Loading...
1/*
2 * linux/kernel/printk.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
11 * Ted Ts'o, 2/11/93.
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
17 */
18
19#include <linux/kernel.h>
20#include <linux/mm.h>
21#include <linux/tty.h>
22#include <linux/tty_driver.h>
23#include <linux/console.h>
24#include <linux/init.h>
25#include <linux/jiffies.h>
26#include <linux/nmi.h>
27#include <linux/module.h>
28#include <linux/moduleparam.h>
29#include <linux/delay.h>
30#include <linux/smp.h>
31#include <linux/security.h>
32#include <linux/bootmem.h>
33#include <linux/memblock.h>
34#include <linux/syscalls.h>
35#include <linux/kexec.h>
36#include <linux/kdb.h>
37#include <linux/ratelimit.h>
38#include <linux/kmsg_dump.h>
39#include <linux/syslog.h>
40#include <linux/cpu.h>
41#include <linux/notifier.h>
42#include <linux/rculist.h>
43#include <linux/poll.h>
44#include <linux/irq_work.h>
45#include <linux/utsname.h>
46#include <linux/ctype.h>
47#include <linux/uio.h>
48
49#include <linux/uaccess.h>
50#include <asm/sections.h>
51
52#define CREATE_TRACE_POINTS
53#include <trace/events/printk.h>
54
55#include "console_cmdline.h"
56#include "braille.h"
57#include "internal.h"
58
59int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
64};
65
66/*
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
69 */
70int oops_in_progress;
71EXPORT_SYMBOL(oops_in_progress);
72
73/*
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
76 * driver system.
77 */
78static DEFINE_SEMAPHORE(console_sem);
79struct console *console_drivers;
80EXPORT_SYMBOL_GPL(console_drivers);
81
82#ifdef CONFIG_LOCKDEP
83static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
85};
86#endif
87
88enum devkmsg_log_bits {
89 __DEVKMSG_LOG_BIT_ON = 0,
90 __DEVKMSG_LOG_BIT_OFF,
91 __DEVKMSG_LOG_BIT_LOCK,
92};
93
94enum devkmsg_log_masks {
95 DEVKMSG_LOG_MASK_ON = BIT(__DEVKMSG_LOG_BIT_ON),
96 DEVKMSG_LOG_MASK_OFF = BIT(__DEVKMSG_LOG_BIT_OFF),
97 DEVKMSG_LOG_MASK_LOCK = BIT(__DEVKMSG_LOG_BIT_LOCK),
98};
99
100/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
101#define DEVKMSG_LOG_MASK_DEFAULT 0
102
103static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
104
105static int __control_devkmsg(char *str)
106{
107 if (!str)
108 return -EINVAL;
109
110 if (!strncmp(str, "on", 2)) {
111 devkmsg_log = DEVKMSG_LOG_MASK_ON;
112 return 2;
113 } else if (!strncmp(str, "off", 3)) {
114 devkmsg_log = DEVKMSG_LOG_MASK_OFF;
115 return 3;
116 } else if (!strncmp(str, "ratelimit", 9)) {
117 devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
118 return 9;
119 }
120 return -EINVAL;
121}
122
123static int __init control_devkmsg(char *str)
124{
125 if (__control_devkmsg(str) < 0)
126 return 1;
127
128 /*
129 * Set sysctl string accordingly:
130 */
131 if (devkmsg_log == DEVKMSG_LOG_MASK_ON) {
132 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
133 strncpy(devkmsg_log_str, "on", 2);
134 } else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF) {
135 memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
136 strncpy(devkmsg_log_str, "off", 3);
137 }
138 /* else "ratelimit" which is set by default. */
139
140 /*
141 * Sysctl cannot change it anymore. The kernel command line setting of
142 * this parameter is to force the setting to be permanent throughout the
143 * runtime of the system. This is a precation measure against userspace
144 * trying to be a smarta** and attempting to change it up on us.
145 */
146 devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
147
148 return 0;
149}
150__setup("printk.devkmsg=", control_devkmsg);
151
152char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
153
154int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
155 void __user *buffer, size_t *lenp, loff_t *ppos)
156{
157 char old_str[DEVKMSG_STR_MAX_SIZE];
158 unsigned int old;
159 int err;
160
161 if (write) {
162 if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
163 return -EINVAL;
164
165 old = devkmsg_log;
166 strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
167 }
168
169 err = proc_dostring(table, write, buffer, lenp, ppos);
170 if (err)
171 return err;
172
173 if (write) {
174 err = __control_devkmsg(devkmsg_log_str);
175
176 /*
177 * Do not accept an unknown string OR a known string with
178 * trailing crap...
179 */
180 if (err < 0 || (err + 1 != *lenp)) {
181
182 /* ... and restore old setting. */
183 devkmsg_log = old;
184 strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
185
186 return -EINVAL;
187 }
188 }
189
190 return 0;
191}
192
193/*
194 * Number of registered extended console drivers.
195 *
196 * If extended consoles are present, in-kernel cont reassembly is disabled
197 * and each fragment is stored as a separate log entry with proper
198 * continuation flag so that every emitted message has full metadata. This
199 * doesn't change the result for regular consoles or /proc/kmsg. For
200 * /dev/kmsg, as long as the reader concatenates messages according to
201 * consecutive continuation flags, the end result should be the same too.
202 */
203static int nr_ext_console_drivers;
204
205/*
206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
207 * macros instead of functions so that _RET_IP_ contains useful information.
208 */
209#define down_console_sem() do { \
210 down(&console_sem);\
211 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
212} while (0)
213
214static int __down_trylock_console_sem(unsigned long ip)
215{
216 if (down_trylock(&console_sem))
217 return 1;
218 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
219 return 0;
220}
221#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
222
223#define up_console_sem() do { \
224 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
225 up(&console_sem);\
226} while (0)
227
228/*
229 * This is used for debugging the mess that is the VT code by
230 * keeping track if we have the console semaphore held. It's
231 * definitely not the perfect debug tool (we don't know if _WE_
232 * hold it and are racing, but it helps tracking those weird code
233 * paths in the console code where we end up in places I want
234 * locked without the console sempahore held).
235 */
236static int console_locked, console_suspended;
237
238/*
239 * If exclusive_console is non-NULL then only this console is to be printed to.
240 */
241static struct console *exclusive_console;
242
243/*
244 * Array of consoles built from command line options (console=)
245 */
246
247#define MAX_CMDLINECONSOLES 8
248
249static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
250
251static int selected_console = -1;
252static int preferred_console = -1;
253int console_set_on_cmdline;
254EXPORT_SYMBOL(console_set_on_cmdline);
255
256/* Flag: console code may call schedule() */
257static int console_may_schedule;
258
259/*
260 * The printk log buffer consists of a chain of concatenated variable
261 * length records. Every record starts with a record header, containing
262 * the overall length of the record.
263 *
264 * The heads to the first and last entry in the buffer, as well as the
265 * sequence numbers of these entries are maintained when messages are
266 * stored.
267 *
268 * If the heads indicate available messages, the length in the header
269 * tells the start next message. A length == 0 for the next message
270 * indicates a wrap-around to the beginning of the buffer.
271 *
272 * Every record carries the monotonic timestamp in microseconds, as well as
273 * the standard userspace syslog level and syslog facility. The usual
274 * kernel messages use LOG_KERN; userspace-injected messages always carry
275 * a matching syslog facility, by default LOG_USER. The origin of every
276 * message can be reliably determined that way.
277 *
278 * The human readable log message directly follows the message header. The
279 * length of the message text is stored in the header, the stored message
280 * is not terminated.
281 *
282 * Optionally, a message can carry a dictionary of properties (key/value pairs),
283 * to provide userspace with a machine-readable message context.
284 *
285 * Examples for well-defined, commonly used property names are:
286 * DEVICE=b12:8 device identifier
287 * b12:8 block dev_t
288 * c127:3 char dev_t
289 * n8 netdev ifindex
290 * +sound:card0 subsystem:devname
291 * SUBSYSTEM=pci driver-core subsystem name
292 *
293 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
294 * follows directly after a '=' character. Every property is terminated by
295 * a '\0' character. The last property is not terminated.
296 *
297 * Example of a message structure:
298 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
299 * 0008 34 00 record is 52 bytes long
300 * 000a 0b 00 text is 11 bytes long
301 * 000c 1f 00 dictionary is 23 bytes long
302 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
303 * 0010 69 74 27 73 20 61 20 6c "it's a l"
304 * 69 6e 65 "ine"
305 * 001b 44 45 56 49 43 "DEVIC"
306 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
307 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
308 * 67 "g"
309 * 0032 00 00 00 padding to next message header
310 *
311 * The 'struct printk_log' buffer header must never be directly exported to
312 * userspace, it is a kernel-private implementation detail that might
313 * need to be changed in the future, when the requirements change.
314 *
315 * /dev/kmsg exports the structured data in the following line format:
316 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
317 *
318 * Users of the export format should ignore possible additional values
319 * separated by ',', and find the message after the ';' character.
320 *
321 * The optional key/value pairs are attached as continuation lines starting
322 * with a space character and terminated by a newline. All possible
323 * non-prinatable characters are escaped in the "\xff" notation.
324 */
325
326enum log_flags {
327 LOG_NOCONS = 1, /* already flushed, do not print to console */
328 LOG_NEWLINE = 2, /* text ended with a newline */
329 LOG_PREFIX = 4, /* text started with a prefix */
330 LOG_CONT = 8, /* text is a fragment of a continuation line */
331};
332
333struct printk_log {
334 u64 ts_nsec; /* timestamp in nanoseconds */
335 u16 len; /* length of entire record */
336 u16 text_len; /* length of text buffer */
337 u16 dict_len; /* length of dictionary buffer */
338 u8 facility; /* syslog facility */
339 u8 flags:5; /* internal record flags */
340 u8 level:3; /* syslog level */
341}
342#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
343__packed __aligned(4)
344#endif
345;
346
347/*
348 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
349 * within the scheduler's rq lock. It must be released before calling
350 * console_unlock() or anything else that might wake up a process.
351 */
352DEFINE_RAW_SPINLOCK(logbuf_lock);
353
354#ifdef CONFIG_PRINTK
355DECLARE_WAIT_QUEUE_HEAD(log_wait);
356/* the next printk record to read by syslog(READ) or /proc/kmsg */
357static u64 syslog_seq;
358static u32 syslog_idx;
359static size_t syslog_partial;
360
361/* index and sequence number of the first record stored in the buffer */
362static u64 log_first_seq;
363static u32 log_first_idx;
364
365/* index and sequence number of the next record to store in the buffer */
366static u64 log_next_seq;
367static u32 log_next_idx;
368
369/* the next printk record to write to the console */
370static u64 console_seq;
371static u32 console_idx;
372
373/* the next printk record to read after the last 'clear' command */
374static u64 clear_seq;
375static u32 clear_idx;
376
377#define PREFIX_MAX 32
378#define LOG_LINE_MAX (1024 - PREFIX_MAX)
379
380#define LOG_LEVEL(v) ((v) & 0x07)
381#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
382
383/* record buffer */
384#define LOG_ALIGN __alignof__(struct printk_log)
385#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
386static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
387static char *log_buf = __log_buf;
388static u32 log_buf_len = __LOG_BUF_LEN;
389
390/* Return log buffer address */
391char *log_buf_addr_get(void)
392{
393 return log_buf;
394}
395
396/* Return log buffer size */
397u32 log_buf_len_get(void)
398{
399 return log_buf_len;
400}
401
402/* human readable text of the record */
403static char *log_text(const struct printk_log *msg)
404{
405 return (char *)msg + sizeof(struct printk_log);
406}
407
408/* optional key/value pair dictionary attached to the record */
409static char *log_dict(const struct printk_log *msg)
410{
411 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
412}
413
414/* get record by index; idx must point to valid msg */
415static struct printk_log *log_from_idx(u32 idx)
416{
417 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
418
419 /*
420 * A length == 0 record is the end of buffer marker. Wrap around and
421 * read the message at the start of the buffer.
422 */
423 if (!msg->len)
424 return (struct printk_log *)log_buf;
425 return msg;
426}
427
428/* get next record; idx must point to valid msg */
429static u32 log_next(u32 idx)
430{
431 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
432
433 /* length == 0 indicates the end of the buffer; wrap */
434 /*
435 * A length == 0 record is the end of buffer marker. Wrap around and
436 * read the message at the start of the buffer as *this* one, and
437 * return the one after that.
438 */
439 if (!msg->len) {
440 msg = (struct printk_log *)log_buf;
441 return msg->len;
442 }
443 return idx + msg->len;
444}
445
446/*
447 * Check whether there is enough free space for the given message.
448 *
449 * The same values of first_idx and next_idx mean that the buffer
450 * is either empty or full.
451 *
452 * If the buffer is empty, we must respect the position of the indexes.
453 * They cannot be reset to the beginning of the buffer.
454 */
455static int logbuf_has_space(u32 msg_size, bool empty)
456{
457 u32 free;
458
459 if (log_next_idx > log_first_idx || empty)
460 free = max(log_buf_len - log_next_idx, log_first_idx);
461 else
462 free = log_first_idx - log_next_idx;
463
464 /*
465 * We need space also for an empty header that signalizes wrapping
466 * of the buffer.
467 */
468 return free >= msg_size + sizeof(struct printk_log);
469}
470
471static int log_make_free_space(u32 msg_size)
472{
473 while (log_first_seq < log_next_seq &&
474 !logbuf_has_space(msg_size, false)) {
475 /* drop old messages until we have enough contiguous space */
476 log_first_idx = log_next(log_first_idx);
477 log_first_seq++;
478 }
479
480 if (clear_seq < log_first_seq) {
481 clear_seq = log_first_seq;
482 clear_idx = log_first_idx;
483 }
484
485 /* sequence numbers are equal, so the log buffer is empty */
486 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
487 return 0;
488
489 return -ENOMEM;
490}
491
492/* compute the message size including the padding bytes */
493static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
494{
495 u32 size;
496
497 size = sizeof(struct printk_log) + text_len + dict_len;
498 *pad_len = (-size) & (LOG_ALIGN - 1);
499 size += *pad_len;
500
501 return size;
502}
503
504/*
505 * Define how much of the log buffer we could take at maximum. The value
506 * must be greater than two. Note that only half of the buffer is available
507 * when the index points to the middle.
508 */
509#define MAX_LOG_TAKE_PART 4
510static const char trunc_msg[] = "<truncated>";
511
512static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
513 u16 *dict_len, u32 *pad_len)
514{
515 /*
516 * The message should not take the whole buffer. Otherwise, it might
517 * get removed too soon.
518 */
519 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
520 if (*text_len > max_text_len)
521 *text_len = max_text_len;
522 /* enable the warning message */
523 *trunc_msg_len = strlen(trunc_msg);
524 /* disable the "dict" completely */
525 *dict_len = 0;
526 /* compute the size again, count also the warning message */
527 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
528}
529
530/* insert record into the buffer, discard old ones, update heads */
531static int log_store(int facility, int level,
532 enum log_flags flags, u64 ts_nsec,
533 const char *dict, u16 dict_len,
534 const char *text, u16 text_len)
535{
536 struct printk_log *msg;
537 u32 size, pad_len;
538 u16 trunc_msg_len = 0;
539
540 /* number of '\0' padding bytes to next message */
541 size = msg_used_size(text_len, dict_len, &pad_len);
542
543 if (log_make_free_space(size)) {
544 /* truncate the message if it is too long for empty buffer */
545 size = truncate_msg(&text_len, &trunc_msg_len,
546 &dict_len, &pad_len);
547 /* survive when the log buffer is too small for trunc_msg */
548 if (log_make_free_space(size))
549 return 0;
550 }
551
552 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
553 /*
554 * This message + an additional empty header does not fit
555 * at the end of the buffer. Add an empty header with len == 0
556 * to signify a wrap around.
557 */
558 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
559 log_next_idx = 0;
560 }
561
562 /* fill message */
563 msg = (struct printk_log *)(log_buf + log_next_idx);
564 memcpy(log_text(msg), text, text_len);
565 msg->text_len = text_len;
566 if (trunc_msg_len) {
567 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
568 msg->text_len += trunc_msg_len;
569 }
570 memcpy(log_dict(msg), dict, dict_len);
571 msg->dict_len = dict_len;
572 msg->facility = facility;
573 msg->level = level & 7;
574 msg->flags = flags & 0x1f;
575 if (ts_nsec > 0)
576 msg->ts_nsec = ts_nsec;
577 else
578 msg->ts_nsec = local_clock();
579 memset(log_dict(msg) + dict_len, 0, pad_len);
580 msg->len = size;
581
582 /* insert message */
583 log_next_idx += msg->len;
584 log_next_seq++;
585
586 return msg->text_len;
587}
588
589int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
590
591static int syslog_action_restricted(int type)
592{
593 if (dmesg_restrict)
594 return 1;
595 /*
596 * Unless restricted, we allow "read all" and "get buffer size"
597 * for everybody.
598 */
599 return type != SYSLOG_ACTION_READ_ALL &&
600 type != SYSLOG_ACTION_SIZE_BUFFER;
601}
602
603int check_syslog_permissions(int type, int source)
604{
605 /*
606 * If this is from /proc/kmsg and we've already opened it, then we've
607 * already done the capabilities checks at open time.
608 */
609 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
610 goto ok;
611
612 if (syslog_action_restricted(type)) {
613 if (capable(CAP_SYSLOG))
614 goto ok;
615 /*
616 * For historical reasons, accept CAP_SYS_ADMIN too, with
617 * a warning.
618 */
619 if (capable(CAP_SYS_ADMIN)) {
620 pr_warn_once("%s (%d): Attempt to access syslog with "
621 "CAP_SYS_ADMIN but no CAP_SYSLOG "
622 "(deprecated).\n",
623 current->comm, task_pid_nr(current));
624 goto ok;
625 }
626 return -EPERM;
627 }
628ok:
629 return security_syslog(type);
630}
631EXPORT_SYMBOL_GPL(check_syslog_permissions);
632
633static void append_char(char **pp, char *e, char c)
634{
635 if (*pp < e)
636 *(*pp)++ = c;
637}
638
639static ssize_t msg_print_ext_header(char *buf, size_t size,
640 struct printk_log *msg, u64 seq)
641{
642 u64 ts_usec = msg->ts_nsec;
643
644 do_div(ts_usec, 1000);
645
646 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
647 (msg->facility << 3) | msg->level, seq, ts_usec,
648 msg->flags & LOG_CONT ? 'c' : '-');
649}
650
651static ssize_t msg_print_ext_body(char *buf, size_t size,
652 char *dict, size_t dict_len,
653 char *text, size_t text_len)
654{
655 char *p = buf, *e = buf + size;
656 size_t i;
657
658 /* escape non-printable characters */
659 for (i = 0; i < text_len; i++) {
660 unsigned char c = text[i];
661
662 if (c < ' ' || c >= 127 || c == '\\')
663 p += scnprintf(p, e - p, "\\x%02x", c);
664 else
665 append_char(&p, e, c);
666 }
667 append_char(&p, e, '\n');
668
669 if (dict_len) {
670 bool line = true;
671
672 for (i = 0; i < dict_len; i++) {
673 unsigned char c = dict[i];
674
675 if (line) {
676 append_char(&p, e, ' ');
677 line = false;
678 }
679
680 if (c == '\0') {
681 append_char(&p, e, '\n');
682 line = true;
683 continue;
684 }
685
686 if (c < ' ' || c >= 127 || c == '\\') {
687 p += scnprintf(p, e - p, "\\x%02x", c);
688 continue;
689 }
690
691 append_char(&p, e, c);
692 }
693 append_char(&p, e, '\n');
694 }
695
696 return p - buf;
697}
698
699/* /dev/kmsg - userspace message inject/listen interface */
700struct devkmsg_user {
701 u64 seq;
702 u32 idx;
703 struct ratelimit_state rs;
704 struct mutex lock;
705 char buf[CONSOLE_EXT_LOG_MAX];
706};
707
708static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
709{
710 char *buf, *line;
711 int level = default_message_loglevel;
712 int facility = 1; /* LOG_USER */
713 struct file *file = iocb->ki_filp;
714 struct devkmsg_user *user = file->private_data;
715 size_t len = iov_iter_count(from);
716 ssize_t ret = len;
717
718 if (!user || len > LOG_LINE_MAX)
719 return -EINVAL;
720
721 /* Ignore when user logging is disabled. */
722 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
723 return len;
724
725 /* Ratelimit when not explicitly enabled. */
726 if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
727 if (!___ratelimit(&user->rs, current->comm))
728 return ret;
729 }
730
731 buf = kmalloc(len+1, GFP_KERNEL);
732 if (buf == NULL)
733 return -ENOMEM;
734
735 buf[len] = '\0';
736 if (!copy_from_iter_full(buf, len, from)) {
737 kfree(buf);
738 return -EFAULT;
739 }
740
741 /*
742 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
743 * the decimal value represents 32bit, the lower 3 bit are the log
744 * level, the rest are the log facility.
745 *
746 * If no prefix or no userspace facility is specified, we
747 * enforce LOG_USER, to be able to reliably distinguish
748 * kernel-generated messages from userspace-injected ones.
749 */
750 line = buf;
751 if (line[0] == '<') {
752 char *endp = NULL;
753 unsigned int u;
754
755 u = simple_strtoul(line + 1, &endp, 10);
756 if (endp && endp[0] == '>') {
757 level = LOG_LEVEL(u);
758 if (LOG_FACILITY(u) != 0)
759 facility = LOG_FACILITY(u);
760 endp++;
761 len -= endp - line;
762 line = endp;
763 }
764 }
765
766 printk_emit(facility, level, NULL, 0, "%s", line);
767 kfree(buf);
768 return ret;
769}
770
771static ssize_t devkmsg_read(struct file *file, char __user *buf,
772 size_t count, loff_t *ppos)
773{
774 struct devkmsg_user *user = file->private_data;
775 struct printk_log *msg;
776 size_t len;
777 ssize_t ret;
778
779 if (!user)
780 return -EBADF;
781
782 ret = mutex_lock_interruptible(&user->lock);
783 if (ret)
784 return ret;
785 raw_spin_lock_irq(&logbuf_lock);
786 while (user->seq == log_next_seq) {
787 if (file->f_flags & O_NONBLOCK) {
788 ret = -EAGAIN;
789 raw_spin_unlock_irq(&logbuf_lock);
790 goto out;
791 }
792
793 raw_spin_unlock_irq(&logbuf_lock);
794 ret = wait_event_interruptible(log_wait,
795 user->seq != log_next_seq);
796 if (ret)
797 goto out;
798 raw_spin_lock_irq(&logbuf_lock);
799 }
800
801 if (user->seq < log_first_seq) {
802 /* our last seen message is gone, return error and reset */
803 user->idx = log_first_idx;
804 user->seq = log_first_seq;
805 ret = -EPIPE;
806 raw_spin_unlock_irq(&logbuf_lock);
807 goto out;
808 }
809
810 msg = log_from_idx(user->idx);
811 len = msg_print_ext_header(user->buf, sizeof(user->buf),
812 msg, user->seq);
813 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
814 log_dict(msg), msg->dict_len,
815 log_text(msg), msg->text_len);
816
817 user->idx = log_next(user->idx);
818 user->seq++;
819 raw_spin_unlock_irq(&logbuf_lock);
820
821 if (len > count) {
822 ret = -EINVAL;
823 goto out;
824 }
825
826 if (copy_to_user(buf, user->buf, len)) {
827 ret = -EFAULT;
828 goto out;
829 }
830 ret = len;
831out:
832 mutex_unlock(&user->lock);
833 return ret;
834}
835
836static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
837{
838 struct devkmsg_user *user = file->private_data;
839 loff_t ret = 0;
840
841 if (!user)
842 return -EBADF;
843 if (offset)
844 return -ESPIPE;
845
846 raw_spin_lock_irq(&logbuf_lock);
847 switch (whence) {
848 case SEEK_SET:
849 /* the first record */
850 user->idx = log_first_idx;
851 user->seq = log_first_seq;
852 break;
853 case SEEK_DATA:
854 /*
855 * The first record after the last SYSLOG_ACTION_CLEAR,
856 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
857 * changes no global state, and does not clear anything.
858 */
859 user->idx = clear_idx;
860 user->seq = clear_seq;
861 break;
862 case SEEK_END:
863 /* after the last record */
864 user->idx = log_next_idx;
865 user->seq = log_next_seq;
866 break;
867 default:
868 ret = -EINVAL;
869 }
870 raw_spin_unlock_irq(&logbuf_lock);
871 return ret;
872}
873
874static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
875{
876 struct devkmsg_user *user = file->private_data;
877 int ret = 0;
878
879 if (!user)
880 return POLLERR|POLLNVAL;
881
882 poll_wait(file, &log_wait, wait);
883
884 raw_spin_lock_irq(&logbuf_lock);
885 if (user->seq < log_next_seq) {
886 /* return error when data has vanished underneath us */
887 if (user->seq < log_first_seq)
888 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
889 else
890 ret = POLLIN|POLLRDNORM;
891 }
892 raw_spin_unlock_irq(&logbuf_lock);
893
894 return ret;
895}
896
897static int devkmsg_open(struct inode *inode, struct file *file)
898{
899 struct devkmsg_user *user;
900 int err;
901
902 if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
903 return -EPERM;
904
905 /* write-only does not need any file context */
906 if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
907 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
908 SYSLOG_FROM_READER);
909 if (err)
910 return err;
911 }
912
913 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
914 if (!user)
915 return -ENOMEM;
916
917 ratelimit_default_init(&user->rs);
918 ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
919
920 mutex_init(&user->lock);
921
922 raw_spin_lock_irq(&logbuf_lock);
923 user->idx = log_first_idx;
924 user->seq = log_first_seq;
925 raw_spin_unlock_irq(&logbuf_lock);
926
927 file->private_data = user;
928 return 0;
929}
930
931static int devkmsg_release(struct inode *inode, struct file *file)
932{
933 struct devkmsg_user *user = file->private_data;
934
935 if (!user)
936 return 0;
937
938 ratelimit_state_exit(&user->rs);
939
940 mutex_destroy(&user->lock);
941 kfree(user);
942 return 0;
943}
944
945const struct file_operations kmsg_fops = {
946 .open = devkmsg_open,
947 .read = devkmsg_read,
948 .write_iter = devkmsg_write,
949 .llseek = devkmsg_llseek,
950 .poll = devkmsg_poll,
951 .release = devkmsg_release,
952};
953
954#ifdef CONFIG_KEXEC_CORE
955/*
956 * This appends the listed symbols to /proc/vmcore
957 *
958 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
959 * obtain access to symbols that are otherwise very difficult to locate. These
960 * symbols are specifically used so that utilities can access and extract the
961 * dmesg log from a vmcore file after a crash.
962 */
963void log_buf_kexec_setup(void)
964{
965 VMCOREINFO_SYMBOL(log_buf);
966 VMCOREINFO_SYMBOL(log_buf_len);
967 VMCOREINFO_SYMBOL(log_first_idx);
968 VMCOREINFO_SYMBOL(clear_idx);
969 VMCOREINFO_SYMBOL(log_next_idx);
970 /*
971 * Export struct printk_log size and field offsets. User space tools can
972 * parse it and detect any changes to structure down the line.
973 */
974 VMCOREINFO_STRUCT_SIZE(printk_log);
975 VMCOREINFO_OFFSET(printk_log, ts_nsec);
976 VMCOREINFO_OFFSET(printk_log, len);
977 VMCOREINFO_OFFSET(printk_log, text_len);
978 VMCOREINFO_OFFSET(printk_log, dict_len);
979}
980#endif
981
982/* requested log_buf_len from kernel cmdline */
983static unsigned long __initdata new_log_buf_len;
984
985/* we practice scaling the ring buffer by powers of 2 */
986static void __init log_buf_len_update(unsigned size)
987{
988 if (size)
989 size = roundup_pow_of_two(size);
990 if (size > log_buf_len)
991 new_log_buf_len = size;
992}
993
994/* save requested log_buf_len since it's too early to process it */
995static int __init log_buf_len_setup(char *str)
996{
997 unsigned size = memparse(str, &str);
998
999 log_buf_len_update(size);
1000
1001 return 0;
1002}
1003early_param("log_buf_len", log_buf_len_setup);
1004
1005#ifdef CONFIG_SMP
1006#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1007
1008static void __init log_buf_add_cpu(void)
1009{
1010 unsigned int cpu_extra;
1011
1012 /*
1013 * archs should set up cpu_possible_bits properly with
1014 * set_cpu_possible() after setup_arch() but just in
1015 * case lets ensure this is valid.
1016 */
1017 if (num_possible_cpus() == 1)
1018 return;
1019
1020 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1021
1022 /* by default this will only continue through for large > 64 CPUs */
1023 if (cpu_extra <= __LOG_BUF_LEN / 2)
1024 return;
1025
1026 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1027 __LOG_CPU_MAX_BUF_LEN);
1028 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1029 cpu_extra);
1030 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1031
1032 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1033}
1034#else /* !CONFIG_SMP */
1035static inline void log_buf_add_cpu(void) {}
1036#endif /* CONFIG_SMP */
1037
1038void __init setup_log_buf(int early)
1039{
1040 unsigned long flags;
1041 char *new_log_buf;
1042 int free;
1043
1044 if (log_buf != __log_buf)
1045 return;
1046
1047 if (!early && !new_log_buf_len)
1048 log_buf_add_cpu();
1049
1050 if (!new_log_buf_len)
1051 return;
1052
1053 if (early) {
1054 new_log_buf =
1055 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1056 } else {
1057 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1058 LOG_ALIGN);
1059 }
1060
1061 if (unlikely(!new_log_buf)) {
1062 pr_err("log_buf_len: %ld bytes not available\n",
1063 new_log_buf_len);
1064 return;
1065 }
1066
1067 raw_spin_lock_irqsave(&logbuf_lock, flags);
1068 log_buf_len = new_log_buf_len;
1069 log_buf = new_log_buf;
1070 new_log_buf_len = 0;
1071 free = __LOG_BUF_LEN - log_next_idx;
1072 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1073 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1074
1075 pr_info("log_buf_len: %d bytes\n", log_buf_len);
1076 pr_info("early log buf free: %d(%d%%)\n",
1077 free, (free * 100) / __LOG_BUF_LEN);
1078}
1079
1080static bool __read_mostly ignore_loglevel;
1081
1082static int __init ignore_loglevel_setup(char *str)
1083{
1084 ignore_loglevel = true;
1085 pr_info("debug: ignoring loglevel setting.\n");
1086
1087 return 0;
1088}
1089
1090early_param("ignore_loglevel", ignore_loglevel_setup);
1091module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1092MODULE_PARM_DESC(ignore_loglevel,
1093 "ignore loglevel setting (prints all kernel messages to the console)");
1094
1095static bool suppress_message_printing(int level)
1096{
1097 return (level >= console_loglevel && !ignore_loglevel);
1098}
1099
1100#ifdef CONFIG_BOOT_PRINTK_DELAY
1101
1102static int boot_delay; /* msecs delay after each printk during bootup */
1103static unsigned long long loops_per_msec; /* based on boot_delay */
1104
1105static int __init boot_delay_setup(char *str)
1106{
1107 unsigned long lpj;
1108
1109 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
1110 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1111
1112 get_option(&str, &boot_delay);
1113 if (boot_delay > 10 * 1000)
1114 boot_delay = 0;
1115
1116 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1117 "HZ: %d, loops_per_msec: %llu\n",
1118 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1119 return 0;
1120}
1121early_param("boot_delay", boot_delay_setup);
1122
1123static void boot_delay_msec(int level)
1124{
1125 unsigned long long k;
1126 unsigned long timeout;
1127
1128 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1129 || suppress_message_printing(level)) {
1130 return;
1131 }
1132
1133 k = (unsigned long long)loops_per_msec * boot_delay;
1134
1135 timeout = jiffies + msecs_to_jiffies(boot_delay);
1136 while (k) {
1137 k--;
1138 cpu_relax();
1139 /*
1140 * use (volatile) jiffies to prevent
1141 * compiler reduction; loop termination via jiffies
1142 * is secondary and may or may not happen.
1143 */
1144 if (time_after(jiffies, timeout))
1145 break;
1146 touch_nmi_watchdog();
1147 }
1148}
1149#else
1150static inline void boot_delay_msec(int level)
1151{
1152}
1153#endif
1154
1155static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1156module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1157
1158static size_t print_time(u64 ts, char *buf)
1159{
1160 unsigned long rem_nsec;
1161
1162 if (!printk_time)
1163 return 0;
1164
1165 rem_nsec = do_div(ts, 1000000000);
1166
1167 if (!buf)
1168 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1169
1170 return sprintf(buf, "[%5lu.%06lu] ",
1171 (unsigned long)ts, rem_nsec / 1000);
1172}
1173
1174static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1175{
1176 size_t len = 0;
1177 unsigned int prefix = (msg->facility << 3) | msg->level;
1178
1179 if (syslog) {
1180 if (buf) {
1181 len += sprintf(buf, "<%u>", prefix);
1182 } else {
1183 len += 3;
1184 if (prefix > 999)
1185 len += 3;
1186 else if (prefix > 99)
1187 len += 2;
1188 else if (prefix > 9)
1189 len++;
1190 }
1191 }
1192
1193 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1194 return len;
1195}
1196
1197static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1198{
1199 const char *text = log_text(msg);
1200 size_t text_size = msg->text_len;
1201 size_t len = 0;
1202
1203 do {
1204 const char *next = memchr(text, '\n', text_size);
1205 size_t text_len;
1206
1207 if (next) {
1208 text_len = next - text;
1209 next++;
1210 text_size -= next - text;
1211 } else {
1212 text_len = text_size;
1213 }
1214
1215 if (buf) {
1216 if (print_prefix(msg, syslog, NULL) +
1217 text_len + 1 >= size - len)
1218 break;
1219
1220 len += print_prefix(msg, syslog, buf + len);
1221 memcpy(buf + len, text, text_len);
1222 len += text_len;
1223 buf[len++] = '\n';
1224 } else {
1225 /* SYSLOG_ACTION_* buffer size only calculation */
1226 len += print_prefix(msg, syslog, NULL);
1227 len += text_len;
1228 len++;
1229 }
1230
1231 text = next;
1232 } while (text);
1233
1234 return len;
1235}
1236
1237static int syslog_print(char __user *buf, int size)
1238{
1239 char *text;
1240 struct printk_log *msg;
1241 int len = 0;
1242
1243 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1244 if (!text)
1245 return -ENOMEM;
1246
1247 while (size > 0) {
1248 size_t n;
1249 size_t skip;
1250
1251 raw_spin_lock_irq(&logbuf_lock);
1252 if (syslog_seq < log_first_seq) {
1253 /* messages are gone, move to first one */
1254 syslog_seq = log_first_seq;
1255 syslog_idx = log_first_idx;
1256 syslog_partial = 0;
1257 }
1258 if (syslog_seq == log_next_seq) {
1259 raw_spin_unlock_irq(&logbuf_lock);
1260 break;
1261 }
1262
1263 skip = syslog_partial;
1264 msg = log_from_idx(syslog_idx);
1265 n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1266 if (n - syslog_partial <= size) {
1267 /* message fits into buffer, move forward */
1268 syslog_idx = log_next(syslog_idx);
1269 syslog_seq++;
1270 n -= syslog_partial;
1271 syslog_partial = 0;
1272 } else if (!len){
1273 /* partial read(), remember position */
1274 n = size;
1275 syslog_partial += n;
1276 } else
1277 n = 0;
1278 raw_spin_unlock_irq(&logbuf_lock);
1279
1280 if (!n)
1281 break;
1282
1283 if (copy_to_user(buf, text + skip, n)) {
1284 if (!len)
1285 len = -EFAULT;
1286 break;
1287 }
1288
1289 len += n;
1290 size -= n;
1291 buf += n;
1292 }
1293
1294 kfree(text);
1295 return len;
1296}
1297
1298static int syslog_print_all(char __user *buf, int size, bool clear)
1299{
1300 char *text;
1301 int len = 0;
1302
1303 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1304 if (!text)
1305 return -ENOMEM;
1306
1307 raw_spin_lock_irq(&logbuf_lock);
1308 if (buf) {
1309 u64 next_seq;
1310 u64 seq;
1311 u32 idx;
1312
1313 /*
1314 * Find first record that fits, including all following records,
1315 * into the user-provided buffer for this dump.
1316 */
1317 seq = clear_seq;
1318 idx = clear_idx;
1319 while (seq < log_next_seq) {
1320 struct printk_log *msg = log_from_idx(idx);
1321
1322 len += msg_print_text(msg, true, NULL, 0);
1323 idx = log_next(idx);
1324 seq++;
1325 }
1326
1327 /* move first record forward until length fits into the buffer */
1328 seq = clear_seq;
1329 idx = clear_idx;
1330 while (len > size && seq < log_next_seq) {
1331 struct printk_log *msg = log_from_idx(idx);
1332
1333 len -= msg_print_text(msg, true, NULL, 0);
1334 idx = log_next(idx);
1335 seq++;
1336 }
1337
1338 /* last message fitting into this dump */
1339 next_seq = log_next_seq;
1340
1341 len = 0;
1342 while (len >= 0 && seq < next_seq) {
1343 struct printk_log *msg = log_from_idx(idx);
1344 int textlen;
1345
1346 textlen = msg_print_text(msg, true, text,
1347 LOG_LINE_MAX + PREFIX_MAX);
1348 if (textlen < 0) {
1349 len = textlen;
1350 break;
1351 }
1352 idx = log_next(idx);
1353 seq++;
1354
1355 raw_spin_unlock_irq(&logbuf_lock);
1356 if (copy_to_user(buf + len, text, textlen))
1357 len = -EFAULT;
1358 else
1359 len += textlen;
1360 raw_spin_lock_irq(&logbuf_lock);
1361
1362 if (seq < log_first_seq) {
1363 /* messages are gone, move to next one */
1364 seq = log_first_seq;
1365 idx = log_first_idx;
1366 }
1367 }
1368 }
1369
1370 if (clear) {
1371 clear_seq = log_next_seq;
1372 clear_idx = log_next_idx;
1373 }
1374 raw_spin_unlock_irq(&logbuf_lock);
1375
1376 kfree(text);
1377 return len;
1378}
1379
1380int do_syslog(int type, char __user *buf, int len, int source)
1381{
1382 bool clear = false;
1383 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1384 int error;
1385
1386 error = check_syslog_permissions(type, source);
1387 if (error)
1388 goto out;
1389
1390 switch (type) {
1391 case SYSLOG_ACTION_CLOSE: /* Close log */
1392 break;
1393 case SYSLOG_ACTION_OPEN: /* Open log */
1394 break;
1395 case SYSLOG_ACTION_READ: /* Read from log */
1396 error = -EINVAL;
1397 if (!buf || len < 0)
1398 goto out;
1399 error = 0;
1400 if (!len)
1401 goto out;
1402 if (!access_ok(VERIFY_WRITE, buf, len)) {
1403 error = -EFAULT;
1404 goto out;
1405 }
1406 error = wait_event_interruptible(log_wait,
1407 syslog_seq != log_next_seq);
1408 if (error)
1409 goto out;
1410 error = syslog_print(buf, len);
1411 break;
1412 /* Read/clear last kernel messages */
1413 case SYSLOG_ACTION_READ_CLEAR:
1414 clear = true;
1415 /* FALL THRU */
1416 /* Read last kernel messages */
1417 case SYSLOG_ACTION_READ_ALL:
1418 error = -EINVAL;
1419 if (!buf || len < 0)
1420 goto out;
1421 error = 0;
1422 if (!len)
1423 goto out;
1424 if (!access_ok(VERIFY_WRITE, buf, len)) {
1425 error = -EFAULT;
1426 goto out;
1427 }
1428 error = syslog_print_all(buf, len, clear);
1429 break;
1430 /* Clear ring buffer */
1431 case SYSLOG_ACTION_CLEAR:
1432 syslog_print_all(NULL, 0, true);
1433 break;
1434 /* Disable logging to console */
1435 case SYSLOG_ACTION_CONSOLE_OFF:
1436 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1437 saved_console_loglevel = console_loglevel;
1438 console_loglevel = minimum_console_loglevel;
1439 break;
1440 /* Enable logging to console */
1441 case SYSLOG_ACTION_CONSOLE_ON:
1442 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1443 console_loglevel = saved_console_loglevel;
1444 saved_console_loglevel = LOGLEVEL_DEFAULT;
1445 }
1446 break;
1447 /* Set level of messages printed to console */
1448 case SYSLOG_ACTION_CONSOLE_LEVEL:
1449 error = -EINVAL;
1450 if (len < 1 || len > 8)
1451 goto out;
1452 if (len < minimum_console_loglevel)
1453 len = minimum_console_loglevel;
1454 console_loglevel = len;
1455 /* Implicitly re-enable logging to console */
1456 saved_console_loglevel = LOGLEVEL_DEFAULT;
1457 error = 0;
1458 break;
1459 /* Number of chars in the log buffer */
1460 case SYSLOG_ACTION_SIZE_UNREAD:
1461 raw_spin_lock_irq(&logbuf_lock);
1462 if (syslog_seq < log_first_seq) {
1463 /* messages are gone, move to first one */
1464 syslog_seq = log_first_seq;
1465 syslog_idx = log_first_idx;
1466 syslog_partial = 0;
1467 }
1468 if (source == SYSLOG_FROM_PROC) {
1469 /*
1470 * Short-cut for poll(/"proc/kmsg") which simply checks
1471 * for pending data, not the size; return the count of
1472 * records, not the length.
1473 */
1474 error = log_next_seq - syslog_seq;
1475 } else {
1476 u64 seq = syslog_seq;
1477 u32 idx = syslog_idx;
1478
1479 error = 0;
1480 while (seq < log_next_seq) {
1481 struct printk_log *msg = log_from_idx(idx);
1482
1483 error += msg_print_text(msg, true, NULL, 0);
1484 idx = log_next(idx);
1485 seq++;
1486 }
1487 error -= syslog_partial;
1488 }
1489 raw_spin_unlock_irq(&logbuf_lock);
1490 break;
1491 /* Size of the log buffer */
1492 case SYSLOG_ACTION_SIZE_BUFFER:
1493 error = log_buf_len;
1494 break;
1495 default:
1496 error = -EINVAL;
1497 break;
1498 }
1499out:
1500 return error;
1501}
1502
1503SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1504{
1505 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1506}
1507
1508/*
1509 * Call the console drivers, asking them to write out
1510 * log_buf[start] to log_buf[end - 1].
1511 * The console_lock must be held.
1512 */
1513static void call_console_drivers(int level,
1514 const char *ext_text, size_t ext_len,
1515 const char *text, size_t len)
1516{
1517 struct console *con;
1518
1519 trace_console_rcuidle(text, len);
1520
1521 if (!console_drivers)
1522 return;
1523
1524 for_each_console(con) {
1525 if (exclusive_console && con != exclusive_console)
1526 continue;
1527 if (!(con->flags & CON_ENABLED))
1528 continue;
1529 if (!con->write)
1530 continue;
1531 if (!cpu_online(smp_processor_id()) &&
1532 !(con->flags & CON_ANYTIME))
1533 continue;
1534 if (con->flags & CON_EXTENDED)
1535 con->write(con, ext_text, ext_len);
1536 else
1537 con->write(con, text, len);
1538 }
1539}
1540
1541/*
1542 * Zap console related locks when oopsing.
1543 * To leave time for slow consoles to print a full oops,
1544 * only zap at most once every 30 seconds.
1545 */
1546static void zap_locks(void)
1547{
1548 static unsigned long oops_timestamp;
1549
1550 if (time_after_eq(jiffies, oops_timestamp) &&
1551 !time_after(jiffies, oops_timestamp + 30 * HZ))
1552 return;
1553
1554 oops_timestamp = jiffies;
1555
1556 debug_locks_off();
1557 /* If a crash is occurring, make sure we can't deadlock */
1558 raw_spin_lock_init(&logbuf_lock);
1559 /* And make sure that we print immediately */
1560 sema_init(&console_sem, 1);
1561}
1562
1563int printk_delay_msec __read_mostly;
1564
1565static inline void printk_delay(void)
1566{
1567 if (unlikely(printk_delay_msec)) {
1568 int m = printk_delay_msec;
1569
1570 while (m--) {
1571 mdelay(1);
1572 touch_nmi_watchdog();
1573 }
1574 }
1575}
1576
1577/*
1578 * Continuation lines are buffered, and not committed to the record buffer
1579 * until the line is complete, or a race forces it. The line fragments
1580 * though, are printed immediately to the consoles to ensure everything has
1581 * reached the console in case of a kernel crash.
1582 */
1583static struct cont {
1584 char buf[LOG_LINE_MAX];
1585 size_t len; /* length == 0 means unused buffer */
1586 struct task_struct *owner; /* task of first print*/
1587 u64 ts_nsec; /* time of first print */
1588 u8 level; /* log level of first message */
1589 u8 facility; /* log facility of first message */
1590 enum log_flags flags; /* prefix, newline flags */
1591} cont;
1592
1593static void cont_flush(void)
1594{
1595 if (cont.len == 0)
1596 return;
1597
1598 log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1599 NULL, 0, cont.buf, cont.len);
1600 cont.len = 0;
1601}
1602
1603static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1604{
1605 /*
1606 * If ext consoles are present, flush and skip in-kernel
1607 * continuation. See nr_ext_console_drivers definition. Also, if
1608 * the line gets too long, split it up in separate records.
1609 */
1610 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1611 cont_flush();
1612 return false;
1613 }
1614
1615 if (!cont.len) {
1616 cont.facility = facility;
1617 cont.level = level;
1618 cont.owner = current;
1619 cont.ts_nsec = local_clock();
1620 cont.flags = flags;
1621 }
1622
1623 memcpy(cont.buf + cont.len, text, len);
1624 cont.len += len;
1625
1626 // The original flags come from the first line,
1627 // but later continuations can add a newline.
1628 if (flags & LOG_NEWLINE) {
1629 cont.flags |= LOG_NEWLINE;
1630 cont_flush();
1631 }
1632
1633 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1634 cont_flush();
1635
1636 return true;
1637}
1638
1639static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1640{
1641 /*
1642 * If an earlier line was buffered, and we're a continuation
1643 * write from the same process, try to add it to the buffer.
1644 */
1645 if (cont.len) {
1646 if (cont.owner == current && (lflags & LOG_CONT)) {
1647 if (cont_add(facility, level, lflags, text, text_len))
1648 return text_len;
1649 }
1650 /* Otherwise, make sure it's flushed */
1651 cont_flush();
1652 }
1653
1654 /* Skip empty continuation lines that couldn't be added - they just flush */
1655 if (!text_len && (lflags & LOG_CONT))
1656 return 0;
1657
1658 /* If it doesn't end in a newline, try to buffer the current line */
1659 if (!(lflags & LOG_NEWLINE)) {
1660 if (cont_add(facility, level, lflags, text, text_len))
1661 return text_len;
1662 }
1663
1664 /* Store it in the record log */
1665 return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1666}
1667
1668asmlinkage int vprintk_emit(int facility, int level,
1669 const char *dict, size_t dictlen,
1670 const char *fmt, va_list args)
1671{
1672 static bool recursion_bug;
1673 static char textbuf[LOG_LINE_MAX];
1674 char *text = textbuf;
1675 size_t text_len = 0;
1676 enum log_flags lflags = 0;
1677 unsigned long flags;
1678 int this_cpu;
1679 int printed_len = 0;
1680 int nmi_message_lost;
1681 bool in_sched = false;
1682 /* cpu currently holding logbuf_lock in this function */
1683 static unsigned int logbuf_cpu = UINT_MAX;
1684
1685 if (level == LOGLEVEL_SCHED) {
1686 level = LOGLEVEL_DEFAULT;
1687 in_sched = true;
1688 }
1689
1690 boot_delay_msec(level);
1691 printk_delay();
1692
1693 local_irq_save(flags);
1694 this_cpu = smp_processor_id();
1695
1696 /*
1697 * Ouch, printk recursed into itself!
1698 */
1699 if (unlikely(logbuf_cpu == this_cpu)) {
1700 /*
1701 * If a crash is occurring during printk() on this CPU,
1702 * then try to get the crash message out but make sure
1703 * we can't deadlock. Otherwise just return to avoid the
1704 * recursion and return - but flag the recursion so that
1705 * it can be printed at the next appropriate moment:
1706 */
1707 if (!oops_in_progress && !lockdep_recursing(current)) {
1708 recursion_bug = true;
1709 local_irq_restore(flags);
1710 return 0;
1711 }
1712 zap_locks();
1713 }
1714
1715 lockdep_off();
1716 /* This stops the holder of console_sem just where we want him */
1717 raw_spin_lock(&logbuf_lock);
1718 logbuf_cpu = this_cpu;
1719
1720 if (unlikely(recursion_bug)) {
1721 static const char recursion_msg[] =
1722 "BUG: recent printk recursion!";
1723
1724 recursion_bug = false;
1725 /* emit KERN_CRIT message */
1726 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1727 NULL, 0, recursion_msg,
1728 strlen(recursion_msg));
1729 }
1730
1731 nmi_message_lost = get_nmi_message_lost();
1732 if (unlikely(nmi_message_lost)) {
1733 text_len = scnprintf(textbuf, sizeof(textbuf),
1734 "BAD LUCK: lost %d message(s) from NMI context!",
1735 nmi_message_lost);
1736 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1737 NULL, 0, textbuf, text_len);
1738 }
1739
1740 /*
1741 * The printf needs to come first; we need the syslog
1742 * prefix which might be passed-in as a parameter.
1743 */
1744 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1745
1746 /* mark and strip a trailing newline */
1747 if (text_len && text[text_len-1] == '\n') {
1748 text_len--;
1749 lflags |= LOG_NEWLINE;
1750 }
1751
1752 /* strip kernel syslog prefix and extract log level or control flags */
1753 if (facility == 0) {
1754 int kern_level;
1755
1756 while ((kern_level = printk_get_level(text)) != 0) {
1757 switch (kern_level) {
1758 case '0' ... '7':
1759 if (level == LOGLEVEL_DEFAULT)
1760 level = kern_level - '0';
1761 /* fallthrough */
1762 case 'd': /* KERN_DEFAULT */
1763 lflags |= LOG_PREFIX;
1764 break;
1765 case 'c': /* KERN_CONT */
1766 lflags |= LOG_CONT;
1767 }
1768
1769 text_len -= 2;
1770 text += 2;
1771 }
1772 }
1773
1774 if (level == LOGLEVEL_DEFAULT)
1775 level = default_message_loglevel;
1776
1777 if (dict)
1778 lflags |= LOG_PREFIX|LOG_NEWLINE;
1779
1780 printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1781
1782 logbuf_cpu = UINT_MAX;
1783 raw_spin_unlock(&logbuf_lock);
1784 lockdep_on();
1785 local_irq_restore(flags);
1786
1787 /* If called from the scheduler, we can not call up(). */
1788 if (!in_sched) {
1789 lockdep_off();
1790 /*
1791 * Try to acquire and then immediately release the console
1792 * semaphore. The release will print out buffers and wake up
1793 * /dev/kmsg and syslog() users.
1794 */
1795 if (console_trylock())
1796 console_unlock();
1797 lockdep_on();
1798 }
1799
1800 return printed_len;
1801}
1802EXPORT_SYMBOL(vprintk_emit);
1803
1804asmlinkage int vprintk(const char *fmt, va_list args)
1805{
1806 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1807}
1808EXPORT_SYMBOL(vprintk);
1809
1810asmlinkage int printk_emit(int facility, int level,
1811 const char *dict, size_t dictlen,
1812 const char *fmt, ...)
1813{
1814 va_list args;
1815 int r;
1816
1817 va_start(args, fmt);
1818 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1819 va_end(args);
1820
1821 return r;
1822}
1823EXPORT_SYMBOL(printk_emit);
1824
1825int vprintk_default(const char *fmt, va_list args)
1826{
1827 int r;
1828
1829#ifdef CONFIG_KGDB_KDB
1830 /* Allow to pass printk() to kdb but avoid a recursion. */
1831 if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1832 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1833 return r;
1834 }
1835#endif
1836 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1837
1838 return r;
1839}
1840EXPORT_SYMBOL_GPL(vprintk_default);
1841
1842/**
1843 * printk - print a kernel message
1844 * @fmt: format string
1845 *
1846 * This is printk(). It can be called from any context. We want it to work.
1847 *
1848 * We try to grab the console_lock. If we succeed, it's easy - we log the
1849 * output and call the console drivers. If we fail to get the semaphore, we
1850 * place the output into the log buffer and return. The current holder of
1851 * the console_sem will notice the new output in console_unlock(); and will
1852 * send it to the consoles before releasing the lock.
1853 *
1854 * One effect of this deferred printing is that code which calls printk() and
1855 * then changes console_loglevel may break. This is because console_loglevel
1856 * is inspected when the actual printing occurs.
1857 *
1858 * See also:
1859 * printf(3)
1860 *
1861 * See the vsnprintf() documentation for format string extensions over C99.
1862 */
1863asmlinkage __visible int printk(const char *fmt, ...)
1864{
1865 va_list args;
1866 int r;
1867
1868 va_start(args, fmt);
1869 r = vprintk_func(fmt, args);
1870 va_end(args);
1871
1872 return r;
1873}
1874EXPORT_SYMBOL(printk);
1875
1876#else /* CONFIG_PRINTK */
1877
1878#define LOG_LINE_MAX 0
1879#define PREFIX_MAX 0
1880
1881static u64 syslog_seq;
1882static u32 syslog_idx;
1883static u64 console_seq;
1884static u32 console_idx;
1885static u64 log_first_seq;
1886static u32 log_first_idx;
1887static u64 log_next_seq;
1888static char *log_text(const struct printk_log *msg) { return NULL; }
1889static char *log_dict(const struct printk_log *msg) { return NULL; }
1890static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1891static u32 log_next(u32 idx) { return 0; }
1892static ssize_t msg_print_ext_header(char *buf, size_t size,
1893 struct printk_log *msg,
1894 u64 seq) { return 0; }
1895static ssize_t msg_print_ext_body(char *buf, size_t size,
1896 char *dict, size_t dict_len,
1897 char *text, size_t text_len) { return 0; }
1898static void call_console_drivers(int level,
1899 const char *ext_text, size_t ext_len,
1900 const char *text, size_t len) {}
1901static size_t msg_print_text(const struct printk_log *msg,
1902 bool syslog, char *buf, size_t size) { return 0; }
1903static bool suppress_message_printing(int level) { return false; }
1904
1905/* Still needs to be defined for users */
1906DEFINE_PER_CPU(printk_func_t, printk_func);
1907
1908#endif /* CONFIG_PRINTK */
1909
1910#ifdef CONFIG_EARLY_PRINTK
1911struct console *early_console;
1912
1913asmlinkage __visible void early_printk(const char *fmt, ...)
1914{
1915 va_list ap;
1916 char buf[512];
1917 int n;
1918
1919 if (!early_console)
1920 return;
1921
1922 va_start(ap, fmt);
1923 n = vscnprintf(buf, sizeof(buf), fmt, ap);
1924 va_end(ap);
1925
1926 early_console->write(early_console, buf, n);
1927}
1928#endif
1929
1930static int __add_preferred_console(char *name, int idx, char *options,
1931 char *brl_options)
1932{
1933 struct console_cmdline *c;
1934 int i;
1935
1936 /*
1937 * See if this tty is not yet registered, and
1938 * if we have a slot free.
1939 */
1940 for (i = 0, c = console_cmdline;
1941 i < MAX_CMDLINECONSOLES && c->name[0];
1942 i++, c++) {
1943 if (strcmp(c->name, name) == 0 && c->index == idx) {
1944 if (!brl_options)
1945 selected_console = i;
1946 return 0;
1947 }
1948 }
1949 if (i == MAX_CMDLINECONSOLES)
1950 return -E2BIG;
1951 if (!brl_options)
1952 selected_console = i;
1953 strlcpy(c->name, name, sizeof(c->name));
1954 c->options = options;
1955 braille_set_options(c, brl_options);
1956
1957 c->index = idx;
1958 return 0;
1959}
1960/*
1961 * Set up a console. Called via do_early_param() in init/main.c
1962 * for each "console=" parameter in the boot command line.
1963 */
1964static int __init console_setup(char *str)
1965{
1966 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1967 char *s, *options, *brl_options = NULL;
1968 int idx;
1969
1970 if (_braille_console_setup(&str, &brl_options))
1971 return 1;
1972
1973 /*
1974 * Decode str into name, index, options.
1975 */
1976 if (str[0] >= '0' && str[0] <= '9') {
1977 strcpy(buf, "ttyS");
1978 strncpy(buf + 4, str, sizeof(buf) - 5);
1979 } else {
1980 strncpy(buf, str, sizeof(buf) - 1);
1981 }
1982 buf[sizeof(buf) - 1] = 0;
1983 options = strchr(str, ',');
1984 if (options)
1985 *(options++) = 0;
1986#ifdef __sparc__
1987 if (!strcmp(str, "ttya"))
1988 strcpy(buf, "ttyS0");
1989 if (!strcmp(str, "ttyb"))
1990 strcpy(buf, "ttyS1");
1991#endif
1992 for (s = buf; *s; s++)
1993 if (isdigit(*s) || *s == ',')
1994 break;
1995 idx = simple_strtoul(s, NULL, 10);
1996 *s = 0;
1997
1998 __add_preferred_console(buf, idx, options, brl_options);
1999 console_set_on_cmdline = 1;
2000 return 1;
2001}
2002__setup("console=", console_setup);
2003
2004/**
2005 * add_preferred_console - add a device to the list of preferred consoles.
2006 * @name: device name
2007 * @idx: device index
2008 * @options: options for this console
2009 *
2010 * The last preferred console added will be used for kernel messages
2011 * and stdin/out/err for init. Normally this is used by console_setup
2012 * above to handle user-supplied console arguments; however it can also
2013 * be used by arch-specific code either to override the user or more
2014 * commonly to provide a default console (ie from PROM variables) when
2015 * the user has not supplied one.
2016 */
2017int add_preferred_console(char *name, int idx, char *options)
2018{
2019 return __add_preferred_console(name, idx, options, NULL);
2020}
2021
2022bool console_suspend_enabled = true;
2023EXPORT_SYMBOL(console_suspend_enabled);
2024
2025static int __init console_suspend_disable(char *str)
2026{
2027 console_suspend_enabled = false;
2028 return 1;
2029}
2030__setup("no_console_suspend", console_suspend_disable);
2031module_param_named(console_suspend, console_suspend_enabled,
2032 bool, S_IRUGO | S_IWUSR);
2033MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2034 " and hibernate operations");
2035
2036/**
2037 * suspend_console - suspend the console subsystem
2038 *
2039 * This disables printk() while we go into suspend states
2040 */
2041void suspend_console(void)
2042{
2043 if (!console_suspend_enabled)
2044 return;
2045 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2046 console_lock();
2047 console_suspended = 1;
2048 up_console_sem();
2049}
2050
2051void resume_console(void)
2052{
2053 if (!console_suspend_enabled)
2054 return;
2055 down_console_sem();
2056 console_suspended = 0;
2057 console_unlock();
2058}
2059
2060/**
2061 * console_cpu_notify - print deferred console messages after CPU hotplug
2062 * @cpu: unused
2063 *
2064 * If printk() is called from a CPU that is not online yet, the messages
2065 * will be spooled but will not show up on the console. This function is
2066 * called when a new CPU comes online (or fails to come up), and ensures
2067 * that any such output gets printed.
2068 */
2069static int console_cpu_notify(unsigned int cpu)
2070{
2071 if (!cpuhp_tasks_frozen) {
2072 console_lock();
2073 console_unlock();
2074 }
2075 return 0;
2076}
2077
2078/**
2079 * console_lock - lock the console system for exclusive use.
2080 *
2081 * Acquires a lock which guarantees that the caller has
2082 * exclusive access to the console system and the console_drivers list.
2083 *
2084 * Can sleep, returns nothing.
2085 */
2086void console_lock(void)
2087{
2088 might_sleep();
2089
2090 down_console_sem();
2091 if (console_suspended)
2092 return;
2093 console_locked = 1;
2094 console_may_schedule = 1;
2095}
2096EXPORT_SYMBOL(console_lock);
2097
2098/**
2099 * console_trylock - try to lock the console system for exclusive use.
2100 *
2101 * Try to acquire a lock which guarantees that the caller has exclusive
2102 * access to the console system and the console_drivers list.
2103 *
2104 * returns 1 on success, and 0 on failure to acquire the lock.
2105 */
2106int console_trylock(void)
2107{
2108 if (down_trylock_console_sem())
2109 return 0;
2110 if (console_suspended) {
2111 up_console_sem();
2112 return 0;
2113 }
2114 console_locked = 1;
2115 /*
2116 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2117 * safe to schedule (e.g. calling printk while holding a spin_lock),
2118 * because preempt_disable()/preempt_enable() are just barriers there
2119 * and preempt_count() is always 0.
2120 *
2121 * RCU read sections have a separate preemption counter when
2122 * PREEMPT_RCU enabled thus we must take extra care and check
2123 * rcu_preempt_depth(), otherwise RCU read sections modify
2124 * preempt_count().
2125 */
2126 console_may_schedule = !oops_in_progress &&
2127 preemptible() &&
2128 !rcu_preempt_depth();
2129 return 1;
2130}
2131EXPORT_SYMBOL(console_trylock);
2132
2133int is_console_locked(void)
2134{
2135 return console_locked;
2136}
2137
2138/*
2139 * Check if we have any console that is capable of printing while cpu is
2140 * booting or shutting down. Requires console_sem.
2141 */
2142static int have_callable_console(void)
2143{
2144 struct console *con;
2145
2146 for_each_console(con)
2147 if ((con->flags & CON_ENABLED) &&
2148 (con->flags & CON_ANYTIME))
2149 return 1;
2150
2151 return 0;
2152}
2153
2154/*
2155 * Can we actually use the console at this time on this cpu?
2156 *
2157 * Console drivers may assume that per-cpu resources have been allocated. So
2158 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2159 * call them until this CPU is officially up.
2160 */
2161static inline int can_use_console(void)
2162{
2163 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2164}
2165
2166/**
2167 * console_unlock - unlock the console system
2168 *
2169 * Releases the console_lock which the caller holds on the console system
2170 * and the console driver list.
2171 *
2172 * While the console_lock was held, console output may have been buffered
2173 * by printk(). If this is the case, console_unlock(); emits
2174 * the output prior to releasing the lock.
2175 *
2176 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2177 *
2178 * console_unlock(); may be called from any context.
2179 */
2180void console_unlock(void)
2181{
2182 static char ext_text[CONSOLE_EXT_LOG_MAX];
2183 static char text[LOG_LINE_MAX + PREFIX_MAX];
2184 static u64 seen_seq;
2185 unsigned long flags;
2186 bool wake_klogd = false;
2187 bool do_cond_resched, retry;
2188
2189 if (console_suspended) {
2190 up_console_sem();
2191 return;
2192 }
2193
2194 /*
2195 * Console drivers are called under logbuf_lock, so
2196 * @console_may_schedule should be cleared before; however, we may
2197 * end up dumping a lot of lines, for example, if called from
2198 * console registration path, and should invoke cond_resched()
2199 * between lines if allowable. Not doing so can cause a very long
2200 * scheduling stall on a slow console leading to RCU stall and
2201 * softlockup warnings which exacerbate the issue with more
2202 * messages practically incapacitating the system.
2203 */
2204 do_cond_resched = console_may_schedule;
2205 console_may_schedule = 0;
2206
2207again:
2208 /*
2209 * We released the console_sem lock, so we need to recheck if
2210 * cpu is online and (if not) is there at least one CON_ANYTIME
2211 * console.
2212 */
2213 if (!can_use_console()) {
2214 console_locked = 0;
2215 up_console_sem();
2216 return;
2217 }
2218
2219 for (;;) {
2220 struct printk_log *msg;
2221 size_t ext_len = 0;
2222 size_t len;
2223 int level;
2224
2225 raw_spin_lock_irqsave(&logbuf_lock, flags);
2226 if (seen_seq != log_next_seq) {
2227 wake_klogd = true;
2228 seen_seq = log_next_seq;
2229 }
2230
2231 if (console_seq < log_first_seq) {
2232 len = sprintf(text, "** %u printk messages dropped ** ",
2233 (unsigned)(log_first_seq - console_seq));
2234
2235 /* messages are gone, move to first one */
2236 console_seq = log_first_seq;
2237 console_idx = log_first_idx;
2238 } else {
2239 len = 0;
2240 }
2241skip:
2242 if (console_seq == log_next_seq)
2243 break;
2244
2245 msg = log_from_idx(console_idx);
2246 level = msg->level;
2247 if (suppress_message_printing(level)) {
2248 /*
2249 * Skip record we have buffered and already printed
2250 * directly to the console when we received it, and
2251 * record that has level above the console loglevel.
2252 */
2253 console_idx = log_next(console_idx);
2254 console_seq++;
2255 goto skip;
2256 }
2257
2258 len += msg_print_text(msg, false, text + len, sizeof(text) - len);
2259 if (nr_ext_console_drivers) {
2260 ext_len = msg_print_ext_header(ext_text,
2261 sizeof(ext_text),
2262 msg, console_seq);
2263 ext_len += msg_print_ext_body(ext_text + ext_len,
2264 sizeof(ext_text) - ext_len,
2265 log_dict(msg), msg->dict_len,
2266 log_text(msg), msg->text_len);
2267 }
2268 console_idx = log_next(console_idx);
2269 console_seq++;
2270 raw_spin_unlock(&logbuf_lock);
2271
2272 stop_critical_timings(); /* don't trace print latency */
2273 call_console_drivers(level, ext_text, ext_len, text, len);
2274 start_critical_timings();
2275 local_irq_restore(flags);
2276
2277 if (do_cond_resched)
2278 cond_resched();
2279 }
2280 console_locked = 0;
2281
2282 /* Release the exclusive_console once it is used */
2283 if (unlikely(exclusive_console))
2284 exclusive_console = NULL;
2285
2286 raw_spin_unlock(&logbuf_lock);
2287
2288 up_console_sem();
2289
2290 /*
2291 * Someone could have filled up the buffer again, so re-check if there's
2292 * something to flush. In case we cannot trylock the console_sem again,
2293 * there's a new owner and the console_unlock() from them will do the
2294 * flush, no worries.
2295 */
2296 raw_spin_lock(&logbuf_lock);
2297 retry = console_seq != log_next_seq;
2298 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2299
2300 if (retry && console_trylock())
2301 goto again;
2302
2303 if (wake_klogd)
2304 wake_up_klogd();
2305}
2306EXPORT_SYMBOL(console_unlock);
2307
2308/**
2309 * console_conditional_schedule - yield the CPU if required
2310 *
2311 * If the console code is currently allowed to sleep, and
2312 * if this CPU should yield the CPU to another task, do
2313 * so here.
2314 *
2315 * Must be called within console_lock();.
2316 */
2317void __sched console_conditional_schedule(void)
2318{
2319 if (console_may_schedule)
2320 cond_resched();
2321}
2322EXPORT_SYMBOL(console_conditional_schedule);
2323
2324void console_unblank(void)
2325{
2326 struct console *c;
2327
2328 /*
2329 * console_unblank can no longer be called in interrupt context unless
2330 * oops_in_progress is set to 1..
2331 */
2332 if (oops_in_progress) {
2333 if (down_trylock_console_sem() != 0)
2334 return;
2335 } else
2336 console_lock();
2337
2338 console_locked = 1;
2339 console_may_schedule = 0;
2340 for_each_console(c)
2341 if ((c->flags & CON_ENABLED) && c->unblank)
2342 c->unblank();
2343 console_unlock();
2344}
2345
2346/**
2347 * console_flush_on_panic - flush console content on panic
2348 *
2349 * Immediately output all pending messages no matter what.
2350 */
2351void console_flush_on_panic(void)
2352{
2353 /*
2354 * If someone else is holding the console lock, trylock will fail
2355 * and may_schedule may be set. Ignore and proceed to unlock so
2356 * that messages are flushed out. As this can be called from any
2357 * context and we don't want to get preempted while flushing,
2358 * ensure may_schedule is cleared.
2359 */
2360 console_trylock();
2361 console_may_schedule = 0;
2362 console_unlock();
2363}
2364
2365/*
2366 * Return the console tty driver structure and its associated index
2367 */
2368struct tty_driver *console_device(int *index)
2369{
2370 struct console *c;
2371 struct tty_driver *driver = NULL;
2372
2373 console_lock();
2374 for_each_console(c) {
2375 if (!c->device)
2376 continue;
2377 driver = c->device(c, index);
2378 if (driver)
2379 break;
2380 }
2381 console_unlock();
2382 return driver;
2383}
2384
2385/*
2386 * Prevent further output on the passed console device so that (for example)
2387 * serial drivers can disable console output before suspending a port, and can
2388 * re-enable output afterwards.
2389 */
2390void console_stop(struct console *console)
2391{
2392 console_lock();
2393 console->flags &= ~CON_ENABLED;
2394 console_unlock();
2395}
2396EXPORT_SYMBOL(console_stop);
2397
2398void console_start(struct console *console)
2399{
2400 console_lock();
2401 console->flags |= CON_ENABLED;
2402 console_unlock();
2403}
2404EXPORT_SYMBOL(console_start);
2405
2406static int __read_mostly keep_bootcon;
2407
2408static int __init keep_bootcon_setup(char *str)
2409{
2410 keep_bootcon = 1;
2411 pr_info("debug: skip boot console de-registration.\n");
2412
2413 return 0;
2414}
2415
2416early_param("keep_bootcon", keep_bootcon_setup);
2417
2418/*
2419 * The console driver calls this routine during kernel initialization
2420 * to register the console printing procedure with printk() and to
2421 * print any messages that were printed by the kernel before the
2422 * console driver was initialized.
2423 *
2424 * This can happen pretty early during the boot process (because of
2425 * early_printk) - sometimes before setup_arch() completes - be careful
2426 * of what kernel features are used - they may not be initialised yet.
2427 *
2428 * There are two types of consoles - bootconsoles (early_printk) and
2429 * "real" consoles (everything which is not a bootconsole) which are
2430 * handled differently.
2431 * - Any number of bootconsoles can be registered at any time.
2432 * - As soon as a "real" console is registered, all bootconsoles
2433 * will be unregistered automatically.
2434 * - Once a "real" console is registered, any attempt to register a
2435 * bootconsoles will be rejected
2436 */
2437void register_console(struct console *newcon)
2438{
2439 int i;
2440 unsigned long flags;
2441 struct console *bcon = NULL;
2442 struct console_cmdline *c;
2443
2444 if (console_drivers)
2445 for_each_console(bcon)
2446 if (WARN(bcon == newcon,
2447 "console '%s%d' already registered\n",
2448 bcon->name, bcon->index))
2449 return;
2450
2451 /*
2452 * before we register a new CON_BOOT console, make sure we don't
2453 * already have a valid console
2454 */
2455 if (console_drivers && newcon->flags & CON_BOOT) {
2456 /* find the last or real console */
2457 for_each_console(bcon) {
2458 if (!(bcon->flags & CON_BOOT)) {
2459 pr_info("Too late to register bootconsole %s%d\n",
2460 newcon->name, newcon->index);
2461 return;
2462 }
2463 }
2464 }
2465
2466 if (console_drivers && console_drivers->flags & CON_BOOT)
2467 bcon = console_drivers;
2468
2469 if (preferred_console < 0 || bcon || !console_drivers)
2470 preferred_console = selected_console;
2471
2472 /*
2473 * See if we want to use this console driver. If we
2474 * didn't select a console we take the first one
2475 * that registers here.
2476 */
2477 if (preferred_console < 0) {
2478 if (newcon->index < 0)
2479 newcon->index = 0;
2480 if (newcon->setup == NULL ||
2481 newcon->setup(newcon, NULL) == 0) {
2482 newcon->flags |= CON_ENABLED;
2483 if (newcon->device) {
2484 newcon->flags |= CON_CONSDEV;
2485 preferred_console = 0;
2486 }
2487 }
2488 }
2489
2490 /*
2491 * See if this console matches one we selected on
2492 * the command line.
2493 */
2494 for (i = 0, c = console_cmdline;
2495 i < MAX_CMDLINECONSOLES && c->name[0];
2496 i++, c++) {
2497 if (!newcon->match ||
2498 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2499 /* default matching */
2500 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2501 if (strcmp(c->name, newcon->name) != 0)
2502 continue;
2503 if (newcon->index >= 0 &&
2504 newcon->index != c->index)
2505 continue;
2506 if (newcon->index < 0)
2507 newcon->index = c->index;
2508
2509 if (_braille_register_console(newcon, c))
2510 return;
2511
2512 if (newcon->setup &&
2513 newcon->setup(newcon, c->options) != 0)
2514 break;
2515 }
2516
2517 newcon->flags |= CON_ENABLED;
2518 if (i == selected_console) {
2519 newcon->flags |= CON_CONSDEV;
2520 preferred_console = selected_console;
2521 }
2522 break;
2523 }
2524
2525 if (!(newcon->flags & CON_ENABLED))
2526 return;
2527
2528 /*
2529 * If we have a bootconsole, and are switching to a real console,
2530 * don't print everything out again, since when the boot console, and
2531 * the real console are the same physical device, it's annoying to
2532 * see the beginning boot messages twice
2533 */
2534 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2535 newcon->flags &= ~CON_PRINTBUFFER;
2536
2537 /*
2538 * Put this console in the list - keep the
2539 * preferred driver at the head of the list.
2540 */
2541 console_lock();
2542 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2543 newcon->next = console_drivers;
2544 console_drivers = newcon;
2545 if (newcon->next)
2546 newcon->next->flags &= ~CON_CONSDEV;
2547 } else {
2548 newcon->next = console_drivers->next;
2549 console_drivers->next = newcon;
2550 }
2551
2552 if (newcon->flags & CON_EXTENDED)
2553 if (!nr_ext_console_drivers++)
2554 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2555
2556 if (newcon->flags & CON_PRINTBUFFER) {
2557 /*
2558 * console_unlock(); will print out the buffered messages
2559 * for us.
2560 */
2561 raw_spin_lock_irqsave(&logbuf_lock, flags);
2562 console_seq = syslog_seq;
2563 console_idx = syslog_idx;
2564 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2565 /*
2566 * We're about to replay the log buffer. Only do this to the
2567 * just-registered console to avoid excessive message spam to
2568 * the already-registered consoles.
2569 */
2570 exclusive_console = newcon;
2571 }
2572 console_unlock();
2573 console_sysfs_notify();
2574
2575 /*
2576 * By unregistering the bootconsoles after we enable the real console
2577 * we get the "console xxx enabled" message on all the consoles -
2578 * boot consoles, real consoles, etc - this is to ensure that end
2579 * users know there might be something in the kernel's log buffer that
2580 * went to the bootconsole (that they do not see on the real console)
2581 */
2582 pr_info("%sconsole [%s%d] enabled\n",
2583 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2584 newcon->name, newcon->index);
2585 if (bcon &&
2586 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2587 !keep_bootcon) {
2588 /* We need to iterate through all boot consoles, to make
2589 * sure we print everything out, before we unregister them.
2590 */
2591 for_each_console(bcon)
2592 if (bcon->flags & CON_BOOT)
2593 unregister_console(bcon);
2594 }
2595}
2596EXPORT_SYMBOL(register_console);
2597
2598int unregister_console(struct console *console)
2599{
2600 struct console *a, *b;
2601 int res;
2602
2603 pr_info("%sconsole [%s%d] disabled\n",
2604 (console->flags & CON_BOOT) ? "boot" : "" ,
2605 console->name, console->index);
2606
2607 res = _braille_unregister_console(console);
2608 if (res)
2609 return res;
2610
2611 res = 1;
2612 console_lock();
2613 if (console_drivers == console) {
2614 console_drivers=console->next;
2615 res = 0;
2616 } else if (console_drivers) {
2617 for (a=console_drivers->next, b=console_drivers ;
2618 a; b=a, a=b->next) {
2619 if (a == console) {
2620 b->next = a->next;
2621 res = 0;
2622 break;
2623 }
2624 }
2625 }
2626
2627 if (!res && (console->flags & CON_EXTENDED))
2628 nr_ext_console_drivers--;
2629
2630 /*
2631 * If this isn't the last console and it has CON_CONSDEV set, we
2632 * need to set it on the next preferred console.
2633 */
2634 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2635 console_drivers->flags |= CON_CONSDEV;
2636
2637 console->flags &= ~CON_ENABLED;
2638 console_unlock();
2639 console_sysfs_notify();
2640 return res;
2641}
2642EXPORT_SYMBOL(unregister_console);
2643
2644/*
2645 * Some boot consoles access data that is in the init section and which will
2646 * be discarded after the initcalls have been run. To make sure that no code
2647 * will access this data, unregister the boot consoles in a late initcall.
2648 *
2649 * If for some reason, such as deferred probe or the driver being a loadable
2650 * module, the real console hasn't registered yet at this point, there will
2651 * be a brief interval in which no messages are logged to the console, which
2652 * makes it difficult to diagnose problems that occur during this time.
2653 *
2654 * To mitigate this problem somewhat, only unregister consoles whose memory
2655 * intersects with the init section. Note that code exists elsewhere to get
2656 * rid of the boot console as soon as the proper console shows up, so there
2657 * won't be side-effects from postponing the removal.
2658 */
2659static int __init printk_late_init(void)
2660{
2661 struct console *con;
2662 int ret;
2663
2664 for_each_console(con) {
2665 if (!keep_bootcon && con->flags & CON_BOOT) {
2666 /*
2667 * Make sure to unregister boot consoles whose data
2668 * resides in the init section before the init section
2669 * is discarded. Boot consoles whose data will stick
2670 * around will automatically be unregistered when the
2671 * proper console replaces them.
2672 */
2673 if (init_section_intersects(con, sizeof(*con)))
2674 unregister_console(con);
2675 }
2676 }
2677 ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2678 console_cpu_notify);
2679 WARN_ON(ret < 0);
2680 ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2681 console_cpu_notify, NULL);
2682 WARN_ON(ret < 0);
2683 return 0;
2684}
2685late_initcall(printk_late_init);
2686
2687#if defined CONFIG_PRINTK
2688/*
2689 * Delayed printk version, for scheduler-internal messages:
2690 */
2691#define PRINTK_PENDING_WAKEUP 0x01
2692#define PRINTK_PENDING_OUTPUT 0x02
2693
2694static DEFINE_PER_CPU(int, printk_pending);
2695
2696static void wake_up_klogd_work_func(struct irq_work *irq_work)
2697{
2698 int pending = __this_cpu_xchg(printk_pending, 0);
2699
2700 if (pending & PRINTK_PENDING_OUTPUT) {
2701 /* If trylock fails, someone else is doing the printing */
2702 if (console_trylock())
2703 console_unlock();
2704 }
2705
2706 if (pending & PRINTK_PENDING_WAKEUP)
2707 wake_up_interruptible(&log_wait);
2708}
2709
2710static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2711 .func = wake_up_klogd_work_func,
2712 .flags = IRQ_WORK_LAZY,
2713};
2714
2715void wake_up_klogd(void)
2716{
2717 preempt_disable();
2718 if (waitqueue_active(&log_wait)) {
2719 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2720 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2721 }
2722 preempt_enable();
2723}
2724
2725int printk_deferred(const char *fmt, ...)
2726{
2727 va_list args;
2728 int r;
2729
2730 preempt_disable();
2731 va_start(args, fmt);
2732 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2733 va_end(args);
2734
2735 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2736 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2737 preempt_enable();
2738
2739 return r;
2740}
2741
2742/*
2743 * printk rate limiting, lifted from the networking subsystem.
2744 *
2745 * This enforces a rate limit: not more than 10 kernel messages
2746 * every 5s to make a denial-of-service attack impossible.
2747 */
2748DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2749
2750int __printk_ratelimit(const char *func)
2751{
2752 return ___ratelimit(&printk_ratelimit_state, func);
2753}
2754EXPORT_SYMBOL(__printk_ratelimit);
2755
2756/**
2757 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2758 * @caller_jiffies: pointer to caller's state
2759 * @interval_msecs: minimum interval between prints
2760 *
2761 * printk_timed_ratelimit() returns true if more than @interval_msecs
2762 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2763 * returned true.
2764 */
2765bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2766 unsigned int interval_msecs)
2767{
2768 unsigned long elapsed = jiffies - *caller_jiffies;
2769
2770 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2771 return false;
2772
2773 *caller_jiffies = jiffies;
2774 return true;
2775}
2776EXPORT_SYMBOL(printk_timed_ratelimit);
2777
2778static DEFINE_SPINLOCK(dump_list_lock);
2779static LIST_HEAD(dump_list);
2780
2781/**
2782 * kmsg_dump_register - register a kernel log dumper.
2783 * @dumper: pointer to the kmsg_dumper structure
2784 *
2785 * Adds a kernel log dumper to the system. The dump callback in the
2786 * structure will be called when the kernel oopses or panics and must be
2787 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2788 */
2789int kmsg_dump_register(struct kmsg_dumper *dumper)
2790{
2791 unsigned long flags;
2792 int err = -EBUSY;
2793
2794 /* The dump callback needs to be set */
2795 if (!dumper->dump)
2796 return -EINVAL;
2797
2798 spin_lock_irqsave(&dump_list_lock, flags);
2799 /* Don't allow registering multiple times */
2800 if (!dumper->registered) {
2801 dumper->registered = 1;
2802 list_add_tail_rcu(&dumper->list, &dump_list);
2803 err = 0;
2804 }
2805 spin_unlock_irqrestore(&dump_list_lock, flags);
2806
2807 return err;
2808}
2809EXPORT_SYMBOL_GPL(kmsg_dump_register);
2810
2811/**
2812 * kmsg_dump_unregister - unregister a kmsg dumper.
2813 * @dumper: pointer to the kmsg_dumper structure
2814 *
2815 * Removes a dump device from the system. Returns zero on success and
2816 * %-EINVAL otherwise.
2817 */
2818int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2819{
2820 unsigned long flags;
2821 int err = -EINVAL;
2822
2823 spin_lock_irqsave(&dump_list_lock, flags);
2824 if (dumper->registered) {
2825 dumper->registered = 0;
2826 list_del_rcu(&dumper->list);
2827 err = 0;
2828 }
2829 spin_unlock_irqrestore(&dump_list_lock, flags);
2830 synchronize_rcu();
2831
2832 return err;
2833}
2834EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2835
2836static bool always_kmsg_dump;
2837module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2838
2839/**
2840 * kmsg_dump - dump kernel log to kernel message dumpers.
2841 * @reason: the reason (oops, panic etc) for dumping
2842 *
2843 * Call each of the registered dumper's dump() callback, which can
2844 * retrieve the kmsg records with kmsg_dump_get_line() or
2845 * kmsg_dump_get_buffer().
2846 */
2847void kmsg_dump(enum kmsg_dump_reason reason)
2848{
2849 struct kmsg_dumper *dumper;
2850 unsigned long flags;
2851
2852 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2853 return;
2854
2855 rcu_read_lock();
2856 list_for_each_entry_rcu(dumper, &dump_list, list) {
2857 if (dumper->max_reason && reason > dumper->max_reason)
2858 continue;
2859
2860 /* initialize iterator with data about the stored records */
2861 dumper->active = true;
2862
2863 raw_spin_lock_irqsave(&logbuf_lock, flags);
2864 dumper->cur_seq = clear_seq;
2865 dumper->cur_idx = clear_idx;
2866 dumper->next_seq = log_next_seq;
2867 dumper->next_idx = log_next_idx;
2868 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2869
2870 /* invoke dumper which will iterate over records */
2871 dumper->dump(dumper, reason);
2872
2873 /* reset iterator */
2874 dumper->active = false;
2875 }
2876 rcu_read_unlock();
2877}
2878
2879/**
2880 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2881 * @dumper: registered kmsg dumper
2882 * @syslog: include the "<4>" prefixes
2883 * @line: buffer to copy the line to
2884 * @size: maximum size of the buffer
2885 * @len: length of line placed into buffer
2886 *
2887 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2888 * record, and copy one record into the provided buffer.
2889 *
2890 * Consecutive calls will return the next available record moving
2891 * towards the end of the buffer with the youngest messages.
2892 *
2893 * A return value of FALSE indicates that there are no more records to
2894 * read.
2895 *
2896 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2897 */
2898bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2899 char *line, size_t size, size_t *len)
2900{
2901 struct printk_log *msg;
2902 size_t l = 0;
2903 bool ret = false;
2904
2905 if (!dumper->active)
2906 goto out;
2907
2908 if (dumper->cur_seq < log_first_seq) {
2909 /* messages are gone, move to first available one */
2910 dumper->cur_seq = log_first_seq;
2911 dumper->cur_idx = log_first_idx;
2912 }
2913
2914 /* last entry */
2915 if (dumper->cur_seq >= log_next_seq)
2916 goto out;
2917
2918 msg = log_from_idx(dumper->cur_idx);
2919 l = msg_print_text(msg, syslog, line, size);
2920
2921 dumper->cur_idx = log_next(dumper->cur_idx);
2922 dumper->cur_seq++;
2923 ret = true;
2924out:
2925 if (len)
2926 *len = l;
2927 return ret;
2928}
2929
2930/**
2931 * kmsg_dump_get_line - retrieve one kmsg log line
2932 * @dumper: registered kmsg dumper
2933 * @syslog: include the "<4>" prefixes
2934 * @line: buffer to copy the line to
2935 * @size: maximum size of the buffer
2936 * @len: length of line placed into buffer
2937 *
2938 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2939 * record, and copy one record into the provided buffer.
2940 *
2941 * Consecutive calls will return the next available record moving
2942 * towards the end of the buffer with the youngest messages.
2943 *
2944 * A return value of FALSE indicates that there are no more records to
2945 * read.
2946 */
2947bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2948 char *line, size_t size, size_t *len)
2949{
2950 unsigned long flags;
2951 bool ret;
2952
2953 raw_spin_lock_irqsave(&logbuf_lock, flags);
2954 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2955 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2956
2957 return ret;
2958}
2959EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2960
2961/**
2962 * kmsg_dump_get_buffer - copy kmsg log lines
2963 * @dumper: registered kmsg dumper
2964 * @syslog: include the "<4>" prefixes
2965 * @buf: buffer to copy the line to
2966 * @size: maximum size of the buffer
2967 * @len: length of line placed into buffer
2968 *
2969 * Start at the end of the kmsg buffer and fill the provided buffer
2970 * with as many of the the *youngest* kmsg records that fit into it.
2971 * If the buffer is large enough, all available kmsg records will be
2972 * copied with a single call.
2973 *
2974 * Consecutive calls will fill the buffer with the next block of
2975 * available older records, not including the earlier retrieved ones.
2976 *
2977 * A return value of FALSE indicates that there are no more records to
2978 * read.
2979 */
2980bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2981 char *buf, size_t size, size_t *len)
2982{
2983 unsigned long flags;
2984 u64 seq;
2985 u32 idx;
2986 u64 next_seq;
2987 u32 next_idx;
2988 size_t l = 0;
2989 bool ret = false;
2990
2991 if (!dumper->active)
2992 goto out;
2993
2994 raw_spin_lock_irqsave(&logbuf_lock, flags);
2995 if (dumper->cur_seq < log_first_seq) {
2996 /* messages are gone, move to first available one */
2997 dumper->cur_seq = log_first_seq;
2998 dumper->cur_idx = log_first_idx;
2999 }
3000
3001 /* last entry */
3002 if (dumper->cur_seq >= dumper->next_seq) {
3003 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3004 goto out;
3005 }
3006
3007 /* calculate length of entire buffer */
3008 seq = dumper->cur_seq;
3009 idx = dumper->cur_idx;
3010 while (seq < dumper->next_seq) {
3011 struct printk_log *msg = log_from_idx(idx);
3012
3013 l += msg_print_text(msg, true, NULL, 0);
3014 idx = log_next(idx);
3015 seq++;
3016 }
3017
3018 /* move first record forward until length fits into the buffer */
3019 seq = dumper->cur_seq;
3020 idx = dumper->cur_idx;
3021 while (l > size && seq < dumper->next_seq) {
3022 struct printk_log *msg = log_from_idx(idx);
3023
3024 l -= msg_print_text(msg, true, NULL, 0);
3025 idx = log_next(idx);
3026 seq++;
3027 }
3028
3029 /* last message in next interation */
3030 next_seq = seq;
3031 next_idx = idx;
3032
3033 l = 0;
3034 while (seq < dumper->next_seq) {
3035 struct printk_log *msg = log_from_idx(idx);
3036
3037 l += msg_print_text(msg, syslog, buf + l, size - l);
3038 idx = log_next(idx);
3039 seq++;
3040 }
3041
3042 dumper->next_seq = next_seq;
3043 dumper->next_idx = next_idx;
3044 ret = true;
3045 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3046out:
3047 if (len)
3048 *len = l;
3049 return ret;
3050}
3051EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3052
3053/**
3054 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3055 * @dumper: registered kmsg dumper
3056 *
3057 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3058 * kmsg_dump_get_buffer() can be called again and used multiple
3059 * times within the same dumper.dump() callback.
3060 *
3061 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3062 */
3063void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3064{
3065 dumper->cur_seq = clear_seq;
3066 dumper->cur_idx = clear_idx;
3067 dumper->next_seq = log_next_seq;
3068 dumper->next_idx = log_next_idx;
3069}
3070
3071/**
3072 * kmsg_dump_rewind - reset the interator
3073 * @dumper: registered kmsg dumper
3074 *
3075 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3076 * kmsg_dump_get_buffer() can be called again and used multiple
3077 * times within the same dumper.dump() callback.
3078 */
3079void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3080{
3081 unsigned long flags;
3082
3083 raw_spin_lock_irqsave(&logbuf_lock, flags);
3084 kmsg_dump_rewind_nolock(dumper);
3085 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3086}
3087EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3088
3089static char dump_stack_arch_desc_str[128];
3090
3091/**
3092 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3093 * @fmt: printf-style format string
3094 * @...: arguments for the format string
3095 *
3096 * The configured string will be printed right after utsname during task
3097 * dumps. Usually used to add arch-specific system identifiers. If an
3098 * arch wants to make use of such an ID string, it should initialize this
3099 * as soon as possible during boot.
3100 */
3101void __init dump_stack_set_arch_desc(const char *fmt, ...)
3102{
3103 va_list args;
3104
3105 va_start(args, fmt);
3106 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3107 fmt, args);
3108 va_end(args);
3109}
3110
3111/**
3112 * dump_stack_print_info - print generic debug info for dump_stack()
3113 * @log_lvl: log level
3114 *
3115 * Arch-specific dump_stack() implementations can use this function to
3116 * print out the same debug information as the generic dump_stack().
3117 */
3118void dump_stack_print_info(const char *log_lvl)
3119{
3120 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3121 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3122 print_tainted(), init_utsname()->release,
3123 (int)strcspn(init_utsname()->version, " "),
3124 init_utsname()->version);
3125
3126 if (dump_stack_arch_desc_str[0] != '\0')
3127 printk("%sHardware name: %s\n",
3128 log_lvl, dump_stack_arch_desc_str);
3129
3130 print_worker_info(log_lvl, current);
3131}
3132
3133/**
3134 * show_regs_print_info - print generic debug info for show_regs()
3135 * @log_lvl: log level
3136 *
3137 * show_regs() implementations can use this function to print out generic
3138 * debug information.
3139 */
3140void show_regs_print_info(const char *log_lvl)
3141{
3142 dump_stack_print_info(log_lvl);
3143
3144 printk("%stask: %p task.stack: %p\n",
3145 log_lvl, current, task_stack_page(current));
3146}
3147
3148#endif
1/*
2 * linux/kernel/printk.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * Modified to make sys_syslog() more flexible: added commands to
7 * return the last 4k of kernel messages, regardless of whether
8 * they've been read or not. Added option to suppress kernel printk's
9 * to the console. Added hook for sending the console messages
10 * elsewhere, in preparation for a serial line console (someday).
11 * Ted Ts'o, 2/11/93.
12 * Modified for sysctl support, 1/8/97, Chris Horn.
13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
14 * manfred@colorfullife.com
15 * Rewrote bits to get rid of console_lock
16 * 01Mar01 Andrew Morton
17 */
18
19#include <linux/kernel.h>
20#include <linux/mm.h>
21#include <linux/tty.h>
22#include <linux/tty_driver.h>
23#include <linux/console.h>
24#include <linux/init.h>
25#include <linux/jiffies.h>
26#include <linux/nmi.h>
27#include <linux/module.h>
28#include <linux/moduleparam.h>
29#include <linux/interrupt.h> /* For in_interrupt() */
30#include <linux/delay.h>
31#include <linux/smp.h>
32#include <linux/security.h>
33#include <linux/bootmem.h>
34#include <linux/memblock.h>
35#include <linux/syscalls.h>
36#include <linux/kexec.h>
37#include <linux/kdb.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/notifier.h>
43#include <linux/rculist.h>
44#include <linux/poll.h>
45#include <linux/irq_work.h>
46#include <linux/utsname.h>
47#include <linux/ctype.h>
48#include <linux/uio.h>
49
50#include <asm/uaccess.h>
51#include <asm-generic/sections.h>
52
53#define CREATE_TRACE_POINTS
54#include <trace/events/printk.h>
55
56#include "console_cmdline.h"
57#include "braille.h"
58
59int console_printk[4] = {
60 CONSOLE_LOGLEVEL_DEFAULT, /* console_loglevel */
61 MESSAGE_LOGLEVEL_DEFAULT, /* default_message_loglevel */
62 CONSOLE_LOGLEVEL_MIN, /* minimum_console_loglevel */
63 CONSOLE_LOGLEVEL_DEFAULT, /* default_console_loglevel */
64};
65
66/*
67 * Low level drivers may need that to know if they can schedule in
68 * their unblank() callback or not. So let's export it.
69 */
70int oops_in_progress;
71EXPORT_SYMBOL(oops_in_progress);
72
73/*
74 * console_sem protects the console_drivers list, and also
75 * provides serialisation for access to the entire console
76 * driver system.
77 */
78static DEFINE_SEMAPHORE(console_sem);
79struct console *console_drivers;
80EXPORT_SYMBOL_GPL(console_drivers);
81
82#ifdef CONFIG_LOCKDEP
83static struct lockdep_map console_lock_dep_map = {
84 .name = "console_lock"
85};
86#endif
87
88/*
89 * Number of registered extended console drivers.
90 *
91 * If extended consoles are present, in-kernel cont reassembly is disabled
92 * and each fragment is stored as a separate log entry with proper
93 * continuation flag so that every emitted message has full metadata. This
94 * doesn't change the result for regular consoles or /proc/kmsg. For
95 * /dev/kmsg, as long as the reader concatenates messages according to
96 * consecutive continuation flags, the end result should be the same too.
97 */
98static int nr_ext_console_drivers;
99
100/*
101 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
102 * macros instead of functions so that _RET_IP_ contains useful information.
103 */
104#define down_console_sem() do { \
105 down(&console_sem);\
106 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
107} while (0)
108
109static int __down_trylock_console_sem(unsigned long ip)
110{
111 if (down_trylock(&console_sem))
112 return 1;
113 mutex_acquire(&console_lock_dep_map, 0, 1, ip);
114 return 0;
115}
116#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
117
118#define up_console_sem() do { \
119 mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
120 up(&console_sem);\
121} while (0)
122
123/*
124 * This is used for debugging the mess that is the VT code by
125 * keeping track if we have the console semaphore held. It's
126 * definitely not the perfect debug tool (we don't know if _WE_
127 * hold it and are racing, but it helps tracking those weird code
128 * paths in the console code where we end up in places I want
129 * locked without the console sempahore held).
130 */
131static int console_locked, console_suspended;
132
133/*
134 * If exclusive_console is non-NULL then only this console is to be printed to.
135 */
136static struct console *exclusive_console;
137
138/*
139 * Array of consoles built from command line options (console=)
140 */
141
142#define MAX_CMDLINECONSOLES 8
143
144static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
145
146static int selected_console = -1;
147static int preferred_console = -1;
148int console_set_on_cmdline;
149EXPORT_SYMBOL(console_set_on_cmdline);
150
151/* Flag: console code may call schedule() */
152static int console_may_schedule;
153
154/*
155 * The printk log buffer consists of a chain of concatenated variable
156 * length records. Every record starts with a record header, containing
157 * the overall length of the record.
158 *
159 * The heads to the first and last entry in the buffer, as well as the
160 * sequence numbers of these entries are maintained when messages are
161 * stored.
162 *
163 * If the heads indicate available messages, the length in the header
164 * tells the start next message. A length == 0 for the next message
165 * indicates a wrap-around to the beginning of the buffer.
166 *
167 * Every record carries the monotonic timestamp in microseconds, as well as
168 * the standard userspace syslog level and syslog facility. The usual
169 * kernel messages use LOG_KERN; userspace-injected messages always carry
170 * a matching syslog facility, by default LOG_USER. The origin of every
171 * message can be reliably determined that way.
172 *
173 * The human readable log message directly follows the message header. The
174 * length of the message text is stored in the header, the stored message
175 * is not terminated.
176 *
177 * Optionally, a message can carry a dictionary of properties (key/value pairs),
178 * to provide userspace with a machine-readable message context.
179 *
180 * Examples for well-defined, commonly used property names are:
181 * DEVICE=b12:8 device identifier
182 * b12:8 block dev_t
183 * c127:3 char dev_t
184 * n8 netdev ifindex
185 * +sound:card0 subsystem:devname
186 * SUBSYSTEM=pci driver-core subsystem name
187 *
188 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
189 * follows directly after a '=' character. Every property is terminated by
190 * a '\0' character. The last property is not terminated.
191 *
192 * Example of a message structure:
193 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
194 * 0008 34 00 record is 52 bytes long
195 * 000a 0b 00 text is 11 bytes long
196 * 000c 1f 00 dictionary is 23 bytes long
197 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
198 * 0010 69 74 27 73 20 61 20 6c "it's a l"
199 * 69 6e 65 "ine"
200 * 001b 44 45 56 49 43 "DEVIC"
201 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
202 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
203 * 67 "g"
204 * 0032 00 00 00 padding to next message header
205 *
206 * The 'struct printk_log' buffer header must never be directly exported to
207 * userspace, it is a kernel-private implementation detail that might
208 * need to be changed in the future, when the requirements change.
209 *
210 * /dev/kmsg exports the structured data in the following line format:
211 * "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
212 *
213 * Users of the export format should ignore possible additional values
214 * separated by ',', and find the message after the ';' character.
215 *
216 * The optional key/value pairs are attached as continuation lines starting
217 * with a space character and terminated by a newline. All possible
218 * non-prinatable characters are escaped in the "\xff" notation.
219 */
220
221enum log_flags {
222 LOG_NOCONS = 1, /* already flushed, do not print to console */
223 LOG_NEWLINE = 2, /* text ended with a newline */
224 LOG_PREFIX = 4, /* text started with a prefix */
225 LOG_CONT = 8, /* text is a fragment of a continuation line */
226};
227
228struct printk_log {
229 u64 ts_nsec; /* timestamp in nanoseconds */
230 u16 len; /* length of entire record */
231 u16 text_len; /* length of text buffer */
232 u16 dict_len; /* length of dictionary buffer */
233 u8 facility; /* syslog facility */
234 u8 flags:5; /* internal record flags */
235 u8 level:3; /* syslog level */
236}
237#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
238__packed __aligned(4)
239#endif
240;
241
242/*
243 * The logbuf_lock protects kmsg buffer, indices, counters. This can be taken
244 * within the scheduler's rq lock. It must be released before calling
245 * console_unlock() or anything else that might wake up a process.
246 */
247static DEFINE_RAW_SPINLOCK(logbuf_lock);
248
249#ifdef CONFIG_PRINTK
250DECLARE_WAIT_QUEUE_HEAD(log_wait);
251/* the next printk record to read by syslog(READ) or /proc/kmsg */
252static u64 syslog_seq;
253static u32 syslog_idx;
254static enum log_flags syslog_prev;
255static size_t syslog_partial;
256
257/* index and sequence number of the first record stored in the buffer */
258static u64 log_first_seq;
259static u32 log_first_idx;
260
261/* index and sequence number of the next record to store in the buffer */
262static u64 log_next_seq;
263static u32 log_next_idx;
264
265/* the next printk record to write to the console */
266static u64 console_seq;
267static u32 console_idx;
268static enum log_flags console_prev;
269
270/* the next printk record to read after the last 'clear' command */
271static u64 clear_seq;
272static u32 clear_idx;
273
274#define PREFIX_MAX 32
275#define LOG_LINE_MAX (1024 - PREFIX_MAX)
276
277#define LOG_LEVEL(v) ((v) & 0x07)
278#define LOG_FACILITY(v) ((v) >> 3 & 0xff)
279
280/* record buffer */
281#define LOG_ALIGN __alignof__(struct printk_log)
282#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
283static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
284static char *log_buf = __log_buf;
285static u32 log_buf_len = __LOG_BUF_LEN;
286
287/* Return log buffer address */
288char *log_buf_addr_get(void)
289{
290 return log_buf;
291}
292
293/* Return log buffer size */
294u32 log_buf_len_get(void)
295{
296 return log_buf_len;
297}
298
299/* human readable text of the record */
300static char *log_text(const struct printk_log *msg)
301{
302 return (char *)msg + sizeof(struct printk_log);
303}
304
305/* optional key/value pair dictionary attached to the record */
306static char *log_dict(const struct printk_log *msg)
307{
308 return (char *)msg + sizeof(struct printk_log) + msg->text_len;
309}
310
311/* get record by index; idx must point to valid msg */
312static struct printk_log *log_from_idx(u32 idx)
313{
314 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
315
316 /*
317 * A length == 0 record is the end of buffer marker. Wrap around and
318 * read the message at the start of the buffer.
319 */
320 if (!msg->len)
321 return (struct printk_log *)log_buf;
322 return msg;
323}
324
325/* get next record; idx must point to valid msg */
326static u32 log_next(u32 idx)
327{
328 struct printk_log *msg = (struct printk_log *)(log_buf + idx);
329
330 /* length == 0 indicates the end of the buffer; wrap */
331 /*
332 * A length == 0 record is the end of buffer marker. Wrap around and
333 * read the message at the start of the buffer as *this* one, and
334 * return the one after that.
335 */
336 if (!msg->len) {
337 msg = (struct printk_log *)log_buf;
338 return msg->len;
339 }
340 return idx + msg->len;
341}
342
343/*
344 * Check whether there is enough free space for the given message.
345 *
346 * The same values of first_idx and next_idx mean that the buffer
347 * is either empty or full.
348 *
349 * If the buffer is empty, we must respect the position of the indexes.
350 * They cannot be reset to the beginning of the buffer.
351 */
352static int logbuf_has_space(u32 msg_size, bool empty)
353{
354 u32 free;
355
356 if (log_next_idx > log_first_idx || empty)
357 free = max(log_buf_len - log_next_idx, log_first_idx);
358 else
359 free = log_first_idx - log_next_idx;
360
361 /*
362 * We need space also for an empty header that signalizes wrapping
363 * of the buffer.
364 */
365 return free >= msg_size + sizeof(struct printk_log);
366}
367
368static int log_make_free_space(u32 msg_size)
369{
370 while (log_first_seq < log_next_seq &&
371 !logbuf_has_space(msg_size, false)) {
372 /* drop old messages until we have enough contiguous space */
373 log_first_idx = log_next(log_first_idx);
374 log_first_seq++;
375 }
376
377 if (clear_seq < log_first_seq) {
378 clear_seq = log_first_seq;
379 clear_idx = log_first_idx;
380 }
381
382 /* sequence numbers are equal, so the log buffer is empty */
383 if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
384 return 0;
385
386 return -ENOMEM;
387}
388
389/* compute the message size including the padding bytes */
390static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
391{
392 u32 size;
393
394 size = sizeof(struct printk_log) + text_len + dict_len;
395 *pad_len = (-size) & (LOG_ALIGN - 1);
396 size += *pad_len;
397
398 return size;
399}
400
401/*
402 * Define how much of the log buffer we could take at maximum. The value
403 * must be greater than two. Note that only half of the buffer is available
404 * when the index points to the middle.
405 */
406#define MAX_LOG_TAKE_PART 4
407static const char trunc_msg[] = "<truncated>";
408
409static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
410 u16 *dict_len, u32 *pad_len)
411{
412 /*
413 * The message should not take the whole buffer. Otherwise, it might
414 * get removed too soon.
415 */
416 u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
417 if (*text_len > max_text_len)
418 *text_len = max_text_len;
419 /* enable the warning message */
420 *trunc_msg_len = strlen(trunc_msg);
421 /* disable the "dict" completely */
422 *dict_len = 0;
423 /* compute the size again, count also the warning message */
424 return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
425}
426
427/* insert record into the buffer, discard old ones, update heads */
428static int log_store(int facility, int level,
429 enum log_flags flags, u64 ts_nsec,
430 const char *dict, u16 dict_len,
431 const char *text, u16 text_len)
432{
433 struct printk_log *msg;
434 u32 size, pad_len;
435 u16 trunc_msg_len = 0;
436
437 /* number of '\0' padding bytes to next message */
438 size = msg_used_size(text_len, dict_len, &pad_len);
439
440 if (log_make_free_space(size)) {
441 /* truncate the message if it is too long for empty buffer */
442 size = truncate_msg(&text_len, &trunc_msg_len,
443 &dict_len, &pad_len);
444 /* survive when the log buffer is too small for trunc_msg */
445 if (log_make_free_space(size))
446 return 0;
447 }
448
449 if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
450 /*
451 * This message + an additional empty header does not fit
452 * at the end of the buffer. Add an empty header with len == 0
453 * to signify a wrap around.
454 */
455 memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
456 log_next_idx = 0;
457 }
458
459 /* fill message */
460 msg = (struct printk_log *)(log_buf + log_next_idx);
461 memcpy(log_text(msg), text, text_len);
462 msg->text_len = text_len;
463 if (trunc_msg_len) {
464 memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
465 msg->text_len += trunc_msg_len;
466 }
467 memcpy(log_dict(msg), dict, dict_len);
468 msg->dict_len = dict_len;
469 msg->facility = facility;
470 msg->level = level & 7;
471 msg->flags = flags & 0x1f;
472 if (ts_nsec > 0)
473 msg->ts_nsec = ts_nsec;
474 else
475 msg->ts_nsec = local_clock();
476 memset(log_dict(msg) + dict_len, 0, pad_len);
477 msg->len = size;
478
479 /* insert message */
480 log_next_idx += msg->len;
481 log_next_seq++;
482
483 return msg->text_len;
484}
485
486int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
487
488static int syslog_action_restricted(int type)
489{
490 if (dmesg_restrict)
491 return 1;
492 /*
493 * Unless restricted, we allow "read all" and "get buffer size"
494 * for everybody.
495 */
496 return type != SYSLOG_ACTION_READ_ALL &&
497 type != SYSLOG_ACTION_SIZE_BUFFER;
498}
499
500int check_syslog_permissions(int type, int source)
501{
502 /*
503 * If this is from /proc/kmsg and we've already opened it, then we've
504 * already done the capabilities checks at open time.
505 */
506 if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
507 goto ok;
508
509 if (syslog_action_restricted(type)) {
510 if (capable(CAP_SYSLOG))
511 goto ok;
512 /*
513 * For historical reasons, accept CAP_SYS_ADMIN too, with
514 * a warning.
515 */
516 if (capable(CAP_SYS_ADMIN)) {
517 pr_warn_once("%s (%d): Attempt to access syslog with "
518 "CAP_SYS_ADMIN but no CAP_SYSLOG "
519 "(deprecated).\n",
520 current->comm, task_pid_nr(current));
521 goto ok;
522 }
523 return -EPERM;
524 }
525ok:
526 return security_syslog(type);
527}
528EXPORT_SYMBOL_GPL(check_syslog_permissions);
529
530static void append_char(char **pp, char *e, char c)
531{
532 if (*pp < e)
533 *(*pp)++ = c;
534}
535
536static ssize_t msg_print_ext_header(char *buf, size_t size,
537 struct printk_log *msg, u64 seq,
538 enum log_flags prev_flags)
539{
540 u64 ts_usec = msg->ts_nsec;
541 char cont = '-';
542
543 do_div(ts_usec, 1000);
544
545 /*
546 * If we couldn't merge continuation line fragments during the print,
547 * export the stored flags to allow an optional external merge of the
548 * records. Merging the records isn't always neccessarily correct, like
549 * when we hit a race during printing. In most cases though, it produces
550 * better readable output. 'c' in the record flags mark the first
551 * fragment of a line, '+' the following.
552 */
553 if (msg->flags & LOG_CONT && !(prev_flags & LOG_CONT))
554 cont = 'c';
555 else if ((msg->flags & LOG_CONT) ||
556 ((prev_flags & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
557 cont = '+';
558
559 return scnprintf(buf, size, "%u,%llu,%llu,%c;",
560 (msg->facility << 3) | msg->level, seq, ts_usec, cont);
561}
562
563static ssize_t msg_print_ext_body(char *buf, size_t size,
564 char *dict, size_t dict_len,
565 char *text, size_t text_len)
566{
567 char *p = buf, *e = buf + size;
568 size_t i;
569
570 /* escape non-printable characters */
571 for (i = 0; i < text_len; i++) {
572 unsigned char c = text[i];
573
574 if (c < ' ' || c >= 127 || c == '\\')
575 p += scnprintf(p, e - p, "\\x%02x", c);
576 else
577 append_char(&p, e, c);
578 }
579 append_char(&p, e, '\n');
580
581 if (dict_len) {
582 bool line = true;
583
584 for (i = 0; i < dict_len; i++) {
585 unsigned char c = dict[i];
586
587 if (line) {
588 append_char(&p, e, ' ');
589 line = false;
590 }
591
592 if (c == '\0') {
593 append_char(&p, e, '\n');
594 line = true;
595 continue;
596 }
597
598 if (c < ' ' || c >= 127 || c == '\\') {
599 p += scnprintf(p, e - p, "\\x%02x", c);
600 continue;
601 }
602
603 append_char(&p, e, c);
604 }
605 append_char(&p, e, '\n');
606 }
607
608 return p - buf;
609}
610
611/* /dev/kmsg - userspace message inject/listen interface */
612struct devkmsg_user {
613 u64 seq;
614 u32 idx;
615 enum log_flags prev;
616 struct mutex lock;
617 char buf[CONSOLE_EXT_LOG_MAX];
618};
619
620static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
621{
622 char *buf, *line;
623 int level = default_message_loglevel;
624 int facility = 1; /* LOG_USER */
625 size_t len = iov_iter_count(from);
626 ssize_t ret = len;
627
628 if (len > LOG_LINE_MAX)
629 return -EINVAL;
630 buf = kmalloc(len+1, GFP_KERNEL);
631 if (buf == NULL)
632 return -ENOMEM;
633
634 buf[len] = '\0';
635 if (copy_from_iter(buf, len, from) != len) {
636 kfree(buf);
637 return -EFAULT;
638 }
639
640 /*
641 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
642 * the decimal value represents 32bit, the lower 3 bit are the log
643 * level, the rest are the log facility.
644 *
645 * If no prefix or no userspace facility is specified, we
646 * enforce LOG_USER, to be able to reliably distinguish
647 * kernel-generated messages from userspace-injected ones.
648 */
649 line = buf;
650 if (line[0] == '<') {
651 char *endp = NULL;
652 unsigned int u;
653
654 u = simple_strtoul(line + 1, &endp, 10);
655 if (endp && endp[0] == '>') {
656 level = LOG_LEVEL(u);
657 if (LOG_FACILITY(u) != 0)
658 facility = LOG_FACILITY(u);
659 endp++;
660 len -= endp - line;
661 line = endp;
662 }
663 }
664
665 printk_emit(facility, level, NULL, 0, "%s", line);
666 kfree(buf);
667 return ret;
668}
669
670static ssize_t devkmsg_read(struct file *file, char __user *buf,
671 size_t count, loff_t *ppos)
672{
673 struct devkmsg_user *user = file->private_data;
674 struct printk_log *msg;
675 size_t len;
676 ssize_t ret;
677
678 if (!user)
679 return -EBADF;
680
681 ret = mutex_lock_interruptible(&user->lock);
682 if (ret)
683 return ret;
684 raw_spin_lock_irq(&logbuf_lock);
685 while (user->seq == log_next_seq) {
686 if (file->f_flags & O_NONBLOCK) {
687 ret = -EAGAIN;
688 raw_spin_unlock_irq(&logbuf_lock);
689 goto out;
690 }
691
692 raw_spin_unlock_irq(&logbuf_lock);
693 ret = wait_event_interruptible(log_wait,
694 user->seq != log_next_seq);
695 if (ret)
696 goto out;
697 raw_spin_lock_irq(&logbuf_lock);
698 }
699
700 if (user->seq < log_first_seq) {
701 /* our last seen message is gone, return error and reset */
702 user->idx = log_first_idx;
703 user->seq = log_first_seq;
704 ret = -EPIPE;
705 raw_spin_unlock_irq(&logbuf_lock);
706 goto out;
707 }
708
709 msg = log_from_idx(user->idx);
710 len = msg_print_ext_header(user->buf, sizeof(user->buf),
711 msg, user->seq, user->prev);
712 len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
713 log_dict(msg), msg->dict_len,
714 log_text(msg), msg->text_len);
715
716 user->prev = msg->flags;
717 user->idx = log_next(user->idx);
718 user->seq++;
719 raw_spin_unlock_irq(&logbuf_lock);
720
721 if (len > count) {
722 ret = -EINVAL;
723 goto out;
724 }
725
726 if (copy_to_user(buf, user->buf, len)) {
727 ret = -EFAULT;
728 goto out;
729 }
730 ret = len;
731out:
732 mutex_unlock(&user->lock);
733 return ret;
734}
735
736static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
737{
738 struct devkmsg_user *user = file->private_data;
739 loff_t ret = 0;
740
741 if (!user)
742 return -EBADF;
743 if (offset)
744 return -ESPIPE;
745
746 raw_spin_lock_irq(&logbuf_lock);
747 switch (whence) {
748 case SEEK_SET:
749 /* the first record */
750 user->idx = log_first_idx;
751 user->seq = log_first_seq;
752 break;
753 case SEEK_DATA:
754 /*
755 * The first record after the last SYSLOG_ACTION_CLEAR,
756 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
757 * changes no global state, and does not clear anything.
758 */
759 user->idx = clear_idx;
760 user->seq = clear_seq;
761 break;
762 case SEEK_END:
763 /* after the last record */
764 user->idx = log_next_idx;
765 user->seq = log_next_seq;
766 break;
767 default:
768 ret = -EINVAL;
769 }
770 raw_spin_unlock_irq(&logbuf_lock);
771 return ret;
772}
773
774static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
775{
776 struct devkmsg_user *user = file->private_data;
777 int ret = 0;
778
779 if (!user)
780 return POLLERR|POLLNVAL;
781
782 poll_wait(file, &log_wait, wait);
783
784 raw_spin_lock_irq(&logbuf_lock);
785 if (user->seq < log_next_seq) {
786 /* return error when data has vanished underneath us */
787 if (user->seq < log_first_seq)
788 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
789 else
790 ret = POLLIN|POLLRDNORM;
791 }
792 raw_spin_unlock_irq(&logbuf_lock);
793
794 return ret;
795}
796
797static int devkmsg_open(struct inode *inode, struct file *file)
798{
799 struct devkmsg_user *user;
800 int err;
801
802 /* write-only does not need any file context */
803 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
804 return 0;
805
806 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
807 SYSLOG_FROM_READER);
808 if (err)
809 return err;
810
811 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
812 if (!user)
813 return -ENOMEM;
814
815 mutex_init(&user->lock);
816
817 raw_spin_lock_irq(&logbuf_lock);
818 user->idx = log_first_idx;
819 user->seq = log_first_seq;
820 raw_spin_unlock_irq(&logbuf_lock);
821
822 file->private_data = user;
823 return 0;
824}
825
826static int devkmsg_release(struct inode *inode, struct file *file)
827{
828 struct devkmsg_user *user = file->private_data;
829
830 if (!user)
831 return 0;
832
833 mutex_destroy(&user->lock);
834 kfree(user);
835 return 0;
836}
837
838const struct file_operations kmsg_fops = {
839 .open = devkmsg_open,
840 .read = devkmsg_read,
841 .write_iter = devkmsg_write,
842 .llseek = devkmsg_llseek,
843 .poll = devkmsg_poll,
844 .release = devkmsg_release,
845};
846
847#ifdef CONFIG_KEXEC_CORE
848/*
849 * This appends the listed symbols to /proc/vmcore
850 *
851 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
852 * obtain access to symbols that are otherwise very difficult to locate. These
853 * symbols are specifically used so that utilities can access and extract the
854 * dmesg log from a vmcore file after a crash.
855 */
856void log_buf_kexec_setup(void)
857{
858 VMCOREINFO_SYMBOL(log_buf);
859 VMCOREINFO_SYMBOL(log_buf_len);
860 VMCOREINFO_SYMBOL(log_first_idx);
861 VMCOREINFO_SYMBOL(clear_idx);
862 VMCOREINFO_SYMBOL(log_next_idx);
863 /*
864 * Export struct printk_log size and field offsets. User space tools can
865 * parse it and detect any changes to structure down the line.
866 */
867 VMCOREINFO_STRUCT_SIZE(printk_log);
868 VMCOREINFO_OFFSET(printk_log, ts_nsec);
869 VMCOREINFO_OFFSET(printk_log, len);
870 VMCOREINFO_OFFSET(printk_log, text_len);
871 VMCOREINFO_OFFSET(printk_log, dict_len);
872}
873#endif
874
875/* requested log_buf_len from kernel cmdline */
876static unsigned long __initdata new_log_buf_len;
877
878/* we practice scaling the ring buffer by powers of 2 */
879static void __init log_buf_len_update(unsigned size)
880{
881 if (size)
882 size = roundup_pow_of_two(size);
883 if (size > log_buf_len)
884 new_log_buf_len = size;
885}
886
887/* save requested log_buf_len since it's too early to process it */
888static int __init log_buf_len_setup(char *str)
889{
890 unsigned size = memparse(str, &str);
891
892 log_buf_len_update(size);
893
894 return 0;
895}
896early_param("log_buf_len", log_buf_len_setup);
897
898#ifdef CONFIG_SMP
899#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
900
901static void __init log_buf_add_cpu(void)
902{
903 unsigned int cpu_extra;
904
905 /*
906 * archs should set up cpu_possible_bits properly with
907 * set_cpu_possible() after setup_arch() but just in
908 * case lets ensure this is valid.
909 */
910 if (num_possible_cpus() == 1)
911 return;
912
913 cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
914
915 /* by default this will only continue through for large > 64 CPUs */
916 if (cpu_extra <= __LOG_BUF_LEN / 2)
917 return;
918
919 pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
920 __LOG_CPU_MAX_BUF_LEN);
921 pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
922 cpu_extra);
923 pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
924
925 log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
926}
927#else /* !CONFIG_SMP */
928static inline void log_buf_add_cpu(void) {}
929#endif /* CONFIG_SMP */
930
931void __init setup_log_buf(int early)
932{
933 unsigned long flags;
934 char *new_log_buf;
935 int free;
936
937 if (log_buf != __log_buf)
938 return;
939
940 if (!early && !new_log_buf_len)
941 log_buf_add_cpu();
942
943 if (!new_log_buf_len)
944 return;
945
946 if (early) {
947 new_log_buf =
948 memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
949 } else {
950 new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
951 LOG_ALIGN);
952 }
953
954 if (unlikely(!new_log_buf)) {
955 pr_err("log_buf_len: %ld bytes not available\n",
956 new_log_buf_len);
957 return;
958 }
959
960 raw_spin_lock_irqsave(&logbuf_lock, flags);
961 log_buf_len = new_log_buf_len;
962 log_buf = new_log_buf;
963 new_log_buf_len = 0;
964 free = __LOG_BUF_LEN - log_next_idx;
965 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
966 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
967
968 pr_info("log_buf_len: %d bytes\n", log_buf_len);
969 pr_info("early log buf free: %d(%d%%)\n",
970 free, (free * 100) / __LOG_BUF_LEN);
971}
972
973static bool __read_mostly ignore_loglevel;
974
975static int __init ignore_loglevel_setup(char *str)
976{
977 ignore_loglevel = true;
978 pr_info("debug: ignoring loglevel setting.\n");
979
980 return 0;
981}
982
983early_param("ignore_loglevel", ignore_loglevel_setup);
984module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
985MODULE_PARM_DESC(ignore_loglevel,
986 "ignore loglevel setting (prints all kernel messages to the console)");
987
988#ifdef CONFIG_BOOT_PRINTK_DELAY
989
990static int boot_delay; /* msecs delay after each printk during bootup */
991static unsigned long long loops_per_msec; /* based on boot_delay */
992
993static int __init boot_delay_setup(char *str)
994{
995 unsigned long lpj;
996
997 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
998 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
999
1000 get_option(&str, &boot_delay);
1001 if (boot_delay > 10 * 1000)
1002 boot_delay = 0;
1003
1004 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1005 "HZ: %d, loops_per_msec: %llu\n",
1006 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1007 return 0;
1008}
1009early_param("boot_delay", boot_delay_setup);
1010
1011static void boot_delay_msec(int level)
1012{
1013 unsigned long long k;
1014 unsigned long timeout;
1015
1016 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1017 || (level >= console_loglevel && !ignore_loglevel)) {
1018 return;
1019 }
1020
1021 k = (unsigned long long)loops_per_msec * boot_delay;
1022
1023 timeout = jiffies + msecs_to_jiffies(boot_delay);
1024 while (k) {
1025 k--;
1026 cpu_relax();
1027 /*
1028 * use (volatile) jiffies to prevent
1029 * compiler reduction; loop termination via jiffies
1030 * is secondary and may or may not happen.
1031 */
1032 if (time_after(jiffies, timeout))
1033 break;
1034 touch_nmi_watchdog();
1035 }
1036}
1037#else
1038static inline void boot_delay_msec(int level)
1039{
1040}
1041#endif
1042
1043static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1044module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1045
1046static size_t print_time(u64 ts, char *buf)
1047{
1048 unsigned long rem_nsec;
1049
1050 if (!printk_time)
1051 return 0;
1052
1053 rem_nsec = do_div(ts, 1000000000);
1054
1055 if (!buf)
1056 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1057
1058 return sprintf(buf, "[%5lu.%06lu] ",
1059 (unsigned long)ts, rem_nsec / 1000);
1060}
1061
1062static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1063{
1064 size_t len = 0;
1065 unsigned int prefix = (msg->facility << 3) | msg->level;
1066
1067 if (syslog) {
1068 if (buf) {
1069 len += sprintf(buf, "<%u>", prefix);
1070 } else {
1071 len += 3;
1072 if (prefix > 999)
1073 len += 3;
1074 else if (prefix > 99)
1075 len += 2;
1076 else if (prefix > 9)
1077 len++;
1078 }
1079 }
1080
1081 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1082 return len;
1083}
1084
1085static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1086 bool syslog, char *buf, size_t size)
1087{
1088 const char *text = log_text(msg);
1089 size_t text_size = msg->text_len;
1090 bool prefix = true;
1091 bool newline = true;
1092 size_t len = 0;
1093
1094 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
1095 prefix = false;
1096
1097 if (msg->flags & LOG_CONT) {
1098 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
1099 prefix = false;
1100
1101 if (!(msg->flags & LOG_NEWLINE))
1102 newline = false;
1103 }
1104
1105 do {
1106 const char *next = memchr(text, '\n', text_size);
1107 size_t text_len;
1108
1109 if (next) {
1110 text_len = next - text;
1111 next++;
1112 text_size -= next - text;
1113 } else {
1114 text_len = text_size;
1115 }
1116
1117 if (buf) {
1118 if (print_prefix(msg, syslog, NULL) +
1119 text_len + 1 >= size - len)
1120 break;
1121
1122 if (prefix)
1123 len += print_prefix(msg, syslog, buf + len);
1124 memcpy(buf + len, text, text_len);
1125 len += text_len;
1126 if (next || newline)
1127 buf[len++] = '\n';
1128 } else {
1129 /* SYSLOG_ACTION_* buffer size only calculation */
1130 if (prefix)
1131 len += print_prefix(msg, syslog, NULL);
1132 len += text_len;
1133 if (next || newline)
1134 len++;
1135 }
1136
1137 prefix = true;
1138 text = next;
1139 } while (text);
1140
1141 return len;
1142}
1143
1144static int syslog_print(char __user *buf, int size)
1145{
1146 char *text;
1147 struct printk_log *msg;
1148 int len = 0;
1149
1150 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1151 if (!text)
1152 return -ENOMEM;
1153
1154 while (size > 0) {
1155 size_t n;
1156 size_t skip;
1157
1158 raw_spin_lock_irq(&logbuf_lock);
1159 if (syslog_seq < log_first_seq) {
1160 /* messages are gone, move to first one */
1161 syslog_seq = log_first_seq;
1162 syslog_idx = log_first_idx;
1163 syslog_prev = 0;
1164 syslog_partial = 0;
1165 }
1166 if (syslog_seq == log_next_seq) {
1167 raw_spin_unlock_irq(&logbuf_lock);
1168 break;
1169 }
1170
1171 skip = syslog_partial;
1172 msg = log_from_idx(syslog_idx);
1173 n = msg_print_text(msg, syslog_prev, true, text,
1174 LOG_LINE_MAX + PREFIX_MAX);
1175 if (n - syslog_partial <= size) {
1176 /* message fits into buffer, move forward */
1177 syslog_idx = log_next(syslog_idx);
1178 syslog_seq++;
1179 syslog_prev = msg->flags;
1180 n -= syslog_partial;
1181 syslog_partial = 0;
1182 } else if (!len){
1183 /* partial read(), remember position */
1184 n = size;
1185 syslog_partial += n;
1186 } else
1187 n = 0;
1188 raw_spin_unlock_irq(&logbuf_lock);
1189
1190 if (!n)
1191 break;
1192
1193 if (copy_to_user(buf, text + skip, n)) {
1194 if (!len)
1195 len = -EFAULT;
1196 break;
1197 }
1198
1199 len += n;
1200 size -= n;
1201 buf += n;
1202 }
1203
1204 kfree(text);
1205 return len;
1206}
1207
1208static int syslog_print_all(char __user *buf, int size, bool clear)
1209{
1210 char *text;
1211 int len = 0;
1212
1213 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1214 if (!text)
1215 return -ENOMEM;
1216
1217 raw_spin_lock_irq(&logbuf_lock);
1218 if (buf) {
1219 u64 next_seq;
1220 u64 seq;
1221 u32 idx;
1222 enum log_flags prev;
1223
1224 /*
1225 * Find first record that fits, including all following records,
1226 * into the user-provided buffer for this dump.
1227 */
1228 seq = clear_seq;
1229 idx = clear_idx;
1230 prev = 0;
1231 while (seq < log_next_seq) {
1232 struct printk_log *msg = log_from_idx(idx);
1233
1234 len += msg_print_text(msg, prev, true, NULL, 0);
1235 prev = msg->flags;
1236 idx = log_next(idx);
1237 seq++;
1238 }
1239
1240 /* move first record forward until length fits into the buffer */
1241 seq = clear_seq;
1242 idx = clear_idx;
1243 prev = 0;
1244 while (len > size && seq < log_next_seq) {
1245 struct printk_log *msg = log_from_idx(idx);
1246
1247 len -= msg_print_text(msg, prev, true, NULL, 0);
1248 prev = msg->flags;
1249 idx = log_next(idx);
1250 seq++;
1251 }
1252
1253 /* last message fitting into this dump */
1254 next_seq = log_next_seq;
1255
1256 len = 0;
1257 while (len >= 0 && seq < next_seq) {
1258 struct printk_log *msg = log_from_idx(idx);
1259 int textlen;
1260
1261 textlen = msg_print_text(msg, prev, true, text,
1262 LOG_LINE_MAX + PREFIX_MAX);
1263 if (textlen < 0) {
1264 len = textlen;
1265 break;
1266 }
1267 idx = log_next(idx);
1268 seq++;
1269 prev = msg->flags;
1270
1271 raw_spin_unlock_irq(&logbuf_lock);
1272 if (copy_to_user(buf + len, text, textlen))
1273 len = -EFAULT;
1274 else
1275 len += textlen;
1276 raw_spin_lock_irq(&logbuf_lock);
1277
1278 if (seq < log_first_seq) {
1279 /* messages are gone, move to next one */
1280 seq = log_first_seq;
1281 idx = log_first_idx;
1282 prev = 0;
1283 }
1284 }
1285 }
1286
1287 if (clear) {
1288 clear_seq = log_next_seq;
1289 clear_idx = log_next_idx;
1290 }
1291 raw_spin_unlock_irq(&logbuf_lock);
1292
1293 kfree(text);
1294 return len;
1295}
1296
1297int do_syslog(int type, char __user *buf, int len, int source)
1298{
1299 bool clear = false;
1300 static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1301 int error;
1302
1303 error = check_syslog_permissions(type, source);
1304 if (error)
1305 goto out;
1306
1307 switch (type) {
1308 case SYSLOG_ACTION_CLOSE: /* Close log */
1309 break;
1310 case SYSLOG_ACTION_OPEN: /* Open log */
1311 break;
1312 case SYSLOG_ACTION_READ: /* Read from log */
1313 error = -EINVAL;
1314 if (!buf || len < 0)
1315 goto out;
1316 error = 0;
1317 if (!len)
1318 goto out;
1319 if (!access_ok(VERIFY_WRITE, buf, len)) {
1320 error = -EFAULT;
1321 goto out;
1322 }
1323 error = wait_event_interruptible(log_wait,
1324 syslog_seq != log_next_seq);
1325 if (error)
1326 goto out;
1327 error = syslog_print(buf, len);
1328 break;
1329 /* Read/clear last kernel messages */
1330 case SYSLOG_ACTION_READ_CLEAR:
1331 clear = true;
1332 /* FALL THRU */
1333 /* Read last kernel messages */
1334 case SYSLOG_ACTION_READ_ALL:
1335 error = -EINVAL;
1336 if (!buf || len < 0)
1337 goto out;
1338 error = 0;
1339 if (!len)
1340 goto out;
1341 if (!access_ok(VERIFY_WRITE, buf, len)) {
1342 error = -EFAULT;
1343 goto out;
1344 }
1345 error = syslog_print_all(buf, len, clear);
1346 break;
1347 /* Clear ring buffer */
1348 case SYSLOG_ACTION_CLEAR:
1349 syslog_print_all(NULL, 0, true);
1350 break;
1351 /* Disable logging to console */
1352 case SYSLOG_ACTION_CONSOLE_OFF:
1353 if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1354 saved_console_loglevel = console_loglevel;
1355 console_loglevel = minimum_console_loglevel;
1356 break;
1357 /* Enable logging to console */
1358 case SYSLOG_ACTION_CONSOLE_ON:
1359 if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1360 console_loglevel = saved_console_loglevel;
1361 saved_console_loglevel = LOGLEVEL_DEFAULT;
1362 }
1363 break;
1364 /* Set level of messages printed to console */
1365 case SYSLOG_ACTION_CONSOLE_LEVEL:
1366 error = -EINVAL;
1367 if (len < 1 || len > 8)
1368 goto out;
1369 if (len < minimum_console_loglevel)
1370 len = minimum_console_loglevel;
1371 console_loglevel = len;
1372 /* Implicitly re-enable logging to console */
1373 saved_console_loglevel = LOGLEVEL_DEFAULT;
1374 error = 0;
1375 break;
1376 /* Number of chars in the log buffer */
1377 case SYSLOG_ACTION_SIZE_UNREAD:
1378 raw_spin_lock_irq(&logbuf_lock);
1379 if (syslog_seq < log_first_seq) {
1380 /* messages are gone, move to first one */
1381 syslog_seq = log_first_seq;
1382 syslog_idx = log_first_idx;
1383 syslog_prev = 0;
1384 syslog_partial = 0;
1385 }
1386 if (source == SYSLOG_FROM_PROC) {
1387 /*
1388 * Short-cut for poll(/"proc/kmsg") which simply checks
1389 * for pending data, not the size; return the count of
1390 * records, not the length.
1391 */
1392 error = log_next_seq - syslog_seq;
1393 } else {
1394 u64 seq = syslog_seq;
1395 u32 idx = syslog_idx;
1396 enum log_flags prev = syslog_prev;
1397
1398 error = 0;
1399 while (seq < log_next_seq) {
1400 struct printk_log *msg = log_from_idx(idx);
1401
1402 error += msg_print_text(msg, prev, true, NULL, 0);
1403 idx = log_next(idx);
1404 seq++;
1405 prev = msg->flags;
1406 }
1407 error -= syslog_partial;
1408 }
1409 raw_spin_unlock_irq(&logbuf_lock);
1410 break;
1411 /* Size of the log buffer */
1412 case SYSLOG_ACTION_SIZE_BUFFER:
1413 error = log_buf_len;
1414 break;
1415 default:
1416 error = -EINVAL;
1417 break;
1418 }
1419out:
1420 return error;
1421}
1422
1423SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1424{
1425 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1426}
1427
1428/*
1429 * Call the console drivers, asking them to write out
1430 * log_buf[start] to log_buf[end - 1].
1431 * The console_lock must be held.
1432 */
1433static void call_console_drivers(int level,
1434 const char *ext_text, size_t ext_len,
1435 const char *text, size_t len)
1436{
1437 struct console *con;
1438
1439 trace_console(text, len);
1440
1441 if (level >= console_loglevel && !ignore_loglevel)
1442 return;
1443 if (!console_drivers)
1444 return;
1445
1446 for_each_console(con) {
1447 if (exclusive_console && con != exclusive_console)
1448 continue;
1449 if (!(con->flags & CON_ENABLED))
1450 continue;
1451 if (!con->write)
1452 continue;
1453 if (!cpu_online(smp_processor_id()) &&
1454 !(con->flags & CON_ANYTIME))
1455 continue;
1456 if (con->flags & CON_EXTENDED)
1457 con->write(con, ext_text, ext_len);
1458 else
1459 con->write(con, text, len);
1460 }
1461}
1462
1463/*
1464 * Zap console related locks when oopsing.
1465 * To leave time for slow consoles to print a full oops,
1466 * only zap at most once every 30 seconds.
1467 */
1468static void zap_locks(void)
1469{
1470 static unsigned long oops_timestamp;
1471
1472 if (time_after_eq(jiffies, oops_timestamp) &&
1473 !time_after(jiffies, oops_timestamp + 30 * HZ))
1474 return;
1475
1476 oops_timestamp = jiffies;
1477
1478 debug_locks_off();
1479 /* If a crash is occurring, make sure we can't deadlock */
1480 raw_spin_lock_init(&logbuf_lock);
1481 /* And make sure that we print immediately */
1482 sema_init(&console_sem, 1);
1483}
1484
1485int printk_delay_msec __read_mostly;
1486
1487static inline void printk_delay(void)
1488{
1489 if (unlikely(printk_delay_msec)) {
1490 int m = printk_delay_msec;
1491
1492 while (m--) {
1493 mdelay(1);
1494 touch_nmi_watchdog();
1495 }
1496 }
1497}
1498
1499/*
1500 * Continuation lines are buffered, and not committed to the record buffer
1501 * until the line is complete, or a race forces it. The line fragments
1502 * though, are printed immediately to the consoles to ensure everything has
1503 * reached the console in case of a kernel crash.
1504 */
1505static struct cont {
1506 char buf[LOG_LINE_MAX];
1507 size_t len; /* length == 0 means unused buffer */
1508 size_t cons; /* bytes written to console */
1509 struct task_struct *owner; /* task of first print*/
1510 u64 ts_nsec; /* time of first print */
1511 u8 level; /* log level of first message */
1512 u8 facility; /* log facility of first message */
1513 enum log_flags flags; /* prefix, newline flags */
1514 bool flushed:1; /* buffer sealed and committed */
1515} cont;
1516
1517static void cont_flush(enum log_flags flags)
1518{
1519 if (cont.flushed)
1520 return;
1521 if (cont.len == 0)
1522 return;
1523
1524 if (cont.cons) {
1525 /*
1526 * If a fragment of this line was directly flushed to the
1527 * console; wait for the console to pick up the rest of the
1528 * line. LOG_NOCONS suppresses a duplicated output.
1529 */
1530 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1531 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1532 cont.flags = flags;
1533 cont.flushed = true;
1534 } else {
1535 /*
1536 * If no fragment of this line ever reached the console,
1537 * just submit it to the store and free the buffer.
1538 */
1539 log_store(cont.facility, cont.level, flags, 0,
1540 NULL, 0, cont.buf, cont.len);
1541 cont.len = 0;
1542 }
1543}
1544
1545static bool cont_add(int facility, int level, const char *text, size_t len)
1546{
1547 if (cont.len && cont.flushed)
1548 return false;
1549
1550 /*
1551 * If ext consoles are present, flush and skip in-kernel
1552 * continuation. See nr_ext_console_drivers definition. Also, if
1553 * the line gets too long, split it up in separate records.
1554 */
1555 if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1556 cont_flush(LOG_CONT);
1557 return false;
1558 }
1559
1560 if (!cont.len) {
1561 cont.facility = facility;
1562 cont.level = level;
1563 cont.owner = current;
1564 cont.ts_nsec = local_clock();
1565 cont.flags = 0;
1566 cont.cons = 0;
1567 cont.flushed = false;
1568 }
1569
1570 memcpy(cont.buf + cont.len, text, len);
1571 cont.len += len;
1572
1573 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1574 cont_flush(LOG_CONT);
1575
1576 return true;
1577}
1578
1579static size_t cont_print_text(char *text, size_t size)
1580{
1581 size_t textlen = 0;
1582 size_t len;
1583
1584 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1585 textlen += print_time(cont.ts_nsec, text);
1586 size -= textlen;
1587 }
1588
1589 len = cont.len - cont.cons;
1590 if (len > 0) {
1591 if (len+1 > size)
1592 len = size-1;
1593 memcpy(text + textlen, cont.buf + cont.cons, len);
1594 textlen += len;
1595 cont.cons = cont.len;
1596 }
1597
1598 if (cont.flushed) {
1599 if (cont.flags & LOG_NEWLINE)
1600 text[textlen++] = '\n';
1601 /* got everything, release buffer */
1602 cont.len = 0;
1603 }
1604 return textlen;
1605}
1606
1607asmlinkage int vprintk_emit(int facility, int level,
1608 const char *dict, size_t dictlen,
1609 const char *fmt, va_list args)
1610{
1611 static bool recursion_bug;
1612 static char textbuf[LOG_LINE_MAX];
1613 char *text = textbuf;
1614 size_t text_len = 0;
1615 enum log_flags lflags = 0;
1616 unsigned long flags;
1617 int this_cpu;
1618 int printed_len = 0;
1619 bool in_sched = false;
1620 /* cpu currently holding logbuf_lock in this function */
1621 static unsigned int logbuf_cpu = UINT_MAX;
1622
1623 if (level == LOGLEVEL_SCHED) {
1624 level = LOGLEVEL_DEFAULT;
1625 in_sched = true;
1626 }
1627
1628 boot_delay_msec(level);
1629 printk_delay();
1630
1631 local_irq_save(flags);
1632 this_cpu = smp_processor_id();
1633
1634 /*
1635 * Ouch, printk recursed into itself!
1636 */
1637 if (unlikely(logbuf_cpu == this_cpu)) {
1638 /*
1639 * If a crash is occurring during printk() on this CPU,
1640 * then try to get the crash message out but make sure
1641 * we can't deadlock. Otherwise just return to avoid the
1642 * recursion and return - but flag the recursion so that
1643 * it can be printed at the next appropriate moment:
1644 */
1645 if (!oops_in_progress && !lockdep_recursing(current)) {
1646 recursion_bug = true;
1647 local_irq_restore(flags);
1648 return 0;
1649 }
1650 zap_locks();
1651 }
1652
1653 lockdep_off();
1654 /* This stops the holder of console_sem just where we want him */
1655 raw_spin_lock(&logbuf_lock);
1656 logbuf_cpu = this_cpu;
1657
1658 if (unlikely(recursion_bug)) {
1659 static const char recursion_msg[] =
1660 "BUG: recent printk recursion!";
1661
1662 recursion_bug = false;
1663 /* emit KERN_CRIT message */
1664 printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1665 NULL, 0, recursion_msg,
1666 strlen(recursion_msg));
1667 }
1668
1669 /*
1670 * The printf needs to come first; we need the syslog
1671 * prefix which might be passed-in as a parameter.
1672 */
1673 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1674
1675 /* mark and strip a trailing newline */
1676 if (text_len && text[text_len-1] == '\n') {
1677 text_len--;
1678 lflags |= LOG_NEWLINE;
1679 }
1680
1681 /* strip kernel syslog prefix and extract log level or control flags */
1682 if (facility == 0) {
1683 int kern_level = printk_get_level(text);
1684
1685 if (kern_level) {
1686 const char *end_of_header = printk_skip_level(text);
1687 switch (kern_level) {
1688 case '0' ... '7':
1689 if (level == LOGLEVEL_DEFAULT)
1690 level = kern_level - '0';
1691 /* fallthrough */
1692 case 'd': /* KERN_DEFAULT */
1693 lflags |= LOG_PREFIX;
1694 }
1695 /*
1696 * No need to check length here because vscnprintf
1697 * put '\0' at the end of the string. Only valid and
1698 * newly printed level is detected.
1699 */
1700 text_len -= end_of_header - text;
1701 text = (char *)end_of_header;
1702 }
1703 }
1704
1705 if (level == LOGLEVEL_DEFAULT)
1706 level = default_message_loglevel;
1707
1708 if (dict)
1709 lflags |= LOG_PREFIX|LOG_NEWLINE;
1710
1711 if (!(lflags & LOG_NEWLINE)) {
1712 /*
1713 * Flush the conflicting buffer. An earlier newline was missing,
1714 * or another task also prints continuation lines.
1715 */
1716 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1717 cont_flush(LOG_NEWLINE);
1718
1719 /* buffer line if possible, otherwise store it right away */
1720 if (cont_add(facility, level, text, text_len))
1721 printed_len += text_len;
1722 else
1723 printed_len += log_store(facility, level,
1724 lflags | LOG_CONT, 0,
1725 dict, dictlen, text, text_len);
1726 } else {
1727 bool stored = false;
1728
1729 /*
1730 * If an earlier newline was missing and it was the same task,
1731 * either merge it with the current buffer and flush, or if
1732 * there was a race with interrupts (prefix == true) then just
1733 * flush it out and store this line separately.
1734 * If the preceding printk was from a different task and missed
1735 * a newline, flush and append the newline.
1736 */
1737 if (cont.len) {
1738 if (cont.owner == current && !(lflags & LOG_PREFIX))
1739 stored = cont_add(facility, level, text,
1740 text_len);
1741 cont_flush(LOG_NEWLINE);
1742 }
1743
1744 if (stored)
1745 printed_len += text_len;
1746 else
1747 printed_len += log_store(facility, level, lflags, 0,
1748 dict, dictlen, text, text_len);
1749 }
1750
1751 logbuf_cpu = UINT_MAX;
1752 raw_spin_unlock(&logbuf_lock);
1753 lockdep_on();
1754 local_irq_restore(flags);
1755
1756 /* If called from the scheduler, we can not call up(). */
1757 if (!in_sched) {
1758 lockdep_off();
1759 /*
1760 * Try to acquire and then immediately release the console
1761 * semaphore. The release will print out buffers and wake up
1762 * /dev/kmsg and syslog() users.
1763 */
1764 if (console_trylock())
1765 console_unlock();
1766 lockdep_on();
1767 }
1768
1769 return printed_len;
1770}
1771EXPORT_SYMBOL(vprintk_emit);
1772
1773asmlinkage int vprintk(const char *fmt, va_list args)
1774{
1775 return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1776}
1777EXPORT_SYMBOL(vprintk);
1778
1779asmlinkage int printk_emit(int facility, int level,
1780 const char *dict, size_t dictlen,
1781 const char *fmt, ...)
1782{
1783 va_list args;
1784 int r;
1785
1786 va_start(args, fmt);
1787 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1788 va_end(args);
1789
1790 return r;
1791}
1792EXPORT_SYMBOL(printk_emit);
1793
1794int vprintk_default(const char *fmt, va_list args)
1795{
1796 int r;
1797
1798#ifdef CONFIG_KGDB_KDB
1799 if (unlikely(kdb_trap_printk)) {
1800 r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1801 return r;
1802 }
1803#endif
1804 r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1805
1806 return r;
1807}
1808EXPORT_SYMBOL_GPL(vprintk_default);
1809
1810/*
1811 * This allows printk to be diverted to another function per cpu.
1812 * This is useful for calling printk functions from within NMI
1813 * without worrying about race conditions that can lock up the
1814 * box.
1815 */
1816DEFINE_PER_CPU(printk_func_t, printk_func) = vprintk_default;
1817
1818/**
1819 * printk - print a kernel message
1820 * @fmt: format string
1821 *
1822 * This is printk(). It can be called from any context. We want it to work.
1823 *
1824 * We try to grab the console_lock. If we succeed, it's easy - we log the
1825 * output and call the console drivers. If we fail to get the semaphore, we
1826 * place the output into the log buffer and return. The current holder of
1827 * the console_sem will notice the new output in console_unlock(); and will
1828 * send it to the consoles before releasing the lock.
1829 *
1830 * One effect of this deferred printing is that code which calls printk() and
1831 * then changes console_loglevel may break. This is because console_loglevel
1832 * is inspected when the actual printing occurs.
1833 *
1834 * See also:
1835 * printf(3)
1836 *
1837 * See the vsnprintf() documentation for format string extensions over C99.
1838 */
1839asmlinkage __visible int printk(const char *fmt, ...)
1840{
1841 printk_func_t vprintk_func;
1842 va_list args;
1843 int r;
1844
1845 va_start(args, fmt);
1846
1847 /*
1848 * If a caller overrides the per_cpu printk_func, then it needs
1849 * to disable preemption when calling printk(). Otherwise
1850 * the printk_func should be set to the default. No need to
1851 * disable preemption here.
1852 */
1853 vprintk_func = this_cpu_read(printk_func);
1854 r = vprintk_func(fmt, args);
1855
1856 va_end(args);
1857
1858 return r;
1859}
1860EXPORT_SYMBOL(printk);
1861
1862#else /* CONFIG_PRINTK */
1863
1864#define LOG_LINE_MAX 0
1865#define PREFIX_MAX 0
1866
1867static u64 syslog_seq;
1868static u32 syslog_idx;
1869static u64 console_seq;
1870static u32 console_idx;
1871static enum log_flags syslog_prev;
1872static u64 log_first_seq;
1873static u32 log_first_idx;
1874static u64 log_next_seq;
1875static enum log_flags console_prev;
1876static struct cont {
1877 size_t len;
1878 size_t cons;
1879 u8 level;
1880 bool flushed:1;
1881} cont;
1882static char *log_text(const struct printk_log *msg) { return NULL; }
1883static char *log_dict(const struct printk_log *msg) { return NULL; }
1884static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1885static u32 log_next(u32 idx) { return 0; }
1886static ssize_t msg_print_ext_header(char *buf, size_t size,
1887 struct printk_log *msg, u64 seq,
1888 enum log_flags prev_flags) { return 0; }
1889static ssize_t msg_print_ext_body(char *buf, size_t size,
1890 char *dict, size_t dict_len,
1891 char *text, size_t text_len) { return 0; }
1892static void call_console_drivers(int level,
1893 const char *ext_text, size_t ext_len,
1894 const char *text, size_t len) {}
1895static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
1896 bool syslog, char *buf, size_t size) { return 0; }
1897static size_t cont_print_text(char *text, size_t size) { return 0; }
1898
1899/* Still needs to be defined for users */
1900DEFINE_PER_CPU(printk_func_t, printk_func);
1901
1902#endif /* CONFIG_PRINTK */
1903
1904#ifdef CONFIG_EARLY_PRINTK
1905struct console *early_console;
1906
1907asmlinkage __visible void early_printk(const char *fmt, ...)
1908{
1909 va_list ap;
1910 char buf[512];
1911 int n;
1912
1913 if (!early_console)
1914 return;
1915
1916 va_start(ap, fmt);
1917 n = vscnprintf(buf, sizeof(buf), fmt, ap);
1918 va_end(ap);
1919
1920 early_console->write(early_console, buf, n);
1921}
1922#endif
1923
1924static int __add_preferred_console(char *name, int idx, char *options,
1925 char *brl_options)
1926{
1927 struct console_cmdline *c;
1928 int i;
1929
1930 /*
1931 * See if this tty is not yet registered, and
1932 * if we have a slot free.
1933 */
1934 for (i = 0, c = console_cmdline;
1935 i < MAX_CMDLINECONSOLES && c->name[0];
1936 i++, c++) {
1937 if (strcmp(c->name, name) == 0 && c->index == idx) {
1938 if (!brl_options)
1939 selected_console = i;
1940 return 0;
1941 }
1942 }
1943 if (i == MAX_CMDLINECONSOLES)
1944 return -E2BIG;
1945 if (!brl_options)
1946 selected_console = i;
1947 strlcpy(c->name, name, sizeof(c->name));
1948 c->options = options;
1949 braille_set_options(c, brl_options);
1950
1951 c->index = idx;
1952 return 0;
1953}
1954/*
1955 * Set up a console. Called via do_early_param() in init/main.c
1956 * for each "console=" parameter in the boot command line.
1957 */
1958static int __init console_setup(char *str)
1959{
1960 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1961 char *s, *options, *brl_options = NULL;
1962 int idx;
1963
1964 if (_braille_console_setup(&str, &brl_options))
1965 return 1;
1966
1967 /*
1968 * Decode str into name, index, options.
1969 */
1970 if (str[0] >= '0' && str[0] <= '9') {
1971 strcpy(buf, "ttyS");
1972 strncpy(buf + 4, str, sizeof(buf) - 5);
1973 } else {
1974 strncpy(buf, str, sizeof(buf) - 1);
1975 }
1976 buf[sizeof(buf) - 1] = 0;
1977 options = strchr(str, ',');
1978 if (options)
1979 *(options++) = 0;
1980#ifdef __sparc__
1981 if (!strcmp(str, "ttya"))
1982 strcpy(buf, "ttyS0");
1983 if (!strcmp(str, "ttyb"))
1984 strcpy(buf, "ttyS1");
1985#endif
1986 for (s = buf; *s; s++)
1987 if (isdigit(*s) || *s == ',')
1988 break;
1989 idx = simple_strtoul(s, NULL, 10);
1990 *s = 0;
1991
1992 __add_preferred_console(buf, idx, options, brl_options);
1993 console_set_on_cmdline = 1;
1994 return 1;
1995}
1996__setup("console=", console_setup);
1997
1998/**
1999 * add_preferred_console - add a device to the list of preferred consoles.
2000 * @name: device name
2001 * @idx: device index
2002 * @options: options for this console
2003 *
2004 * The last preferred console added will be used for kernel messages
2005 * and stdin/out/err for init. Normally this is used by console_setup
2006 * above to handle user-supplied console arguments; however it can also
2007 * be used by arch-specific code either to override the user or more
2008 * commonly to provide a default console (ie from PROM variables) when
2009 * the user has not supplied one.
2010 */
2011int add_preferred_console(char *name, int idx, char *options)
2012{
2013 return __add_preferred_console(name, idx, options, NULL);
2014}
2015
2016bool console_suspend_enabled = true;
2017EXPORT_SYMBOL(console_suspend_enabled);
2018
2019static int __init console_suspend_disable(char *str)
2020{
2021 console_suspend_enabled = false;
2022 return 1;
2023}
2024__setup("no_console_suspend", console_suspend_disable);
2025module_param_named(console_suspend, console_suspend_enabled,
2026 bool, S_IRUGO | S_IWUSR);
2027MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2028 " and hibernate operations");
2029
2030/**
2031 * suspend_console - suspend the console subsystem
2032 *
2033 * This disables printk() while we go into suspend states
2034 */
2035void suspend_console(void)
2036{
2037 if (!console_suspend_enabled)
2038 return;
2039 printk("Suspending console(s) (use no_console_suspend to debug)\n");
2040 console_lock();
2041 console_suspended = 1;
2042 up_console_sem();
2043}
2044
2045void resume_console(void)
2046{
2047 if (!console_suspend_enabled)
2048 return;
2049 down_console_sem();
2050 console_suspended = 0;
2051 console_unlock();
2052}
2053
2054/**
2055 * console_cpu_notify - print deferred console messages after CPU hotplug
2056 * @self: notifier struct
2057 * @action: CPU hotplug event
2058 * @hcpu: unused
2059 *
2060 * If printk() is called from a CPU that is not online yet, the messages
2061 * will be spooled but will not show up on the console. This function is
2062 * called when a new CPU comes online (or fails to come up), and ensures
2063 * that any such output gets printed.
2064 */
2065static int console_cpu_notify(struct notifier_block *self,
2066 unsigned long action, void *hcpu)
2067{
2068 switch (action) {
2069 case CPU_ONLINE:
2070 case CPU_DEAD:
2071 case CPU_DOWN_FAILED:
2072 case CPU_UP_CANCELED:
2073 console_lock();
2074 console_unlock();
2075 }
2076 return NOTIFY_OK;
2077}
2078
2079/**
2080 * console_lock - lock the console system for exclusive use.
2081 *
2082 * Acquires a lock which guarantees that the caller has
2083 * exclusive access to the console system and the console_drivers list.
2084 *
2085 * Can sleep, returns nothing.
2086 */
2087void console_lock(void)
2088{
2089 might_sleep();
2090
2091 down_console_sem();
2092 if (console_suspended)
2093 return;
2094 console_locked = 1;
2095 console_may_schedule = 1;
2096}
2097EXPORT_SYMBOL(console_lock);
2098
2099/**
2100 * console_trylock - try to lock the console system for exclusive use.
2101 *
2102 * Try to acquire a lock which guarantees that the caller has exclusive
2103 * access to the console system and the console_drivers list.
2104 *
2105 * returns 1 on success, and 0 on failure to acquire the lock.
2106 */
2107int console_trylock(void)
2108{
2109 if (down_trylock_console_sem())
2110 return 0;
2111 if (console_suspended) {
2112 up_console_sem();
2113 return 0;
2114 }
2115 console_locked = 1;
2116 /*
2117 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2118 * safe to schedule (e.g. calling printk while holding a spin_lock),
2119 * because preempt_disable()/preempt_enable() are just barriers there
2120 * and preempt_count() is always 0.
2121 *
2122 * RCU read sections have a separate preemption counter when
2123 * PREEMPT_RCU enabled thus we must take extra care and check
2124 * rcu_preempt_depth(), otherwise RCU read sections modify
2125 * preempt_count().
2126 */
2127 console_may_schedule = !oops_in_progress &&
2128 preemptible() &&
2129 !rcu_preempt_depth();
2130 return 1;
2131}
2132EXPORT_SYMBOL(console_trylock);
2133
2134int is_console_locked(void)
2135{
2136 return console_locked;
2137}
2138
2139/*
2140 * Check if we have any console that is capable of printing while cpu is
2141 * booting or shutting down. Requires console_sem.
2142 */
2143static int have_callable_console(void)
2144{
2145 struct console *con;
2146
2147 for_each_console(con)
2148 if ((con->flags & CON_ENABLED) &&
2149 (con->flags & CON_ANYTIME))
2150 return 1;
2151
2152 return 0;
2153}
2154
2155/*
2156 * Can we actually use the console at this time on this cpu?
2157 *
2158 * Console drivers may assume that per-cpu resources have been allocated. So
2159 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2160 * call them until this CPU is officially up.
2161 */
2162static inline int can_use_console(void)
2163{
2164 return cpu_online(raw_smp_processor_id()) || have_callable_console();
2165}
2166
2167static void console_cont_flush(char *text, size_t size)
2168{
2169 unsigned long flags;
2170 size_t len;
2171
2172 raw_spin_lock_irqsave(&logbuf_lock, flags);
2173
2174 if (!cont.len)
2175 goto out;
2176
2177 /*
2178 * We still queue earlier records, likely because the console was
2179 * busy. The earlier ones need to be printed before this one, we
2180 * did not flush any fragment so far, so just let it queue up.
2181 */
2182 if (console_seq < log_next_seq && !cont.cons)
2183 goto out;
2184
2185 len = cont_print_text(text, size);
2186 raw_spin_unlock(&logbuf_lock);
2187 stop_critical_timings();
2188 call_console_drivers(cont.level, NULL, 0, text, len);
2189 start_critical_timings();
2190 local_irq_restore(flags);
2191 return;
2192out:
2193 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2194}
2195
2196/**
2197 * console_unlock - unlock the console system
2198 *
2199 * Releases the console_lock which the caller holds on the console system
2200 * and the console driver list.
2201 *
2202 * While the console_lock was held, console output may have been buffered
2203 * by printk(). If this is the case, console_unlock(); emits
2204 * the output prior to releasing the lock.
2205 *
2206 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2207 *
2208 * console_unlock(); may be called from any context.
2209 */
2210void console_unlock(void)
2211{
2212 static char ext_text[CONSOLE_EXT_LOG_MAX];
2213 static char text[LOG_LINE_MAX + PREFIX_MAX];
2214 static u64 seen_seq;
2215 unsigned long flags;
2216 bool wake_klogd = false;
2217 bool do_cond_resched, retry;
2218
2219 if (console_suspended) {
2220 up_console_sem();
2221 return;
2222 }
2223
2224 /*
2225 * Console drivers are called under logbuf_lock, so
2226 * @console_may_schedule should be cleared before; however, we may
2227 * end up dumping a lot of lines, for example, if called from
2228 * console registration path, and should invoke cond_resched()
2229 * between lines if allowable. Not doing so can cause a very long
2230 * scheduling stall on a slow console leading to RCU stall and
2231 * softlockup warnings which exacerbate the issue with more
2232 * messages practically incapacitating the system.
2233 */
2234 do_cond_resched = console_may_schedule;
2235 console_may_schedule = 0;
2236
2237again:
2238 /*
2239 * We released the console_sem lock, so we need to recheck if
2240 * cpu is online and (if not) is there at least one CON_ANYTIME
2241 * console.
2242 */
2243 if (!can_use_console()) {
2244 console_locked = 0;
2245 up_console_sem();
2246 return;
2247 }
2248
2249 /* flush buffered message fragment immediately to console */
2250 console_cont_flush(text, sizeof(text));
2251
2252 for (;;) {
2253 struct printk_log *msg;
2254 size_t ext_len = 0;
2255 size_t len;
2256 int level;
2257
2258 raw_spin_lock_irqsave(&logbuf_lock, flags);
2259 if (seen_seq != log_next_seq) {
2260 wake_klogd = true;
2261 seen_seq = log_next_seq;
2262 }
2263
2264 if (console_seq < log_first_seq) {
2265 len = sprintf(text, "** %u printk messages dropped ** ",
2266 (unsigned)(log_first_seq - console_seq));
2267
2268 /* messages are gone, move to first one */
2269 console_seq = log_first_seq;
2270 console_idx = log_first_idx;
2271 console_prev = 0;
2272 } else {
2273 len = 0;
2274 }
2275skip:
2276 if (console_seq == log_next_seq)
2277 break;
2278
2279 msg = log_from_idx(console_idx);
2280 if (msg->flags & LOG_NOCONS) {
2281 /*
2282 * Skip record we have buffered and already printed
2283 * directly to the console when we received it.
2284 */
2285 console_idx = log_next(console_idx);
2286 console_seq++;
2287 /*
2288 * We will get here again when we register a new
2289 * CON_PRINTBUFFER console. Clear the flag so we
2290 * will properly dump everything later.
2291 */
2292 msg->flags &= ~LOG_NOCONS;
2293 console_prev = msg->flags;
2294 goto skip;
2295 }
2296
2297 level = msg->level;
2298 len += msg_print_text(msg, console_prev, false,
2299 text + len, sizeof(text) - len);
2300 if (nr_ext_console_drivers) {
2301 ext_len = msg_print_ext_header(ext_text,
2302 sizeof(ext_text),
2303 msg, console_seq, console_prev);
2304 ext_len += msg_print_ext_body(ext_text + ext_len,
2305 sizeof(ext_text) - ext_len,
2306 log_dict(msg), msg->dict_len,
2307 log_text(msg), msg->text_len);
2308 }
2309 console_idx = log_next(console_idx);
2310 console_seq++;
2311 console_prev = msg->flags;
2312 raw_spin_unlock(&logbuf_lock);
2313
2314 stop_critical_timings(); /* don't trace print latency */
2315 call_console_drivers(level, ext_text, ext_len, text, len);
2316 start_critical_timings();
2317 local_irq_restore(flags);
2318
2319 if (do_cond_resched)
2320 cond_resched();
2321 }
2322 console_locked = 0;
2323
2324 /* Release the exclusive_console once it is used */
2325 if (unlikely(exclusive_console))
2326 exclusive_console = NULL;
2327
2328 raw_spin_unlock(&logbuf_lock);
2329
2330 up_console_sem();
2331
2332 /*
2333 * Someone could have filled up the buffer again, so re-check if there's
2334 * something to flush. In case we cannot trylock the console_sem again,
2335 * there's a new owner and the console_unlock() from them will do the
2336 * flush, no worries.
2337 */
2338 raw_spin_lock(&logbuf_lock);
2339 retry = console_seq != log_next_seq;
2340 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2341
2342 if (retry && console_trylock())
2343 goto again;
2344
2345 if (wake_klogd)
2346 wake_up_klogd();
2347}
2348EXPORT_SYMBOL(console_unlock);
2349
2350/**
2351 * console_conditional_schedule - yield the CPU if required
2352 *
2353 * If the console code is currently allowed to sleep, and
2354 * if this CPU should yield the CPU to another task, do
2355 * so here.
2356 *
2357 * Must be called within console_lock();.
2358 */
2359void __sched console_conditional_schedule(void)
2360{
2361 if (console_may_schedule)
2362 cond_resched();
2363}
2364EXPORT_SYMBOL(console_conditional_schedule);
2365
2366void console_unblank(void)
2367{
2368 struct console *c;
2369
2370 /*
2371 * console_unblank can no longer be called in interrupt context unless
2372 * oops_in_progress is set to 1..
2373 */
2374 if (oops_in_progress) {
2375 if (down_trylock_console_sem() != 0)
2376 return;
2377 } else
2378 console_lock();
2379
2380 console_locked = 1;
2381 console_may_schedule = 0;
2382 for_each_console(c)
2383 if ((c->flags & CON_ENABLED) && c->unblank)
2384 c->unblank();
2385 console_unlock();
2386}
2387
2388/**
2389 * console_flush_on_panic - flush console content on panic
2390 *
2391 * Immediately output all pending messages no matter what.
2392 */
2393void console_flush_on_panic(void)
2394{
2395 /*
2396 * If someone else is holding the console lock, trylock will fail
2397 * and may_schedule may be set. Ignore and proceed to unlock so
2398 * that messages are flushed out. As this can be called from any
2399 * context and we don't want to get preempted while flushing,
2400 * ensure may_schedule is cleared.
2401 */
2402 console_trylock();
2403 console_may_schedule = 0;
2404 console_unlock();
2405}
2406
2407/*
2408 * Return the console tty driver structure and its associated index
2409 */
2410struct tty_driver *console_device(int *index)
2411{
2412 struct console *c;
2413 struct tty_driver *driver = NULL;
2414
2415 console_lock();
2416 for_each_console(c) {
2417 if (!c->device)
2418 continue;
2419 driver = c->device(c, index);
2420 if (driver)
2421 break;
2422 }
2423 console_unlock();
2424 return driver;
2425}
2426
2427/*
2428 * Prevent further output on the passed console device so that (for example)
2429 * serial drivers can disable console output before suspending a port, and can
2430 * re-enable output afterwards.
2431 */
2432void console_stop(struct console *console)
2433{
2434 console_lock();
2435 console->flags &= ~CON_ENABLED;
2436 console_unlock();
2437}
2438EXPORT_SYMBOL(console_stop);
2439
2440void console_start(struct console *console)
2441{
2442 console_lock();
2443 console->flags |= CON_ENABLED;
2444 console_unlock();
2445}
2446EXPORT_SYMBOL(console_start);
2447
2448static int __read_mostly keep_bootcon;
2449
2450static int __init keep_bootcon_setup(char *str)
2451{
2452 keep_bootcon = 1;
2453 pr_info("debug: skip boot console de-registration.\n");
2454
2455 return 0;
2456}
2457
2458early_param("keep_bootcon", keep_bootcon_setup);
2459
2460/*
2461 * The console driver calls this routine during kernel initialization
2462 * to register the console printing procedure with printk() and to
2463 * print any messages that were printed by the kernel before the
2464 * console driver was initialized.
2465 *
2466 * This can happen pretty early during the boot process (because of
2467 * early_printk) - sometimes before setup_arch() completes - be careful
2468 * of what kernel features are used - they may not be initialised yet.
2469 *
2470 * There are two types of consoles - bootconsoles (early_printk) and
2471 * "real" consoles (everything which is not a bootconsole) which are
2472 * handled differently.
2473 * - Any number of bootconsoles can be registered at any time.
2474 * - As soon as a "real" console is registered, all bootconsoles
2475 * will be unregistered automatically.
2476 * - Once a "real" console is registered, any attempt to register a
2477 * bootconsoles will be rejected
2478 */
2479void register_console(struct console *newcon)
2480{
2481 int i;
2482 unsigned long flags;
2483 struct console *bcon = NULL;
2484 struct console_cmdline *c;
2485
2486 if (console_drivers)
2487 for_each_console(bcon)
2488 if (WARN(bcon == newcon,
2489 "console '%s%d' already registered\n",
2490 bcon->name, bcon->index))
2491 return;
2492
2493 /*
2494 * before we register a new CON_BOOT console, make sure we don't
2495 * already have a valid console
2496 */
2497 if (console_drivers && newcon->flags & CON_BOOT) {
2498 /* find the last or real console */
2499 for_each_console(bcon) {
2500 if (!(bcon->flags & CON_BOOT)) {
2501 pr_info("Too late to register bootconsole %s%d\n",
2502 newcon->name, newcon->index);
2503 return;
2504 }
2505 }
2506 }
2507
2508 if (console_drivers && console_drivers->flags & CON_BOOT)
2509 bcon = console_drivers;
2510
2511 if (preferred_console < 0 || bcon || !console_drivers)
2512 preferred_console = selected_console;
2513
2514 /*
2515 * See if we want to use this console driver. If we
2516 * didn't select a console we take the first one
2517 * that registers here.
2518 */
2519 if (preferred_console < 0) {
2520 if (newcon->index < 0)
2521 newcon->index = 0;
2522 if (newcon->setup == NULL ||
2523 newcon->setup(newcon, NULL) == 0) {
2524 newcon->flags |= CON_ENABLED;
2525 if (newcon->device) {
2526 newcon->flags |= CON_CONSDEV;
2527 preferred_console = 0;
2528 }
2529 }
2530 }
2531
2532 /*
2533 * See if this console matches one we selected on
2534 * the command line.
2535 */
2536 for (i = 0, c = console_cmdline;
2537 i < MAX_CMDLINECONSOLES && c->name[0];
2538 i++, c++) {
2539 if (!newcon->match ||
2540 newcon->match(newcon, c->name, c->index, c->options) != 0) {
2541 /* default matching */
2542 BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2543 if (strcmp(c->name, newcon->name) != 0)
2544 continue;
2545 if (newcon->index >= 0 &&
2546 newcon->index != c->index)
2547 continue;
2548 if (newcon->index < 0)
2549 newcon->index = c->index;
2550
2551 if (_braille_register_console(newcon, c))
2552 return;
2553
2554 if (newcon->setup &&
2555 newcon->setup(newcon, c->options) != 0)
2556 break;
2557 }
2558
2559 newcon->flags |= CON_ENABLED;
2560 if (i == selected_console) {
2561 newcon->flags |= CON_CONSDEV;
2562 preferred_console = selected_console;
2563 }
2564 break;
2565 }
2566
2567 if (!(newcon->flags & CON_ENABLED))
2568 return;
2569
2570 /*
2571 * If we have a bootconsole, and are switching to a real console,
2572 * don't print everything out again, since when the boot console, and
2573 * the real console are the same physical device, it's annoying to
2574 * see the beginning boot messages twice
2575 */
2576 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2577 newcon->flags &= ~CON_PRINTBUFFER;
2578
2579 /*
2580 * Put this console in the list - keep the
2581 * preferred driver at the head of the list.
2582 */
2583 console_lock();
2584 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2585 newcon->next = console_drivers;
2586 console_drivers = newcon;
2587 if (newcon->next)
2588 newcon->next->flags &= ~CON_CONSDEV;
2589 } else {
2590 newcon->next = console_drivers->next;
2591 console_drivers->next = newcon;
2592 }
2593
2594 if (newcon->flags & CON_EXTENDED)
2595 if (!nr_ext_console_drivers++)
2596 pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2597
2598 if (newcon->flags & CON_PRINTBUFFER) {
2599 /*
2600 * console_unlock(); will print out the buffered messages
2601 * for us.
2602 */
2603 raw_spin_lock_irqsave(&logbuf_lock, flags);
2604 console_seq = syslog_seq;
2605 console_idx = syslog_idx;
2606 console_prev = syslog_prev;
2607 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2608 /*
2609 * We're about to replay the log buffer. Only do this to the
2610 * just-registered console to avoid excessive message spam to
2611 * the already-registered consoles.
2612 */
2613 exclusive_console = newcon;
2614 }
2615 console_unlock();
2616 console_sysfs_notify();
2617
2618 /*
2619 * By unregistering the bootconsoles after we enable the real console
2620 * we get the "console xxx enabled" message on all the consoles -
2621 * boot consoles, real consoles, etc - this is to ensure that end
2622 * users know there might be something in the kernel's log buffer that
2623 * went to the bootconsole (that they do not see on the real console)
2624 */
2625 pr_info("%sconsole [%s%d] enabled\n",
2626 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2627 newcon->name, newcon->index);
2628 if (bcon &&
2629 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2630 !keep_bootcon) {
2631 /* We need to iterate through all boot consoles, to make
2632 * sure we print everything out, before we unregister them.
2633 */
2634 for_each_console(bcon)
2635 if (bcon->flags & CON_BOOT)
2636 unregister_console(bcon);
2637 }
2638}
2639EXPORT_SYMBOL(register_console);
2640
2641int unregister_console(struct console *console)
2642{
2643 struct console *a, *b;
2644 int res;
2645
2646 pr_info("%sconsole [%s%d] disabled\n",
2647 (console->flags & CON_BOOT) ? "boot" : "" ,
2648 console->name, console->index);
2649
2650 res = _braille_unregister_console(console);
2651 if (res)
2652 return res;
2653
2654 res = 1;
2655 console_lock();
2656 if (console_drivers == console) {
2657 console_drivers=console->next;
2658 res = 0;
2659 } else if (console_drivers) {
2660 for (a=console_drivers->next, b=console_drivers ;
2661 a; b=a, a=b->next) {
2662 if (a == console) {
2663 b->next = a->next;
2664 res = 0;
2665 break;
2666 }
2667 }
2668 }
2669
2670 if (!res && (console->flags & CON_EXTENDED))
2671 nr_ext_console_drivers--;
2672
2673 /*
2674 * If this isn't the last console and it has CON_CONSDEV set, we
2675 * need to set it on the next preferred console.
2676 */
2677 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2678 console_drivers->flags |= CON_CONSDEV;
2679
2680 console->flags &= ~CON_ENABLED;
2681 console_unlock();
2682 console_sysfs_notify();
2683 return res;
2684}
2685EXPORT_SYMBOL(unregister_console);
2686
2687/*
2688 * Some boot consoles access data that is in the init section and which will
2689 * be discarded after the initcalls have been run. To make sure that no code
2690 * will access this data, unregister the boot consoles in a late initcall.
2691 *
2692 * If for some reason, such as deferred probe or the driver being a loadable
2693 * module, the real console hasn't registered yet at this point, there will
2694 * be a brief interval in which no messages are logged to the console, which
2695 * makes it difficult to diagnose problems that occur during this time.
2696 *
2697 * To mitigate this problem somewhat, only unregister consoles whose memory
2698 * intersects with the init section. Note that code exists elsewhere to get
2699 * rid of the boot console as soon as the proper console shows up, so there
2700 * won't be side-effects from postponing the removal.
2701 */
2702static int __init printk_late_init(void)
2703{
2704 struct console *con;
2705
2706 for_each_console(con) {
2707 if (!keep_bootcon && con->flags & CON_BOOT) {
2708 /*
2709 * Make sure to unregister boot consoles whose data
2710 * resides in the init section before the init section
2711 * is discarded. Boot consoles whose data will stick
2712 * around will automatically be unregistered when the
2713 * proper console replaces them.
2714 */
2715 if (init_section_intersects(con, sizeof(*con)))
2716 unregister_console(con);
2717 }
2718 }
2719 hotcpu_notifier(console_cpu_notify, 0);
2720 return 0;
2721}
2722late_initcall(printk_late_init);
2723
2724#if defined CONFIG_PRINTK
2725/*
2726 * Delayed printk version, for scheduler-internal messages:
2727 */
2728#define PRINTK_PENDING_WAKEUP 0x01
2729#define PRINTK_PENDING_OUTPUT 0x02
2730
2731static DEFINE_PER_CPU(int, printk_pending);
2732
2733static void wake_up_klogd_work_func(struct irq_work *irq_work)
2734{
2735 int pending = __this_cpu_xchg(printk_pending, 0);
2736
2737 if (pending & PRINTK_PENDING_OUTPUT) {
2738 /* If trylock fails, someone else is doing the printing */
2739 if (console_trylock())
2740 console_unlock();
2741 }
2742
2743 if (pending & PRINTK_PENDING_WAKEUP)
2744 wake_up_interruptible(&log_wait);
2745}
2746
2747static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2748 .func = wake_up_klogd_work_func,
2749 .flags = IRQ_WORK_LAZY,
2750};
2751
2752void wake_up_klogd(void)
2753{
2754 preempt_disable();
2755 if (waitqueue_active(&log_wait)) {
2756 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2757 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2758 }
2759 preempt_enable();
2760}
2761
2762int printk_deferred(const char *fmt, ...)
2763{
2764 va_list args;
2765 int r;
2766
2767 preempt_disable();
2768 va_start(args, fmt);
2769 r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2770 va_end(args);
2771
2772 __this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2773 irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2774 preempt_enable();
2775
2776 return r;
2777}
2778
2779/*
2780 * printk rate limiting, lifted from the networking subsystem.
2781 *
2782 * This enforces a rate limit: not more than 10 kernel messages
2783 * every 5s to make a denial-of-service attack impossible.
2784 */
2785DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2786
2787int __printk_ratelimit(const char *func)
2788{
2789 return ___ratelimit(&printk_ratelimit_state, func);
2790}
2791EXPORT_SYMBOL(__printk_ratelimit);
2792
2793/**
2794 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2795 * @caller_jiffies: pointer to caller's state
2796 * @interval_msecs: minimum interval between prints
2797 *
2798 * printk_timed_ratelimit() returns true if more than @interval_msecs
2799 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2800 * returned true.
2801 */
2802bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2803 unsigned int interval_msecs)
2804{
2805 unsigned long elapsed = jiffies - *caller_jiffies;
2806
2807 if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2808 return false;
2809
2810 *caller_jiffies = jiffies;
2811 return true;
2812}
2813EXPORT_SYMBOL(printk_timed_ratelimit);
2814
2815static DEFINE_SPINLOCK(dump_list_lock);
2816static LIST_HEAD(dump_list);
2817
2818/**
2819 * kmsg_dump_register - register a kernel log dumper.
2820 * @dumper: pointer to the kmsg_dumper structure
2821 *
2822 * Adds a kernel log dumper to the system. The dump callback in the
2823 * structure will be called when the kernel oopses or panics and must be
2824 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2825 */
2826int kmsg_dump_register(struct kmsg_dumper *dumper)
2827{
2828 unsigned long flags;
2829 int err = -EBUSY;
2830
2831 /* The dump callback needs to be set */
2832 if (!dumper->dump)
2833 return -EINVAL;
2834
2835 spin_lock_irqsave(&dump_list_lock, flags);
2836 /* Don't allow registering multiple times */
2837 if (!dumper->registered) {
2838 dumper->registered = 1;
2839 list_add_tail_rcu(&dumper->list, &dump_list);
2840 err = 0;
2841 }
2842 spin_unlock_irqrestore(&dump_list_lock, flags);
2843
2844 return err;
2845}
2846EXPORT_SYMBOL_GPL(kmsg_dump_register);
2847
2848/**
2849 * kmsg_dump_unregister - unregister a kmsg dumper.
2850 * @dumper: pointer to the kmsg_dumper structure
2851 *
2852 * Removes a dump device from the system. Returns zero on success and
2853 * %-EINVAL otherwise.
2854 */
2855int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2856{
2857 unsigned long flags;
2858 int err = -EINVAL;
2859
2860 spin_lock_irqsave(&dump_list_lock, flags);
2861 if (dumper->registered) {
2862 dumper->registered = 0;
2863 list_del_rcu(&dumper->list);
2864 err = 0;
2865 }
2866 spin_unlock_irqrestore(&dump_list_lock, flags);
2867 synchronize_rcu();
2868
2869 return err;
2870}
2871EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2872
2873static bool always_kmsg_dump;
2874module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2875
2876/**
2877 * kmsg_dump - dump kernel log to kernel message dumpers.
2878 * @reason: the reason (oops, panic etc) for dumping
2879 *
2880 * Call each of the registered dumper's dump() callback, which can
2881 * retrieve the kmsg records with kmsg_dump_get_line() or
2882 * kmsg_dump_get_buffer().
2883 */
2884void kmsg_dump(enum kmsg_dump_reason reason)
2885{
2886 struct kmsg_dumper *dumper;
2887 unsigned long flags;
2888
2889 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2890 return;
2891
2892 rcu_read_lock();
2893 list_for_each_entry_rcu(dumper, &dump_list, list) {
2894 if (dumper->max_reason && reason > dumper->max_reason)
2895 continue;
2896
2897 /* initialize iterator with data about the stored records */
2898 dumper->active = true;
2899
2900 raw_spin_lock_irqsave(&logbuf_lock, flags);
2901 dumper->cur_seq = clear_seq;
2902 dumper->cur_idx = clear_idx;
2903 dumper->next_seq = log_next_seq;
2904 dumper->next_idx = log_next_idx;
2905 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2906
2907 /* invoke dumper which will iterate over records */
2908 dumper->dump(dumper, reason);
2909
2910 /* reset iterator */
2911 dumper->active = false;
2912 }
2913 rcu_read_unlock();
2914}
2915
2916/**
2917 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2918 * @dumper: registered kmsg dumper
2919 * @syslog: include the "<4>" prefixes
2920 * @line: buffer to copy the line to
2921 * @size: maximum size of the buffer
2922 * @len: length of line placed into buffer
2923 *
2924 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2925 * record, and copy one record into the provided buffer.
2926 *
2927 * Consecutive calls will return the next available record moving
2928 * towards the end of the buffer with the youngest messages.
2929 *
2930 * A return value of FALSE indicates that there are no more records to
2931 * read.
2932 *
2933 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2934 */
2935bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2936 char *line, size_t size, size_t *len)
2937{
2938 struct printk_log *msg;
2939 size_t l = 0;
2940 bool ret = false;
2941
2942 if (!dumper->active)
2943 goto out;
2944
2945 if (dumper->cur_seq < log_first_seq) {
2946 /* messages are gone, move to first available one */
2947 dumper->cur_seq = log_first_seq;
2948 dumper->cur_idx = log_first_idx;
2949 }
2950
2951 /* last entry */
2952 if (dumper->cur_seq >= log_next_seq)
2953 goto out;
2954
2955 msg = log_from_idx(dumper->cur_idx);
2956 l = msg_print_text(msg, 0, syslog, line, size);
2957
2958 dumper->cur_idx = log_next(dumper->cur_idx);
2959 dumper->cur_seq++;
2960 ret = true;
2961out:
2962 if (len)
2963 *len = l;
2964 return ret;
2965}
2966
2967/**
2968 * kmsg_dump_get_line - retrieve one kmsg log line
2969 * @dumper: registered kmsg dumper
2970 * @syslog: include the "<4>" prefixes
2971 * @line: buffer to copy the line to
2972 * @size: maximum size of the buffer
2973 * @len: length of line placed into buffer
2974 *
2975 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2976 * record, and copy one record into the provided buffer.
2977 *
2978 * Consecutive calls will return the next available record moving
2979 * towards the end of the buffer with the youngest messages.
2980 *
2981 * A return value of FALSE indicates that there are no more records to
2982 * read.
2983 */
2984bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2985 char *line, size_t size, size_t *len)
2986{
2987 unsigned long flags;
2988 bool ret;
2989
2990 raw_spin_lock_irqsave(&logbuf_lock, flags);
2991 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2992 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2993
2994 return ret;
2995}
2996EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2997
2998/**
2999 * kmsg_dump_get_buffer - copy kmsg log lines
3000 * @dumper: registered kmsg dumper
3001 * @syslog: include the "<4>" prefixes
3002 * @buf: buffer to copy the line to
3003 * @size: maximum size of the buffer
3004 * @len: length of line placed into buffer
3005 *
3006 * Start at the end of the kmsg buffer and fill the provided buffer
3007 * with as many of the the *youngest* kmsg records that fit into it.
3008 * If the buffer is large enough, all available kmsg records will be
3009 * copied with a single call.
3010 *
3011 * Consecutive calls will fill the buffer with the next block of
3012 * available older records, not including the earlier retrieved ones.
3013 *
3014 * A return value of FALSE indicates that there are no more records to
3015 * read.
3016 */
3017bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3018 char *buf, size_t size, size_t *len)
3019{
3020 unsigned long flags;
3021 u64 seq;
3022 u32 idx;
3023 u64 next_seq;
3024 u32 next_idx;
3025 enum log_flags prev;
3026 size_t l = 0;
3027 bool ret = false;
3028
3029 if (!dumper->active)
3030 goto out;
3031
3032 raw_spin_lock_irqsave(&logbuf_lock, flags);
3033 if (dumper->cur_seq < log_first_seq) {
3034 /* messages are gone, move to first available one */
3035 dumper->cur_seq = log_first_seq;
3036 dumper->cur_idx = log_first_idx;
3037 }
3038
3039 /* last entry */
3040 if (dumper->cur_seq >= dumper->next_seq) {
3041 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3042 goto out;
3043 }
3044
3045 /* calculate length of entire buffer */
3046 seq = dumper->cur_seq;
3047 idx = dumper->cur_idx;
3048 prev = 0;
3049 while (seq < dumper->next_seq) {
3050 struct printk_log *msg = log_from_idx(idx);
3051
3052 l += msg_print_text(msg, prev, true, NULL, 0);
3053 idx = log_next(idx);
3054 seq++;
3055 prev = msg->flags;
3056 }
3057
3058 /* move first record forward until length fits into the buffer */
3059 seq = dumper->cur_seq;
3060 idx = dumper->cur_idx;
3061 prev = 0;
3062 while (l > size && seq < dumper->next_seq) {
3063 struct printk_log *msg = log_from_idx(idx);
3064
3065 l -= msg_print_text(msg, prev, true, NULL, 0);
3066 idx = log_next(idx);
3067 seq++;
3068 prev = msg->flags;
3069 }
3070
3071 /* last message in next interation */
3072 next_seq = seq;
3073 next_idx = idx;
3074
3075 l = 0;
3076 while (seq < dumper->next_seq) {
3077 struct printk_log *msg = log_from_idx(idx);
3078
3079 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
3080 idx = log_next(idx);
3081 seq++;
3082 prev = msg->flags;
3083 }
3084
3085 dumper->next_seq = next_seq;
3086 dumper->next_idx = next_idx;
3087 ret = true;
3088 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3089out:
3090 if (len)
3091 *len = l;
3092 return ret;
3093}
3094EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3095
3096/**
3097 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3098 * @dumper: registered kmsg dumper
3099 *
3100 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3101 * kmsg_dump_get_buffer() can be called again and used multiple
3102 * times within the same dumper.dump() callback.
3103 *
3104 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3105 */
3106void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3107{
3108 dumper->cur_seq = clear_seq;
3109 dumper->cur_idx = clear_idx;
3110 dumper->next_seq = log_next_seq;
3111 dumper->next_idx = log_next_idx;
3112}
3113
3114/**
3115 * kmsg_dump_rewind - reset the interator
3116 * @dumper: registered kmsg dumper
3117 *
3118 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3119 * kmsg_dump_get_buffer() can be called again and used multiple
3120 * times within the same dumper.dump() callback.
3121 */
3122void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3123{
3124 unsigned long flags;
3125
3126 raw_spin_lock_irqsave(&logbuf_lock, flags);
3127 kmsg_dump_rewind_nolock(dumper);
3128 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3129}
3130EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3131
3132static char dump_stack_arch_desc_str[128];
3133
3134/**
3135 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3136 * @fmt: printf-style format string
3137 * @...: arguments for the format string
3138 *
3139 * The configured string will be printed right after utsname during task
3140 * dumps. Usually used to add arch-specific system identifiers. If an
3141 * arch wants to make use of such an ID string, it should initialize this
3142 * as soon as possible during boot.
3143 */
3144void __init dump_stack_set_arch_desc(const char *fmt, ...)
3145{
3146 va_list args;
3147
3148 va_start(args, fmt);
3149 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3150 fmt, args);
3151 va_end(args);
3152}
3153
3154/**
3155 * dump_stack_print_info - print generic debug info for dump_stack()
3156 * @log_lvl: log level
3157 *
3158 * Arch-specific dump_stack() implementations can use this function to
3159 * print out the same debug information as the generic dump_stack().
3160 */
3161void dump_stack_print_info(const char *log_lvl)
3162{
3163 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3164 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3165 print_tainted(), init_utsname()->release,
3166 (int)strcspn(init_utsname()->version, " "),
3167 init_utsname()->version);
3168
3169 if (dump_stack_arch_desc_str[0] != '\0')
3170 printk("%sHardware name: %s\n",
3171 log_lvl, dump_stack_arch_desc_str);
3172
3173 print_worker_info(log_lvl, current);
3174}
3175
3176/**
3177 * show_regs_print_info - print generic debug info for show_regs()
3178 * @log_lvl: log level
3179 *
3180 * show_regs() implementations can use this function to print out generic
3181 * debug information.
3182 */
3183void show_regs_print_info(const char *log_lvl)
3184{
3185 dump_stack_print_info(log_lvl);
3186
3187 printk("%stask: %p ti: %p task.ti: %p\n",
3188 log_lvl, current, current_thread_info(),
3189 task_thread_info(current));
3190}
3191
3192#endif