
4048 0 : if (len)
4049 0 : *len = l;
4050 0 : return ret;
4051 : }
4052 : EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4053 :
4054 : /**
4055 : * kmsg_dump_get_buffer - copy kmsg log lines
4056 : * @iter: kmsg dump iterator
4057 : * @syslog: include the "<4>" prefixes
4058 : * @buf: buffer to copy the line to
4059 : * @size: maximum size of the buffer
4060 : * @len_out: length of line placed into buffer
4061 : *
4062 : * Start at the end of the kmsg buffer and fill the provided buffer
4063 : * with as many of the *youngest* kmsg records that fit into it.
4064 : * If the buffer is large enough, all available kmsg records will be
4065 : * copied with a single call.
4066 : *
4067 : * Consecutive calls will fill the buffer with the next block of
4068 : * available older records, not including the earlier retrieved ones.
4069 : *
4070 : * A return value of FALSE indicates that there are no more records to
4071 : * read.
4072 : */
4073 0 : bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4074 : char *buf, size_t size, size_t *len_out)
4075 : {
4076 0 : u64 min_seq = latched_seq_read_nolock(&clear_seq);
4077 : struct printk_info info;
4078 : struct printk_record r;
4079 : u64 seq;
4080 : u64 next_seq;
4081 0 : size_t len = 0;
4082 0 : bool ret = false;
4083 0 : bool time = printk_time;
4084 :
4085 0 : if (!buf || !size)
4086 : goto out;
4087 :
4088 0 : if (iter->cur_seq < min_seq)
4089 0 : iter->cur_seq = min_seq;
4090 :
4091 0 : if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4092 0 : if (info.seq != iter->cur_seq) {
4093 : /* messages are gone, move to first available one */
4094 0 : iter->cur_seq = info.seq;
4095 : }
4096 : }
4097 :
4098 : /* last entry */
4099 0 : if (iter->cur_seq >= iter->next_seq)
4100 : goto out;
4101 :
4102 : /*
4103 : * Find first record that fits, including all following records,
4104 : * into the user-provided buffer for this dump. Pass in size-1
4105 : * because this function (by way of record_print_text()) will
4106 : * not write more than size-1 bytes of text into @buf.
4107 : */
4108 0 : seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4109 : size - 1, syslog, time);
4110 :
4111 : /*
4112 : * Next kmsg_dump_get_buffer() invocation will dump block of
4113 : * older records stored right before this one.
4114 : */
4115 0 : next_seq = seq;
4116 :
4117 0 : prb_rec_init_rd(&r, &info, buf, size);
4118 :
4119 0 : len = 0;
4120 0 : prb_for_each_record(seq, prb, seq, &r) {
4121 0 : if (r.info->seq >= iter->next_seq)
4122 : break;
4123 :
4124 0 : len += record_print_text(&r, syslog, time);
4125 :
4126 : /* Adjust record to store to remaining buffer space. */
4127 0 : prb_rec_init_rd(&r, &info, buf + len, size - len);
4128 : }
4129 :
4130 0 : iter->next_seq = next_seq;
4131 0 : ret = true;
4132 : out:
4133 0 : if (len_out)
4134 0 : *len_out = len;
4135 0 : return ret;
4136 : }
4137 : EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4138 :
4139 : /**
4140 : * kmsg_dump_rewind - reset the iterator
4141 : * @iter: kmsg dump iterator
4142 : *
4143 : * Reset the dumper's iterator so that kmsg_dump_get_line() and
4144 : * kmsg_dump_get_buffer() can be called again and used multiple
4145 : * times within the same dumper.dump() callback.
4146 : */
4147 0 : void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4148 : {
4149 0 : iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4150 0 : iter->next_seq = prb_next_seq(prb);
4151 0 : }
4152 : EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4153 :
4154 : #endif
4155 :
4156 : #ifdef CONFIG_SMP
4157 : static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4158 : static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4159 :
4160 : /**
4161 : * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4162 : * spinning lock is not owned by any CPU.
4163 : *
4164 : * Context: Any context.
4165 : */
4166 : void __printk_cpu_sync_wait(void)
4167 : {
4168 : do {
4169 : cpu_relax();
4170 : } while (atomic_read(&printk_cpu_sync_owner) != -1);
4171 : }
4172 : EXPORT_SYMBOL(__printk_cpu_sync_wait);
4173 :
4174 : /**
4175 : * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4176 : * spinning lock.
4177 : *
4178 : * If no processor has the lock, the calling processor takes the lock and
4179 : * becomes the owner. If the calling processor is already the owner of the
4180 : * lock, this function succeeds immediately.
4181 : *
4182 : * Context: Any context. Expects interrupts to be disabled.
4183 : * Return: 1 on success, otherwise 0.
4184 : */
4185 : int __printk_cpu_sync_try_get(void)
4186 : {
4187 : int cpu;
4188 : int old;
4189 :
4190 : cpu = smp_processor_id();
4191 :
4192 : /*
4193 : * Guarantee loads and stores from this CPU when it is the lock owner
4194 : * are _not_ visible to the previous lock owner. This pairs with
4195 : * __printk_cpu_sync_put:B.
4196 : *
4197 : * Memory barrier involvement:
4198 : *
4199 : * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4200 : * then __printk_cpu_sync_put:A can never read from
4201 : * __printk_cpu_sync_try_get:B.
4202 : *
4203 : * Relies on:
4204 : *
4205 : * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4206 : * of the previous CPU
4207 : * matching
4208 : * ACQUIRE from __printk_cpu_sync_try_get:A to
4209 : * __printk_cpu_sync_try_get:B of this CPU
4210 : */
4211 : old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4212 : cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4213 : if (old == -1) {
4214 : /*
4215 : * This CPU is now the owner and begins loading/storing
4216 : * data: LMM(__printk_cpu_sync_try_get:B)
4217 : */
4218 : return 1;
4219 :
4220 : } else if (old == cpu) {
4221 : /* This CPU is already the owner. */
4222 : atomic_inc(&printk_cpu_sync_nested);
4223 : return 1;
4224 : }
4225 :
4226 : return 0;
4227 : }
4228 : EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4229 :
4230 : /**
4231 : * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4232 : *
4233 : * The calling processor must be the owner of the lock.
4234 : *
4235 : * Context: Any context. Expects interrupts to be disabled.
4236 : */
4237 : void __printk_cpu_sync_put(void)
4238 : {
4239 : if (atomic_read(&printk_cpu_sync_nested)) {
4240 : atomic_dec(&printk_cpu_sync_nested);
4241 : return;
4242 : }
4243 :
4244 : /*
4245 : * This CPU is finished loading/storing data:
4246 : * LMM(__printk_cpu_sync_put:A)
4247 : */
4248 :
4249 : /*
4250 : * Guarantee loads and stores from this CPU when it was the
4251 : * lock owner are visible to the next lock owner. This pairs
4252 : * with __printk_cpu_sync_try_get:A.
4253 : *
4254 : * Memory barrier involvement:
4255 : *
4256 : * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4257 : * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4258 : *
4259 : * Relies on:
4260 : *
4261 : * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4262 : * of this CPU
4263 : * matching
4264 : * ACQUIRE from __printk_cpu_sync_try_get:A to
4265 : * __printk_cpu_sync_try_get:B of the next CPU
4266 : */
4267 : atomic_set_release(&printk_cpu_sync_owner,
4268 : -1); /* LMM(__printk_cpu_sync_put:B) */
4269 : }
4270 : EXPORT_SYMBOL(__printk_cpu_sync_put);
4271 : #endif /* CONFIG_SMP */
|
