Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[linux-2.6.git] / kernel / printk.c
blob8212c1aef125f2d4290cbe0e55564d7e9a5a57de
1 /*
2 * linux/kernel/printk.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
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
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/aio.h>
36 #include <linux/syscalls.h>
37 #include <linux/kexec.h>
38 #include <linux/kdb.h>
39 #include <linux/ratelimit.h>
40 #include <linux/kmsg_dump.h>
41 #include <linux/syslog.h>
42 #include <linux/cpu.h>
43 #include <linux/notifier.h>
44 #include <linux/rculist.h>
45 #include <linux/poll.h>
46 #include <linux/irq_work.h>
47 #include <linux/utsname.h>
49 #include <asm/uaccess.h>
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/printk.h>
54 /* printk's without a loglevel use this.. */
55 #define DEFAULT_MESSAGE_LOGLEVEL CONFIG_DEFAULT_MESSAGE_LOGLEVEL
57 /* We show everything that is MORE important than this.. */
58 #define MINIMUM_CONSOLE_LOGLEVEL 1 /* Minimum loglevel we let people use */
59 #define DEFAULT_CONSOLE_LOGLEVEL 7 /* anything MORE serious than KERN_DEBUG */
61 int console_printk[4] = {
62 DEFAULT_CONSOLE_LOGLEVEL, /* console_loglevel */
63 DEFAULT_MESSAGE_LOGLEVEL, /* default_message_loglevel */
64 MINIMUM_CONSOLE_LOGLEVEL, /* minimum_console_loglevel */
65 DEFAULT_CONSOLE_LOGLEVEL, /* default_console_loglevel */
69 * Low level drivers may need that to know if they can schedule in
70 * their unblank() callback or not. So let's export it.
72 int oops_in_progress;
73 EXPORT_SYMBOL(oops_in_progress);
76 * console_sem protects the console_drivers list, and also
77 * provides serialisation for access to the entire console
78 * driver system.
80 static DEFINE_SEMAPHORE(console_sem);
81 struct console *console_drivers;
82 EXPORT_SYMBOL_GPL(console_drivers);
84 #ifdef CONFIG_LOCKDEP
85 static struct lockdep_map console_lock_dep_map = {
86 .name = "console_lock"
88 #endif
91 * This is used for debugging the mess that is the VT code by
92 * keeping track if we have the console semaphore held. It's
93 * definitely not the perfect debug tool (we don't know if _WE_
94 * hold it are racing, but it helps tracking those weird code
95 * path in the console code where we end up in places I want
96 * locked without the console sempahore held
98 static int console_locked, console_suspended;
101 * If exclusive_console is non-NULL then only this console is to be printed to.
103 static struct console *exclusive_console;
106 * Array of consoles built from command line options (console=)
108 struct console_cmdline
110 char name[8]; /* Name of the driver */
111 int index; /* Minor dev. to use */
112 char *options; /* Options for the driver */
113 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
114 char *brl_options; /* Options for braille driver */
115 #endif
118 #define MAX_CMDLINECONSOLES 8
120 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
121 static int selected_console = -1;
122 static int preferred_console = -1;
123 int console_set_on_cmdline;
124 EXPORT_SYMBOL(console_set_on_cmdline);
126 /* Flag: console code may call schedule() */
127 static int console_may_schedule;
130 * The printk log buffer consists of a chain of concatenated variable
131 * length records. Every record starts with a record header, containing
132 * the overall length of the record.
134 * The heads to the first and last entry in the buffer, as well as the
135 * sequence numbers of these both entries are maintained when messages
136 * are stored..
138 * If the heads indicate available messages, the length in the header
139 * tells the start next message. A length == 0 for the next message
140 * indicates a wrap-around to the beginning of the buffer.
142 * Every record carries the monotonic timestamp in microseconds, as well as
143 * the standard userspace syslog level and syslog facility. The usual
144 * kernel messages use LOG_KERN; userspace-injected messages always carry
145 * a matching syslog facility, by default LOG_USER. The origin of every
146 * message can be reliably determined that way.
148 * The human readable log message directly follows the message header. The
149 * length of the message text is stored in the header, the stored message
150 * is not terminated.
152 * Optionally, a message can carry a dictionary of properties (key/value pairs),
153 * to provide userspace with a machine-readable message context.
155 * Examples for well-defined, commonly used property names are:
156 * DEVICE=b12:8 device identifier
157 * b12:8 block dev_t
158 * c127:3 char dev_t
159 * n8 netdev ifindex
160 * +sound:card0 subsystem:devname
161 * SUBSYSTEM=pci driver-core subsystem name
163 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
164 * follows directly after a '=' character. Every property is terminated by
165 * a '\0' character. The last property is not terminated.
167 * Example of a message structure:
168 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
169 * 0008 34 00 record is 52 bytes long
170 * 000a 0b 00 text is 11 bytes long
171 * 000c 1f 00 dictionary is 23 bytes long
172 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
173 * 0010 69 74 27 73 20 61 20 6c "it's a l"
174 * 69 6e 65 "ine"
175 * 001b 44 45 56 49 43 "DEVIC"
176 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
177 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
178 * 67 "g"
179 * 0032 00 00 00 padding to next message header
181 * The 'struct log' buffer header must never be directly exported to
182 * userspace, it is a kernel-private implementation detail that might
183 * need to be changed in the future, when the requirements change.
185 * /dev/kmsg exports the structured data in the following line format:
186 * "level,sequnum,timestamp;<message text>\n"
188 * The optional key/value pairs are attached as continuation lines starting
189 * with a space character and terminated by a newline. All possible
190 * non-prinatable characters are escaped in the "\xff" notation.
192 * Users of the export format should ignore possible additional values
193 * separated by ',', and find the message after the ';' character.
196 enum log_flags {
197 LOG_NOCONS = 1, /* already flushed, do not print to console */
198 LOG_NEWLINE = 2, /* text ended with a newline */
199 LOG_PREFIX = 4, /* text started with a prefix */
200 LOG_CONT = 8, /* text is a fragment of a continuation line */
203 struct log {
204 u64 ts_nsec; /* timestamp in nanoseconds */
205 u16 len; /* length of entire record */
206 u16 text_len; /* length of text buffer */
207 u16 dict_len; /* length of dictionary buffer */
208 u8 facility; /* syslog facility */
209 u8 flags:5; /* internal record flags */
210 u8 level:3; /* syslog level */
214 * The logbuf_lock protects kmsg buffer, indices, counters. It is also
215 * used in interesting ways to provide interlocking in console_unlock();
217 static DEFINE_RAW_SPINLOCK(logbuf_lock);
219 #ifdef CONFIG_PRINTK
220 DECLARE_WAIT_QUEUE_HEAD(log_wait);
221 /* the next printk record to read by syslog(READ) or /proc/kmsg */
222 static u64 syslog_seq;
223 static u32 syslog_idx;
224 static enum log_flags syslog_prev;
225 static size_t syslog_partial;
227 /* index and sequence number of the first record stored in the buffer */
228 static u64 log_first_seq;
229 static u32 log_first_idx;
231 /* index and sequence number of the next record to store in the buffer */
232 static u64 log_next_seq;
233 static u32 log_next_idx;
235 /* the next printk record to write to the console */
236 static u64 console_seq;
237 static u32 console_idx;
238 static enum log_flags console_prev;
240 /* the next printk record to read after the last 'clear' command */
241 static u64 clear_seq;
242 static u32 clear_idx;
244 #define PREFIX_MAX 32
245 #define LOG_LINE_MAX 1024 - PREFIX_MAX
247 /* record buffer */
248 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
249 #define LOG_ALIGN 4
250 #else
251 #define LOG_ALIGN __alignof__(struct log)
252 #endif
253 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
254 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
255 static char *log_buf = __log_buf;
256 static u32 log_buf_len = __LOG_BUF_LEN;
258 /* cpu currently holding logbuf_lock */
259 static volatile unsigned int logbuf_cpu = UINT_MAX;
261 /* human readable text of the record */
262 static char *log_text(const struct log *msg)
264 return (char *)msg + sizeof(struct log);
267 /* optional key/value pair dictionary attached to the record */
268 static char *log_dict(const struct log *msg)
270 return (char *)msg + sizeof(struct log) + msg->text_len;
273 /* get record by index; idx must point to valid msg */
274 static struct log *log_from_idx(u32 idx)
276 struct log *msg = (struct log *)(log_buf + idx);
279 * A length == 0 record is the end of buffer marker. Wrap around and
280 * read the message at the start of the buffer.
282 if (!msg->len)
283 return (struct log *)log_buf;
284 return msg;
287 /* get next record; idx must point to valid msg */
288 static u32 log_next(u32 idx)
290 struct log *msg = (struct log *)(log_buf + idx);
292 /* length == 0 indicates the end of the buffer; wrap */
294 * A length == 0 record is the end of buffer marker. Wrap around and
295 * read the message at the start of the buffer as *this* one, and
296 * return the one after that.
298 if (!msg->len) {
299 msg = (struct log *)log_buf;
300 return msg->len;
302 return idx + msg->len;
305 /* insert record into the buffer, discard old ones, update heads */
306 static void log_store(int facility, int level,
307 enum log_flags flags, u64 ts_nsec,
308 const char *dict, u16 dict_len,
309 const char *text, u16 text_len)
311 struct log *msg;
312 u32 size, pad_len;
314 /* number of '\0' padding bytes to next message */
315 size = sizeof(struct log) + text_len + dict_len;
316 pad_len = (-size) & (LOG_ALIGN - 1);
317 size += pad_len;
319 while (log_first_seq < log_next_seq) {
320 u32 free;
322 if (log_next_idx > log_first_idx)
323 free = max(log_buf_len - log_next_idx, log_first_idx);
324 else
325 free = log_first_idx - log_next_idx;
327 if (free > size + sizeof(struct log))
328 break;
330 /* drop old messages until we have enough contiuous space */
331 log_first_idx = log_next(log_first_idx);
332 log_first_seq++;
335 if (log_next_idx + size + sizeof(struct log) >= log_buf_len) {
337 * This message + an additional empty header does not fit
338 * at the end of the buffer. Add an empty header with len == 0
339 * to signify a wrap around.
341 memset(log_buf + log_next_idx, 0, sizeof(struct log));
342 log_next_idx = 0;
345 /* fill message */
346 msg = (struct log *)(log_buf + log_next_idx);
347 memcpy(log_text(msg), text, text_len);
348 msg->text_len = text_len;
349 memcpy(log_dict(msg), dict, dict_len);
350 msg->dict_len = dict_len;
351 msg->facility = facility;
352 msg->level = level & 7;
353 msg->flags = flags & 0x1f;
354 if (ts_nsec > 0)
355 msg->ts_nsec = ts_nsec;
356 else
357 msg->ts_nsec = local_clock();
358 memset(log_dict(msg) + dict_len, 0, pad_len);
359 msg->len = sizeof(struct log) + text_len + dict_len + pad_len;
361 /* insert message */
362 log_next_idx += msg->len;
363 log_next_seq++;
366 #ifdef CONFIG_SECURITY_DMESG_RESTRICT
367 int dmesg_restrict = 1;
368 #else
369 int dmesg_restrict;
370 #endif
372 static int syslog_action_restricted(int type)
374 if (dmesg_restrict)
375 return 1;
377 * Unless restricted, we allow "read all" and "get buffer size"
378 * for everybody.
380 return type != SYSLOG_ACTION_READ_ALL &&
381 type != SYSLOG_ACTION_SIZE_BUFFER;
384 static int check_syslog_permissions(int type, bool from_file)
387 * If this is from /proc/kmsg and we've already opened it, then we've
388 * already done the capabilities checks at open time.
390 if (from_file && type != SYSLOG_ACTION_OPEN)
391 return 0;
393 if (syslog_action_restricted(type)) {
394 if (capable(CAP_SYSLOG))
395 return 0;
397 * For historical reasons, accept CAP_SYS_ADMIN too, with
398 * a warning.
400 if (capable(CAP_SYS_ADMIN)) {
401 pr_warn_once("%s (%d): Attempt to access syslog with "
402 "CAP_SYS_ADMIN but no CAP_SYSLOG "
403 "(deprecated).\n",
404 current->comm, task_pid_nr(current));
405 return 0;
407 return -EPERM;
409 return security_syslog(type);
413 /* /dev/kmsg - userspace message inject/listen interface */
414 struct devkmsg_user {
415 u64 seq;
416 u32 idx;
417 enum log_flags prev;
418 struct mutex lock;
419 char buf[8192];
422 static ssize_t devkmsg_writev(struct kiocb *iocb, const struct iovec *iv,
423 unsigned long count, loff_t pos)
425 char *buf, *line;
426 int i;
427 int level = default_message_loglevel;
428 int facility = 1; /* LOG_USER */
429 size_t len = iov_length(iv, count);
430 ssize_t ret = len;
432 if (len > LOG_LINE_MAX)
433 return -EINVAL;
434 buf = kmalloc(len+1, GFP_KERNEL);
435 if (buf == NULL)
436 return -ENOMEM;
438 line = buf;
439 for (i = 0; i < count; i++) {
440 if (copy_from_user(line, iv[i].iov_base, iv[i].iov_len)) {
441 ret = -EFAULT;
442 goto out;
444 line += iv[i].iov_len;
448 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
449 * the decimal value represents 32bit, the lower 3 bit are the log
450 * level, the rest are the log facility.
452 * If no prefix or no userspace facility is specified, we
453 * enforce LOG_USER, to be able to reliably distinguish
454 * kernel-generated messages from userspace-injected ones.
456 line = buf;
457 if (line[0] == '<') {
458 char *endp = NULL;
460 i = simple_strtoul(line+1, &endp, 10);
461 if (endp && endp[0] == '>') {
462 level = i & 7;
463 if (i >> 3)
464 facility = i >> 3;
465 endp++;
466 len -= endp - line;
467 line = endp;
470 line[len] = '\0';
472 printk_emit(facility, level, NULL, 0, "%s", line);
473 out:
474 kfree(buf);
475 return ret;
478 static ssize_t devkmsg_read(struct file *file, char __user *buf,
479 size_t count, loff_t *ppos)
481 struct devkmsg_user *user = file->private_data;
482 struct log *msg;
483 u64 ts_usec;
484 size_t i;
485 char cont = '-';
486 size_t len;
487 ssize_t ret;
489 if (!user)
490 return -EBADF;
492 ret = mutex_lock_interruptible(&user->lock);
493 if (ret)
494 return ret;
495 raw_spin_lock_irq(&logbuf_lock);
496 while (user->seq == log_next_seq) {
497 if (file->f_flags & O_NONBLOCK) {
498 ret = -EAGAIN;
499 raw_spin_unlock_irq(&logbuf_lock);
500 goto out;
503 raw_spin_unlock_irq(&logbuf_lock);
504 ret = wait_event_interruptible(log_wait,
505 user->seq != log_next_seq);
506 if (ret)
507 goto out;
508 raw_spin_lock_irq(&logbuf_lock);
511 if (user->seq < log_first_seq) {
512 /* our last seen message is gone, return error and reset */
513 user->idx = log_first_idx;
514 user->seq = log_first_seq;
515 ret = -EPIPE;
516 raw_spin_unlock_irq(&logbuf_lock);
517 goto out;
520 msg = log_from_idx(user->idx);
521 ts_usec = msg->ts_nsec;
522 do_div(ts_usec, 1000);
525 * If we couldn't merge continuation line fragments during the print,
526 * export the stored flags to allow an optional external merge of the
527 * records. Merging the records isn't always neccessarily correct, like
528 * when we hit a race during printing. In most cases though, it produces
529 * better readable output. 'c' in the record flags mark the first
530 * fragment of a line, '+' the following.
532 if (msg->flags & LOG_CONT && !(user->prev & LOG_CONT))
533 cont = 'c';
534 else if ((msg->flags & LOG_CONT) ||
535 ((user->prev & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
536 cont = '+';
538 len = sprintf(user->buf, "%u,%llu,%llu,%c;",
539 (msg->facility << 3) | msg->level,
540 user->seq, ts_usec, cont);
541 user->prev = msg->flags;
543 /* escape non-printable characters */
544 for (i = 0; i < msg->text_len; i++) {
545 unsigned char c = log_text(msg)[i];
547 if (c < ' ' || c >= 127 || c == '\\')
548 len += sprintf(user->buf + len, "\\x%02x", c);
549 else
550 user->buf[len++] = c;
552 user->buf[len++] = '\n';
554 if (msg->dict_len) {
555 bool line = true;
557 for (i = 0; i < msg->dict_len; i++) {
558 unsigned char c = log_dict(msg)[i];
560 if (line) {
561 user->buf[len++] = ' ';
562 line = false;
565 if (c == '\0') {
566 user->buf[len++] = '\n';
567 line = true;
568 continue;
571 if (c < ' ' || c >= 127 || c == '\\') {
572 len += sprintf(user->buf + len, "\\x%02x", c);
573 continue;
576 user->buf[len++] = c;
578 user->buf[len++] = '\n';
581 user->idx = log_next(user->idx);
582 user->seq++;
583 raw_spin_unlock_irq(&logbuf_lock);
585 if (len > count) {
586 ret = -EINVAL;
587 goto out;
590 if (copy_to_user(buf, user->buf, len)) {
591 ret = -EFAULT;
592 goto out;
594 ret = len;
595 out:
596 mutex_unlock(&user->lock);
597 return ret;
600 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
602 struct devkmsg_user *user = file->private_data;
603 loff_t ret = 0;
605 if (!user)
606 return -EBADF;
607 if (offset)
608 return -ESPIPE;
610 raw_spin_lock_irq(&logbuf_lock);
611 switch (whence) {
612 case SEEK_SET:
613 /* the first record */
614 user->idx = log_first_idx;
615 user->seq = log_first_seq;
616 break;
617 case SEEK_DATA:
619 * The first record after the last SYSLOG_ACTION_CLEAR,
620 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
621 * changes no global state, and does not clear anything.
623 user->idx = clear_idx;
624 user->seq = clear_seq;
625 break;
626 case SEEK_END:
627 /* after the last record */
628 user->idx = log_next_idx;
629 user->seq = log_next_seq;
630 break;
631 default:
632 ret = -EINVAL;
634 raw_spin_unlock_irq(&logbuf_lock);
635 return ret;
638 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
640 struct devkmsg_user *user = file->private_data;
641 int ret = 0;
643 if (!user)
644 return POLLERR|POLLNVAL;
646 poll_wait(file, &log_wait, wait);
648 raw_spin_lock_irq(&logbuf_lock);
649 if (user->seq < log_next_seq) {
650 /* return error when data has vanished underneath us */
651 if (user->seq < log_first_seq)
652 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
653 else
654 ret = POLLIN|POLLRDNORM;
656 raw_spin_unlock_irq(&logbuf_lock);
658 return ret;
661 static int devkmsg_open(struct inode *inode, struct file *file)
663 struct devkmsg_user *user;
664 int err;
666 /* write-only does not need any file context */
667 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
668 return 0;
670 err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
671 SYSLOG_FROM_READER);
672 if (err)
673 return err;
675 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
676 if (!user)
677 return -ENOMEM;
679 mutex_init(&user->lock);
681 raw_spin_lock_irq(&logbuf_lock);
682 user->idx = log_first_idx;
683 user->seq = log_first_seq;
684 raw_spin_unlock_irq(&logbuf_lock);
686 file->private_data = user;
687 return 0;
690 static int devkmsg_release(struct inode *inode, struct file *file)
692 struct devkmsg_user *user = file->private_data;
694 if (!user)
695 return 0;
697 mutex_destroy(&user->lock);
698 kfree(user);
699 return 0;
702 const struct file_operations kmsg_fops = {
703 .open = devkmsg_open,
704 .read = devkmsg_read,
705 .aio_write = devkmsg_writev,
706 .llseek = devkmsg_llseek,
707 .poll = devkmsg_poll,
708 .release = devkmsg_release,
711 #ifdef CONFIG_KEXEC
713 * This appends the listed symbols to /proc/vmcoreinfo
715 * /proc/vmcoreinfo is used by various utiilties, like crash and makedumpfile to
716 * obtain access to symbols that are otherwise very difficult to locate. These
717 * symbols are specifically used so that utilities can access and extract the
718 * dmesg log from a vmcore file after a crash.
720 void log_buf_kexec_setup(void)
722 VMCOREINFO_SYMBOL(log_buf);
723 VMCOREINFO_SYMBOL(log_buf_len);
724 VMCOREINFO_SYMBOL(log_first_idx);
725 VMCOREINFO_SYMBOL(log_next_idx);
727 * Export struct log size and field offsets. User space tools can
728 * parse it and detect any changes to structure down the line.
730 VMCOREINFO_STRUCT_SIZE(log);
731 VMCOREINFO_OFFSET(log, ts_nsec);
732 VMCOREINFO_OFFSET(log, len);
733 VMCOREINFO_OFFSET(log, text_len);
734 VMCOREINFO_OFFSET(log, dict_len);
736 #endif
738 /* requested log_buf_len from kernel cmdline */
739 static unsigned long __initdata new_log_buf_len;
741 /* save requested log_buf_len since it's too early to process it */
742 static int __init log_buf_len_setup(char *str)
744 unsigned size = memparse(str, &str);
746 if (size)
747 size = roundup_pow_of_two(size);
748 if (size > log_buf_len)
749 new_log_buf_len = size;
751 return 0;
753 early_param("log_buf_len", log_buf_len_setup);
755 void __init setup_log_buf(int early)
757 unsigned long flags;
758 char *new_log_buf;
759 int free;
761 if (!new_log_buf_len)
762 return;
764 if (early) {
765 unsigned long mem;
767 mem = memblock_alloc(new_log_buf_len, PAGE_SIZE);
768 if (!mem)
769 return;
770 new_log_buf = __va(mem);
771 } else {
772 new_log_buf = alloc_bootmem_nopanic(new_log_buf_len);
775 if (unlikely(!new_log_buf)) {
776 pr_err("log_buf_len: %ld bytes not available\n",
777 new_log_buf_len);
778 return;
781 raw_spin_lock_irqsave(&logbuf_lock, flags);
782 log_buf_len = new_log_buf_len;
783 log_buf = new_log_buf;
784 new_log_buf_len = 0;
785 free = __LOG_BUF_LEN - log_next_idx;
786 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
787 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
789 pr_info("log_buf_len: %d\n", log_buf_len);
790 pr_info("early log buf free: %d(%d%%)\n",
791 free, (free * 100) / __LOG_BUF_LEN);
794 static bool __read_mostly ignore_loglevel;
796 static int __init ignore_loglevel_setup(char *str)
798 ignore_loglevel = 1;
799 printk(KERN_INFO "debug: ignoring loglevel setting.\n");
801 return 0;
804 early_param("ignore_loglevel", ignore_loglevel_setup);
805 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
806 MODULE_PARM_DESC(ignore_loglevel, "ignore loglevel setting, to"
807 "print all kernel messages to the console.");
809 #ifdef CONFIG_BOOT_PRINTK_DELAY
811 static int boot_delay; /* msecs delay after each printk during bootup */
812 static unsigned long long loops_per_msec; /* based on boot_delay */
814 static int __init boot_delay_setup(char *str)
816 unsigned long lpj;
818 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
819 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
821 get_option(&str, &boot_delay);
822 if (boot_delay > 10 * 1000)
823 boot_delay = 0;
825 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
826 "HZ: %d, loops_per_msec: %llu\n",
827 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
828 return 1;
830 __setup("boot_delay=", boot_delay_setup);
832 static void boot_delay_msec(int level)
834 unsigned long long k;
835 unsigned long timeout;
837 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
838 || (level >= console_loglevel && !ignore_loglevel)) {
839 return;
842 k = (unsigned long long)loops_per_msec * boot_delay;
844 timeout = jiffies + msecs_to_jiffies(boot_delay);
845 while (k) {
846 k--;
847 cpu_relax();
849 * use (volatile) jiffies to prevent
850 * compiler reduction; loop termination via jiffies
851 * is secondary and may or may not happen.
853 if (time_after(jiffies, timeout))
854 break;
855 touch_nmi_watchdog();
858 #else
859 static inline void boot_delay_msec(int level)
862 #endif
864 #if defined(CONFIG_PRINTK_TIME)
865 static bool printk_time = 1;
866 #else
867 static bool printk_time;
868 #endif
869 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
871 static size_t print_time(u64 ts, char *buf)
873 unsigned long rem_nsec;
875 if (!printk_time)
876 return 0;
878 rem_nsec = do_div(ts, 1000000000);
880 if (!buf)
881 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
883 return sprintf(buf, "[%5lu.%06lu] ",
884 (unsigned long)ts, rem_nsec / 1000);
887 static size_t print_prefix(const struct log *msg, bool syslog, char *buf)
889 size_t len = 0;
890 unsigned int prefix = (msg->facility << 3) | msg->level;
892 if (syslog) {
893 if (buf) {
894 len += sprintf(buf, "<%u>", prefix);
895 } else {
896 len += 3;
897 if (prefix > 999)
898 len += 3;
899 else if (prefix > 99)
900 len += 2;
901 else if (prefix > 9)
902 len++;
906 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
907 return len;
910 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
911 bool syslog, char *buf, size_t size)
913 const char *text = log_text(msg);
914 size_t text_size = msg->text_len;
915 bool prefix = true;
916 bool newline = true;
917 size_t len = 0;
919 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
920 prefix = false;
922 if (msg->flags & LOG_CONT) {
923 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
924 prefix = false;
926 if (!(msg->flags & LOG_NEWLINE))
927 newline = false;
930 do {
931 const char *next = memchr(text, '\n', text_size);
932 size_t text_len;
934 if (next) {
935 text_len = next - text;
936 next++;
937 text_size -= next - text;
938 } else {
939 text_len = text_size;
942 if (buf) {
943 if (print_prefix(msg, syslog, NULL) +
944 text_len + 1 >= size - len)
945 break;
947 if (prefix)
948 len += print_prefix(msg, syslog, buf + len);
949 memcpy(buf + len, text, text_len);
950 len += text_len;
951 if (next || newline)
952 buf[len++] = '\n';
953 } else {
954 /* SYSLOG_ACTION_* buffer size only calculation */
955 if (prefix)
956 len += print_prefix(msg, syslog, NULL);
957 len += text_len;
958 if (next || newline)
959 len++;
962 prefix = true;
963 text = next;
964 } while (text);
966 return len;
969 static int syslog_print(char __user *buf, int size)
971 char *text;
972 struct log *msg;
973 int len = 0;
975 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
976 if (!text)
977 return -ENOMEM;
979 while (size > 0) {
980 size_t n;
981 size_t skip;
983 raw_spin_lock_irq(&logbuf_lock);
984 if (syslog_seq < log_first_seq) {
985 /* messages are gone, move to first one */
986 syslog_seq = log_first_seq;
987 syslog_idx = log_first_idx;
988 syslog_prev = 0;
989 syslog_partial = 0;
991 if (syslog_seq == log_next_seq) {
992 raw_spin_unlock_irq(&logbuf_lock);
993 break;
996 skip = syslog_partial;
997 msg = log_from_idx(syslog_idx);
998 n = msg_print_text(msg, syslog_prev, true, text,
999 LOG_LINE_MAX + PREFIX_MAX);
1000 if (n - syslog_partial <= size) {
1001 /* message fits into buffer, move forward */
1002 syslog_idx = log_next(syslog_idx);
1003 syslog_seq++;
1004 syslog_prev = msg->flags;
1005 n -= syslog_partial;
1006 syslog_partial = 0;
1007 } else if (!len){
1008 /* partial read(), remember position */
1009 n = size;
1010 syslog_partial += n;
1011 } else
1012 n = 0;
1013 raw_spin_unlock_irq(&logbuf_lock);
1015 if (!n)
1016 break;
1018 if (copy_to_user(buf, text + skip, n)) {
1019 if (!len)
1020 len = -EFAULT;
1021 break;
1024 len += n;
1025 size -= n;
1026 buf += n;
1029 kfree(text);
1030 return len;
1033 static int syslog_print_all(char __user *buf, int size, bool clear)
1035 char *text;
1036 int len = 0;
1038 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1039 if (!text)
1040 return -ENOMEM;
1042 raw_spin_lock_irq(&logbuf_lock);
1043 if (buf) {
1044 u64 next_seq;
1045 u64 seq;
1046 u32 idx;
1047 enum log_flags prev;
1049 if (clear_seq < log_first_seq) {
1050 /* messages are gone, move to first available one */
1051 clear_seq = log_first_seq;
1052 clear_idx = log_first_idx;
1056 * Find first record that fits, including all following records,
1057 * into the user-provided buffer for this dump.
1059 seq = clear_seq;
1060 idx = clear_idx;
1061 prev = 0;
1062 while (seq < log_next_seq) {
1063 struct log *msg = log_from_idx(idx);
1065 len += msg_print_text(msg, prev, true, NULL, 0);
1066 prev = msg->flags;
1067 idx = log_next(idx);
1068 seq++;
1071 /* move first record forward until length fits into the buffer */
1072 seq = clear_seq;
1073 idx = clear_idx;
1074 prev = 0;
1075 while (len > size && seq < log_next_seq) {
1076 struct log *msg = log_from_idx(idx);
1078 len -= msg_print_text(msg, prev, true, NULL, 0);
1079 prev = msg->flags;
1080 idx = log_next(idx);
1081 seq++;
1084 /* last message fitting into this dump */
1085 next_seq = log_next_seq;
1087 len = 0;
1088 prev = 0;
1089 while (len >= 0 && seq < next_seq) {
1090 struct log *msg = log_from_idx(idx);
1091 int textlen;
1093 textlen = msg_print_text(msg, prev, true, text,
1094 LOG_LINE_MAX + PREFIX_MAX);
1095 if (textlen < 0) {
1096 len = textlen;
1097 break;
1099 idx = log_next(idx);
1100 seq++;
1101 prev = msg->flags;
1103 raw_spin_unlock_irq(&logbuf_lock);
1104 if (copy_to_user(buf + len, text, textlen))
1105 len = -EFAULT;
1106 else
1107 len += textlen;
1108 raw_spin_lock_irq(&logbuf_lock);
1110 if (seq < log_first_seq) {
1111 /* messages are gone, move to next one */
1112 seq = log_first_seq;
1113 idx = log_first_idx;
1114 prev = 0;
1119 if (clear) {
1120 clear_seq = log_next_seq;
1121 clear_idx = log_next_idx;
1123 raw_spin_unlock_irq(&logbuf_lock);
1125 kfree(text);
1126 return len;
1129 int do_syslog(int type, char __user *buf, int len, bool from_file)
1131 bool clear = false;
1132 static int saved_console_loglevel = -1;
1133 int error;
1135 error = check_syslog_permissions(type, from_file);
1136 if (error)
1137 goto out;
1139 error = security_syslog(type);
1140 if (error)
1141 return error;
1143 switch (type) {
1144 case SYSLOG_ACTION_CLOSE: /* Close log */
1145 break;
1146 case SYSLOG_ACTION_OPEN: /* Open log */
1147 break;
1148 case SYSLOG_ACTION_READ: /* Read from log */
1149 error = -EINVAL;
1150 if (!buf || len < 0)
1151 goto out;
1152 error = 0;
1153 if (!len)
1154 goto out;
1155 if (!access_ok(VERIFY_WRITE, buf, len)) {
1156 error = -EFAULT;
1157 goto out;
1159 error = wait_event_interruptible(log_wait,
1160 syslog_seq != log_next_seq);
1161 if (error)
1162 goto out;
1163 error = syslog_print(buf, len);
1164 break;
1165 /* Read/clear last kernel messages */
1166 case SYSLOG_ACTION_READ_CLEAR:
1167 clear = true;
1168 /* FALL THRU */
1169 /* Read last kernel messages */
1170 case SYSLOG_ACTION_READ_ALL:
1171 error = -EINVAL;
1172 if (!buf || len < 0)
1173 goto out;
1174 error = 0;
1175 if (!len)
1176 goto out;
1177 if (!access_ok(VERIFY_WRITE, buf, len)) {
1178 error = -EFAULT;
1179 goto out;
1181 error = syslog_print_all(buf, len, clear);
1182 break;
1183 /* Clear ring buffer */
1184 case SYSLOG_ACTION_CLEAR:
1185 syslog_print_all(NULL, 0, true);
1186 break;
1187 /* Disable logging to console */
1188 case SYSLOG_ACTION_CONSOLE_OFF:
1189 if (saved_console_loglevel == -1)
1190 saved_console_loglevel = console_loglevel;
1191 console_loglevel = minimum_console_loglevel;
1192 break;
1193 /* Enable logging to console */
1194 case SYSLOG_ACTION_CONSOLE_ON:
1195 if (saved_console_loglevel != -1) {
1196 console_loglevel = saved_console_loglevel;
1197 saved_console_loglevel = -1;
1199 break;
1200 /* Set level of messages printed to console */
1201 case SYSLOG_ACTION_CONSOLE_LEVEL:
1202 error = -EINVAL;
1203 if (len < 1 || len > 8)
1204 goto out;
1205 if (len < minimum_console_loglevel)
1206 len = minimum_console_loglevel;
1207 console_loglevel = len;
1208 /* Implicitly re-enable logging to console */
1209 saved_console_loglevel = -1;
1210 error = 0;
1211 break;
1212 /* Number of chars in the log buffer */
1213 case SYSLOG_ACTION_SIZE_UNREAD:
1214 raw_spin_lock_irq(&logbuf_lock);
1215 if (syslog_seq < log_first_seq) {
1216 /* messages are gone, move to first one */
1217 syslog_seq = log_first_seq;
1218 syslog_idx = log_first_idx;
1219 syslog_prev = 0;
1220 syslog_partial = 0;
1222 if (from_file) {
1224 * Short-cut for poll(/"proc/kmsg") which simply checks
1225 * for pending data, not the size; return the count of
1226 * records, not the length.
1228 error = log_next_idx - syslog_idx;
1229 } else {
1230 u64 seq = syslog_seq;
1231 u32 idx = syslog_idx;
1232 enum log_flags prev = syslog_prev;
1234 error = 0;
1235 while (seq < log_next_seq) {
1236 struct log *msg = log_from_idx(idx);
1238 error += msg_print_text(msg, prev, true, NULL, 0);
1239 idx = log_next(idx);
1240 seq++;
1241 prev = msg->flags;
1243 error -= syslog_partial;
1245 raw_spin_unlock_irq(&logbuf_lock);
1246 break;
1247 /* Size of the log buffer */
1248 case SYSLOG_ACTION_SIZE_BUFFER:
1249 error = log_buf_len;
1250 break;
1251 default:
1252 error = -EINVAL;
1253 break;
1255 out:
1256 return error;
1259 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1261 return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1265 * Call the console drivers, asking them to write out
1266 * log_buf[start] to log_buf[end - 1].
1267 * The console_lock must be held.
1269 static void call_console_drivers(int level, const char *text, size_t len)
1271 struct console *con;
1273 trace_console(text, len);
1275 if (level >= console_loglevel && !ignore_loglevel)
1276 return;
1277 if (!console_drivers)
1278 return;
1280 for_each_console(con) {
1281 if (exclusive_console && con != exclusive_console)
1282 continue;
1283 if (!(con->flags & CON_ENABLED))
1284 continue;
1285 if (!con->write)
1286 continue;
1287 if (!cpu_online(smp_processor_id()) &&
1288 !(con->flags & CON_ANYTIME))
1289 continue;
1290 con->write(con, text, len);
1295 * Zap console related locks when oopsing. Only zap at most once
1296 * every 10 seconds, to leave time for slow consoles to print a
1297 * full oops.
1299 static void zap_locks(void)
1301 static unsigned long oops_timestamp;
1303 if (time_after_eq(jiffies, oops_timestamp) &&
1304 !time_after(jiffies, oops_timestamp + 30 * HZ))
1305 return;
1307 oops_timestamp = jiffies;
1309 debug_locks_off();
1310 /* If a crash is occurring, make sure we can't deadlock */
1311 raw_spin_lock_init(&logbuf_lock);
1312 /* And make sure that we print immediately */
1313 sema_init(&console_sem, 1);
1316 /* Check if we have any console registered that can be called early in boot. */
1317 static int have_callable_console(void)
1319 struct console *con;
1321 for_each_console(con)
1322 if (con->flags & CON_ANYTIME)
1323 return 1;
1325 return 0;
1329 * Can we actually use the console at this time on this cpu?
1331 * Console drivers may assume that per-cpu resources have
1332 * been allocated. So unless they're explicitly marked as
1333 * being able to cope (CON_ANYTIME) don't call them until
1334 * this CPU is officially up.
1336 static inline int can_use_console(unsigned int cpu)
1338 return cpu_online(cpu) || have_callable_console();
1342 * Try to get console ownership to actually show the kernel
1343 * messages from a 'printk'. Return true (and with the
1344 * console_lock held, and 'console_locked' set) if it
1345 * is successful, false otherwise.
1347 * This gets called with the 'logbuf_lock' spinlock held and
1348 * interrupts disabled. It should return with 'lockbuf_lock'
1349 * released but interrupts still disabled.
1351 static int console_trylock_for_printk(unsigned int cpu)
1352 __releases(&logbuf_lock)
1354 int retval = 0, wake = 0;
1356 if (console_trylock()) {
1357 retval = 1;
1360 * If we can't use the console, we need to release
1361 * the console semaphore by hand to avoid flushing
1362 * the buffer. We need to hold the console semaphore
1363 * in order to do this test safely.
1365 if (!can_use_console(cpu)) {
1366 console_locked = 0;
1367 wake = 1;
1368 retval = 0;
1371 logbuf_cpu = UINT_MAX;
1372 if (wake)
1373 up(&console_sem);
1374 raw_spin_unlock(&logbuf_lock);
1375 return retval;
1378 int printk_delay_msec __read_mostly;
1380 static inline void printk_delay(void)
1382 if (unlikely(printk_delay_msec)) {
1383 int m = printk_delay_msec;
1385 while (m--) {
1386 mdelay(1);
1387 touch_nmi_watchdog();
1393 * Continuation lines are buffered, and not committed to the record buffer
1394 * until the line is complete, or a race forces it. The line fragments
1395 * though, are printed immediately to the consoles to ensure everything has
1396 * reached the console in case of a kernel crash.
1398 static struct cont {
1399 char buf[LOG_LINE_MAX];
1400 size_t len; /* length == 0 means unused buffer */
1401 size_t cons; /* bytes written to console */
1402 struct task_struct *owner; /* task of first print*/
1403 u64 ts_nsec; /* time of first print */
1404 u8 level; /* log level of first message */
1405 u8 facility; /* log level of first message */
1406 enum log_flags flags; /* prefix, newline flags */
1407 bool flushed:1; /* buffer sealed and committed */
1408 } cont;
1410 static void cont_flush(enum log_flags flags)
1412 if (cont.flushed)
1413 return;
1414 if (cont.len == 0)
1415 return;
1417 if (cont.cons) {
1419 * If a fragment of this line was directly flushed to the
1420 * console; wait for the console to pick up the rest of the
1421 * line. LOG_NOCONS suppresses a duplicated output.
1423 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1424 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1425 cont.flags = flags;
1426 cont.flushed = true;
1427 } else {
1429 * If no fragment of this line ever reached the console,
1430 * just submit it to the store and free the buffer.
1432 log_store(cont.facility, cont.level, flags, 0,
1433 NULL, 0, cont.buf, cont.len);
1434 cont.len = 0;
1438 static bool cont_add(int facility, int level, const char *text, size_t len)
1440 if (cont.len && cont.flushed)
1441 return false;
1443 if (cont.len + len > sizeof(cont.buf)) {
1444 /* the line gets too long, split it up in separate records */
1445 cont_flush(LOG_CONT);
1446 return false;
1449 if (!cont.len) {
1450 cont.facility = facility;
1451 cont.level = level;
1452 cont.owner = current;
1453 cont.ts_nsec = local_clock();
1454 cont.flags = 0;
1455 cont.cons = 0;
1456 cont.flushed = false;
1459 memcpy(cont.buf + cont.len, text, len);
1460 cont.len += len;
1462 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1463 cont_flush(LOG_CONT);
1465 return true;
1468 static size_t cont_print_text(char *text, size_t size)
1470 size_t textlen = 0;
1471 size_t len;
1473 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1474 textlen += print_time(cont.ts_nsec, text);
1475 size -= textlen;
1478 len = cont.len - cont.cons;
1479 if (len > 0) {
1480 if (len+1 > size)
1481 len = size-1;
1482 memcpy(text + textlen, cont.buf + cont.cons, len);
1483 textlen += len;
1484 cont.cons = cont.len;
1487 if (cont.flushed) {
1488 if (cont.flags & LOG_NEWLINE)
1489 text[textlen++] = '\n';
1490 /* got everything, release buffer */
1491 cont.len = 0;
1493 return textlen;
1496 asmlinkage int vprintk_emit(int facility, int level,
1497 const char *dict, size_t dictlen,
1498 const char *fmt, va_list args)
1500 static int recursion_bug;
1501 static char textbuf[LOG_LINE_MAX];
1502 char *text = textbuf;
1503 size_t text_len;
1504 enum log_flags lflags = 0;
1505 unsigned long flags;
1506 int this_cpu;
1507 int printed_len = 0;
1509 boot_delay_msec(level);
1510 printk_delay();
1512 /* This stops the holder of console_sem just where we want him */
1513 local_irq_save(flags);
1514 this_cpu = smp_processor_id();
1517 * Ouch, printk recursed into itself!
1519 if (unlikely(logbuf_cpu == this_cpu)) {
1521 * If a crash is occurring during printk() on this CPU,
1522 * then try to get the crash message out but make sure
1523 * we can't deadlock. Otherwise just return to avoid the
1524 * recursion and return - but flag the recursion so that
1525 * it can be printed at the next appropriate moment:
1527 if (!oops_in_progress && !lockdep_recursing(current)) {
1528 recursion_bug = 1;
1529 goto out_restore_irqs;
1531 zap_locks();
1534 lockdep_off();
1535 raw_spin_lock(&logbuf_lock);
1536 logbuf_cpu = this_cpu;
1538 if (recursion_bug) {
1539 static const char recursion_msg[] =
1540 "BUG: recent printk recursion!";
1542 recursion_bug = 0;
1543 printed_len += strlen(recursion_msg);
1544 /* emit KERN_CRIT message */
1545 log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1546 NULL, 0, recursion_msg, printed_len);
1550 * The printf needs to come first; we need the syslog
1551 * prefix which might be passed-in as a parameter.
1553 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1555 /* mark and strip a trailing newline */
1556 if (text_len && text[text_len-1] == '\n') {
1557 text_len--;
1558 lflags |= LOG_NEWLINE;
1561 /* strip kernel syslog prefix and extract log level or control flags */
1562 if (facility == 0) {
1563 int kern_level = printk_get_level(text);
1565 if (kern_level) {
1566 const char *end_of_header = printk_skip_level(text);
1567 switch (kern_level) {
1568 case '0' ... '7':
1569 if (level == -1)
1570 level = kern_level - '0';
1571 case 'd': /* KERN_DEFAULT */
1572 lflags |= LOG_PREFIX;
1573 case 'c': /* KERN_CONT */
1574 break;
1576 text_len -= end_of_header - text;
1577 text = (char *)end_of_header;
1581 if (level == -1)
1582 level = default_message_loglevel;
1584 if (dict)
1585 lflags |= LOG_PREFIX|LOG_NEWLINE;
1587 if (!(lflags & LOG_NEWLINE)) {
1589 * Flush the conflicting buffer. An earlier newline was missing,
1590 * or another task also prints continuation lines.
1592 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1593 cont_flush(LOG_NEWLINE);
1595 /* buffer line if possible, otherwise store it right away */
1596 if (!cont_add(facility, level, text, text_len))
1597 log_store(facility, level, lflags | LOG_CONT, 0,
1598 dict, dictlen, text, text_len);
1599 } else {
1600 bool stored = false;
1603 * If an earlier newline was missing and it was the same task,
1604 * either merge it with the current buffer and flush, or if
1605 * there was a race with interrupts (prefix == true) then just
1606 * flush it out and store this line separately.
1608 if (cont.len && cont.owner == current) {
1609 if (!(lflags & LOG_PREFIX))
1610 stored = cont_add(facility, level, text, text_len);
1611 cont_flush(LOG_NEWLINE);
1614 if (!stored)
1615 log_store(facility, level, lflags, 0,
1616 dict, dictlen, text, text_len);
1618 printed_len += text_len;
1621 * Try to acquire and then immediately release the console semaphore.
1622 * The release will print out buffers and wake up /dev/kmsg and syslog()
1623 * users.
1625 * The console_trylock_for_printk() function will release 'logbuf_lock'
1626 * regardless of whether it actually gets the console semaphore or not.
1628 if (console_trylock_for_printk(this_cpu))
1629 console_unlock();
1631 lockdep_on();
1632 out_restore_irqs:
1633 local_irq_restore(flags);
1635 return printed_len;
1637 EXPORT_SYMBOL(vprintk_emit);
1639 asmlinkage int vprintk(const char *fmt, va_list args)
1641 return vprintk_emit(0, -1, NULL, 0, fmt, args);
1643 EXPORT_SYMBOL(vprintk);
1645 asmlinkage int printk_emit(int facility, int level,
1646 const char *dict, size_t dictlen,
1647 const char *fmt, ...)
1649 va_list args;
1650 int r;
1652 va_start(args, fmt);
1653 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1654 va_end(args);
1656 return r;
1658 EXPORT_SYMBOL(printk_emit);
1661 * printk - print a kernel message
1662 * @fmt: format string
1664 * This is printk(). It can be called from any context. We want it to work.
1666 * We try to grab the console_lock. If we succeed, it's easy - we log the
1667 * output and call the console drivers. If we fail to get the semaphore, we
1668 * place the output into the log buffer and return. The current holder of
1669 * the console_sem will notice the new output in console_unlock(); and will
1670 * send it to the consoles before releasing the lock.
1672 * One effect of this deferred printing is that code which calls printk() and
1673 * then changes console_loglevel may break. This is because console_loglevel
1674 * is inspected when the actual printing occurs.
1676 * See also:
1677 * printf(3)
1679 * See the vsnprintf() documentation for format string extensions over C99.
1681 asmlinkage int printk(const char *fmt, ...)
1683 va_list args;
1684 int r;
1686 #ifdef CONFIG_KGDB_KDB
1687 if (unlikely(kdb_trap_printk)) {
1688 va_start(args, fmt);
1689 r = vkdb_printf(fmt, args);
1690 va_end(args);
1691 return r;
1693 #endif
1694 va_start(args, fmt);
1695 r = vprintk_emit(0, -1, NULL, 0, fmt, args);
1696 va_end(args);
1698 return r;
1700 EXPORT_SYMBOL(printk);
1702 #else /* CONFIG_PRINTK */
1704 #define LOG_LINE_MAX 0
1705 #define PREFIX_MAX 0
1706 #define LOG_LINE_MAX 0
1707 static u64 syslog_seq;
1708 static u32 syslog_idx;
1709 static u64 console_seq;
1710 static u32 console_idx;
1711 static enum log_flags syslog_prev;
1712 static u64 log_first_seq;
1713 static u32 log_first_idx;
1714 static u64 log_next_seq;
1715 static enum log_flags console_prev;
1716 static struct cont {
1717 size_t len;
1718 size_t cons;
1719 u8 level;
1720 bool flushed:1;
1721 } cont;
1722 static struct log *log_from_idx(u32 idx) { return NULL; }
1723 static u32 log_next(u32 idx) { return 0; }
1724 static void call_console_drivers(int level, const char *text, size_t len) {}
1725 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
1726 bool syslog, char *buf, size_t size) { return 0; }
1727 static size_t cont_print_text(char *text, size_t size) { return 0; }
1729 #endif /* CONFIG_PRINTK */
1731 #ifdef CONFIG_EARLY_PRINTK
1732 struct console *early_console;
1734 void early_vprintk(const char *fmt, va_list ap)
1736 if (early_console) {
1737 char buf[512];
1738 int n = vscnprintf(buf, sizeof(buf), fmt, ap);
1740 early_console->write(early_console, buf, n);
1744 asmlinkage void early_printk(const char *fmt, ...)
1746 va_list ap;
1748 va_start(ap, fmt);
1749 early_vprintk(fmt, ap);
1750 va_end(ap);
1752 #endif
1754 static int __add_preferred_console(char *name, int idx, char *options,
1755 char *brl_options)
1757 struct console_cmdline *c;
1758 int i;
1761 * See if this tty is not yet registered, and
1762 * if we have a slot free.
1764 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1765 if (strcmp(console_cmdline[i].name, name) == 0 &&
1766 console_cmdline[i].index == idx) {
1767 if (!brl_options)
1768 selected_console = i;
1769 return 0;
1771 if (i == MAX_CMDLINECONSOLES)
1772 return -E2BIG;
1773 if (!brl_options)
1774 selected_console = i;
1775 c = &console_cmdline[i];
1776 strlcpy(c->name, name, sizeof(c->name));
1777 c->options = options;
1778 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1779 c->brl_options = brl_options;
1780 #endif
1781 c->index = idx;
1782 return 0;
1785 * Set up a list of consoles. Called from init/main.c
1787 static int __init console_setup(char *str)
1789 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for index */
1790 char *s, *options, *brl_options = NULL;
1791 int idx;
1793 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1794 if (!memcmp(str, "brl,", 4)) {
1795 brl_options = "";
1796 str += 4;
1797 } else if (!memcmp(str, "brl=", 4)) {
1798 brl_options = str + 4;
1799 str = strchr(brl_options, ',');
1800 if (!str) {
1801 printk(KERN_ERR "need port name after brl=\n");
1802 return 1;
1804 *(str++) = 0;
1806 #endif
1809 * Decode str into name, index, options.
1811 if (str[0] >= '0' && str[0] <= '9') {
1812 strcpy(buf, "ttyS");
1813 strncpy(buf + 4, str, sizeof(buf) - 5);
1814 } else {
1815 strncpy(buf, str, sizeof(buf) - 1);
1817 buf[sizeof(buf) - 1] = 0;
1818 if ((options = strchr(str, ',')) != NULL)
1819 *(options++) = 0;
1820 #ifdef __sparc__
1821 if (!strcmp(str, "ttya"))
1822 strcpy(buf, "ttyS0");
1823 if (!strcmp(str, "ttyb"))
1824 strcpy(buf, "ttyS1");
1825 #endif
1826 for (s = buf; *s; s++)
1827 if ((*s >= '0' && *s <= '9') || *s == ',')
1828 break;
1829 idx = simple_strtoul(s, NULL, 10);
1830 *s = 0;
1832 __add_preferred_console(buf, idx, options, brl_options);
1833 console_set_on_cmdline = 1;
1834 return 1;
1836 __setup("console=", console_setup);
1839 * add_preferred_console - add a device to the list of preferred consoles.
1840 * @name: device name
1841 * @idx: device index
1842 * @options: options for this console
1844 * The last preferred console added will be used for kernel messages
1845 * and stdin/out/err for init. Normally this is used by console_setup
1846 * above to handle user-supplied console arguments; however it can also
1847 * be used by arch-specific code either to override the user or more
1848 * commonly to provide a default console (ie from PROM variables) when
1849 * the user has not supplied one.
1851 int add_preferred_console(char *name, int idx, char *options)
1853 return __add_preferred_console(name, idx, options, NULL);
1856 int update_console_cmdline(char *name, int idx, char *name_new, int idx_new, char *options)
1858 struct console_cmdline *c;
1859 int i;
1861 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1862 if (strcmp(console_cmdline[i].name, name) == 0 &&
1863 console_cmdline[i].index == idx) {
1864 c = &console_cmdline[i];
1865 strlcpy(c->name, name_new, sizeof(c->name));
1866 c->name[sizeof(c->name) - 1] = 0;
1867 c->options = options;
1868 c->index = idx_new;
1869 return i;
1871 /* not found */
1872 return -1;
1875 bool console_suspend_enabled = 1;
1876 EXPORT_SYMBOL(console_suspend_enabled);
1878 static int __init console_suspend_disable(char *str)
1880 console_suspend_enabled = 0;
1881 return 1;
1883 __setup("no_console_suspend", console_suspend_disable);
1884 module_param_named(console_suspend, console_suspend_enabled,
1885 bool, S_IRUGO | S_IWUSR);
1886 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
1887 " and hibernate operations");
1890 * suspend_console - suspend the console subsystem
1892 * This disables printk() while we go into suspend states
1894 void suspend_console(void)
1896 if (!console_suspend_enabled)
1897 return;
1898 printk("Suspending console(s) (use no_console_suspend to debug)\n");
1899 console_lock();
1900 console_suspended = 1;
1901 up(&console_sem);
1904 void resume_console(void)
1906 if (!console_suspend_enabled)
1907 return;
1908 down(&console_sem);
1909 console_suspended = 0;
1910 console_unlock();
1914 * console_cpu_notify - print deferred console messages after CPU hotplug
1915 * @self: notifier struct
1916 * @action: CPU hotplug event
1917 * @hcpu: unused
1919 * If printk() is called from a CPU that is not online yet, the messages
1920 * will be spooled but will not show up on the console. This function is
1921 * called when a new CPU comes online (or fails to come up), and ensures
1922 * that any such output gets printed.
1924 static int __cpuinit console_cpu_notify(struct notifier_block *self,
1925 unsigned long action, void *hcpu)
1927 switch (action) {
1928 case CPU_ONLINE:
1929 case CPU_DEAD:
1930 case CPU_DOWN_FAILED:
1931 case CPU_UP_CANCELED:
1932 console_lock();
1933 console_unlock();
1935 return NOTIFY_OK;
1939 * console_lock - lock the console system for exclusive use.
1941 * Acquires a lock which guarantees that the caller has
1942 * exclusive access to the console system and the console_drivers list.
1944 * Can sleep, returns nothing.
1946 void console_lock(void)
1948 might_sleep();
1950 down(&console_sem);
1951 if (console_suspended)
1952 return;
1953 console_locked = 1;
1954 console_may_schedule = 1;
1955 mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);
1957 EXPORT_SYMBOL(console_lock);
1960 * console_trylock - try to lock the console system for exclusive use.
1962 * Tried to acquire a lock which guarantees that the caller has
1963 * exclusive access to the console system and the console_drivers list.
1965 * returns 1 on success, and 0 on failure to acquire the lock.
1967 int console_trylock(void)
1969 if (down_trylock(&console_sem))
1970 return 0;
1971 if (console_suspended) {
1972 up(&console_sem);
1973 return 0;
1975 console_locked = 1;
1976 console_may_schedule = 0;
1977 mutex_acquire(&console_lock_dep_map, 0, 1, _RET_IP_);
1978 return 1;
1980 EXPORT_SYMBOL(console_trylock);
1982 int is_console_locked(void)
1984 return console_locked;
1987 static void console_cont_flush(char *text, size_t size)
1989 unsigned long flags;
1990 size_t len;
1992 raw_spin_lock_irqsave(&logbuf_lock, flags);
1994 if (!cont.len)
1995 goto out;
1998 * We still queue earlier records, likely because the console was
1999 * busy. The earlier ones need to be printed before this one, we
2000 * did not flush any fragment so far, so just let it queue up.
2002 if (console_seq < log_next_seq && !cont.cons)
2003 goto out;
2005 len = cont_print_text(text, size);
2006 raw_spin_unlock(&logbuf_lock);
2007 stop_critical_timings();
2008 call_console_drivers(cont.level, text, len);
2009 start_critical_timings();
2010 local_irq_restore(flags);
2011 return;
2012 out:
2013 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2017 * console_unlock - unlock the console system
2019 * Releases the console_lock which the caller holds on the console system
2020 * and the console driver list.
2022 * While the console_lock was held, console output may have been buffered
2023 * by printk(). If this is the case, console_unlock(); emits
2024 * the output prior to releasing the lock.
2026 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2028 * console_unlock(); may be called from any context.
2030 void console_unlock(void)
2032 static char text[LOG_LINE_MAX + PREFIX_MAX];
2033 static u64 seen_seq;
2034 unsigned long flags;
2035 bool wake_klogd = false;
2036 bool retry;
2038 if (console_suspended) {
2039 up(&console_sem);
2040 return;
2043 console_may_schedule = 0;
2045 /* flush buffered message fragment immediately to console */
2046 console_cont_flush(text, sizeof(text));
2047 again:
2048 for (;;) {
2049 struct log *msg;
2050 size_t len;
2051 int level;
2053 raw_spin_lock_irqsave(&logbuf_lock, flags);
2054 if (seen_seq != log_next_seq) {
2055 wake_klogd = true;
2056 seen_seq = log_next_seq;
2059 if (console_seq < log_first_seq) {
2060 /* messages are gone, move to first one */
2061 console_seq = log_first_seq;
2062 console_idx = log_first_idx;
2063 console_prev = 0;
2065 skip:
2066 if (console_seq == log_next_seq)
2067 break;
2069 msg = log_from_idx(console_idx);
2070 if (msg->flags & LOG_NOCONS) {
2072 * Skip record we have buffered and already printed
2073 * directly to the console when we received it.
2075 console_idx = log_next(console_idx);
2076 console_seq++;
2078 * We will get here again when we register a new
2079 * CON_PRINTBUFFER console. Clear the flag so we
2080 * will properly dump everything later.
2082 msg->flags &= ~LOG_NOCONS;
2083 console_prev = msg->flags;
2084 goto skip;
2087 level = msg->level;
2088 len = msg_print_text(msg, console_prev, false,
2089 text, sizeof(text));
2090 console_idx = log_next(console_idx);
2091 console_seq++;
2092 console_prev = msg->flags;
2093 raw_spin_unlock(&logbuf_lock);
2095 stop_critical_timings(); /* don't trace print latency */
2096 call_console_drivers(level, text, len);
2097 start_critical_timings();
2098 local_irq_restore(flags);
2100 console_locked = 0;
2101 mutex_release(&console_lock_dep_map, 1, _RET_IP_);
2103 /* Release the exclusive_console once it is used */
2104 if (unlikely(exclusive_console))
2105 exclusive_console = NULL;
2107 raw_spin_unlock(&logbuf_lock);
2109 up(&console_sem);
2112 * Someone could have filled up the buffer again, so re-check if there's
2113 * something to flush. In case we cannot trylock the console_sem again,
2114 * there's a new owner and the console_unlock() from them will do the
2115 * flush, no worries.
2117 raw_spin_lock(&logbuf_lock);
2118 retry = console_seq != log_next_seq;
2119 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2121 if (retry && console_trylock())
2122 goto again;
2124 if (wake_klogd)
2125 wake_up_klogd();
2127 EXPORT_SYMBOL(console_unlock);
2130 * console_conditional_schedule - yield the CPU if required
2132 * If the console code is currently allowed to sleep, and
2133 * if this CPU should yield the CPU to another task, do
2134 * so here.
2136 * Must be called within console_lock();.
2138 void __sched console_conditional_schedule(void)
2140 if (console_may_schedule)
2141 cond_resched();
2143 EXPORT_SYMBOL(console_conditional_schedule);
2145 void console_unblank(void)
2147 struct console *c;
2150 * console_unblank can no longer be called in interrupt context unless
2151 * oops_in_progress is set to 1..
2153 if (oops_in_progress) {
2154 if (down_trylock(&console_sem) != 0)
2155 return;
2156 } else
2157 console_lock();
2159 console_locked = 1;
2160 console_may_schedule = 0;
2161 for_each_console(c)
2162 if ((c->flags & CON_ENABLED) && c->unblank)
2163 c->unblank();
2164 console_unlock();
2168 * Return the console tty driver structure and its associated index
2170 struct tty_driver *console_device(int *index)
2172 struct console *c;
2173 struct tty_driver *driver = NULL;
2175 console_lock();
2176 for_each_console(c) {
2177 if (!c->device)
2178 continue;
2179 driver = c->device(c, index);
2180 if (driver)
2181 break;
2183 console_unlock();
2184 return driver;
2188 * Prevent further output on the passed console device so that (for example)
2189 * serial drivers can disable console output before suspending a port, and can
2190 * re-enable output afterwards.
2192 void console_stop(struct console *console)
2194 console_lock();
2195 console->flags &= ~CON_ENABLED;
2196 console_unlock();
2198 EXPORT_SYMBOL(console_stop);
2200 void console_start(struct console *console)
2202 console_lock();
2203 console->flags |= CON_ENABLED;
2204 console_unlock();
2206 EXPORT_SYMBOL(console_start);
2208 static int __read_mostly keep_bootcon;
2210 static int __init keep_bootcon_setup(char *str)
2212 keep_bootcon = 1;
2213 printk(KERN_INFO "debug: skip boot console de-registration.\n");
2215 return 0;
2218 early_param("keep_bootcon", keep_bootcon_setup);
2221 * The console driver calls this routine during kernel initialization
2222 * to register the console printing procedure with printk() and to
2223 * print any messages that were printed by the kernel before the
2224 * console driver was initialized.
2226 * This can happen pretty early during the boot process (because of
2227 * early_printk) - sometimes before setup_arch() completes - be careful
2228 * of what kernel features are used - they may not be initialised yet.
2230 * There are two types of consoles - bootconsoles (early_printk) and
2231 * "real" consoles (everything which is not a bootconsole) which are
2232 * handled differently.
2233 * - Any number of bootconsoles can be registered at any time.
2234 * - As soon as a "real" console is registered, all bootconsoles
2235 * will be unregistered automatically.
2236 * - Once a "real" console is registered, any attempt to register a
2237 * bootconsoles will be rejected
2239 void register_console(struct console *newcon)
2241 int i;
2242 unsigned long flags;
2243 struct console *bcon = NULL;
2246 * before we register a new CON_BOOT console, make sure we don't
2247 * already have a valid console
2249 if (console_drivers && newcon->flags & CON_BOOT) {
2250 /* find the last or real console */
2251 for_each_console(bcon) {
2252 if (!(bcon->flags & CON_BOOT)) {
2253 printk(KERN_INFO "Too late to register bootconsole %s%d\n",
2254 newcon->name, newcon->index);
2255 return;
2260 if (console_drivers && console_drivers->flags & CON_BOOT)
2261 bcon = console_drivers;
2263 if (preferred_console < 0 || bcon || !console_drivers)
2264 preferred_console = selected_console;
2266 if (newcon->early_setup)
2267 newcon->early_setup();
2270 * See if we want to use this console driver. If we
2271 * didn't select a console we take the first one
2272 * that registers here.
2274 if (preferred_console < 0) {
2275 if (newcon->index < 0)
2276 newcon->index = 0;
2277 if (newcon->setup == NULL ||
2278 newcon->setup(newcon, NULL) == 0) {
2279 newcon->flags |= CON_ENABLED;
2280 if (newcon->device) {
2281 newcon->flags |= CON_CONSDEV;
2282 preferred_console = 0;
2288 * See if this console matches one we selected on
2289 * the command line.
2291 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0];
2292 i++) {
2293 if (strcmp(console_cmdline[i].name, newcon->name) != 0)
2294 continue;
2295 if (newcon->index >= 0 &&
2296 newcon->index != console_cmdline[i].index)
2297 continue;
2298 if (newcon->index < 0)
2299 newcon->index = console_cmdline[i].index;
2300 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2301 if (console_cmdline[i].brl_options) {
2302 newcon->flags |= CON_BRL;
2303 braille_register_console(newcon,
2304 console_cmdline[i].index,
2305 console_cmdline[i].options,
2306 console_cmdline[i].brl_options);
2307 return;
2309 #endif
2310 if (newcon->setup &&
2311 newcon->setup(newcon, console_cmdline[i].options) != 0)
2312 break;
2313 newcon->flags |= CON_ENABLED;
2314 newcon->index = console_cmdline[i].index;
2315 if (i == selected_console) {
2316 newcon->flags |= CON_CONSDEV;
2317 preferred_console = selected_console;
2319 break;
2322 if (!(newcon->flags & CON_ENABLED))
2323 return;
2326 * If we have a bootconsole, and are switching to a real console,
2327 * don't print everything out again, since when the boot console, and
2328 * the real console are the same physical device, it's annoying to
2329 * see the beginning boot messages twice
2331 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2332 newcon->flags &= ~CON_PRINTBUFFER;
2335 * Put this console in the list - keep the
2336 * preferred driver at the head of the list.
2338 console_lock();
2339 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2340 newcon->next = console_drivers;
2341 console_drivers = newcon;
2342 if (newcon->next)
2343 newcon->next->flags &= ~CON_CONSDEV;
2344 } else {
2345 newcon->next = console_drivers->next;
2346 console_drivers->next = newcon;
2348 if (newcon->flags & CON_PRINTBUFFER) {
2350 * console_unlock(); will print out the buffered messages
2351 * for us.
2353 raw_spin_lock_irqsave(&logbuf_lock, flags);
2354 console_seq = syslog_seq;
2355 console_idx = syslog_idx;
2356 console_prev = syslog_prev;
2357 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2359 * We're about to replay the log buffer. Only do this to the
2360 * just-registered console to avoid excessive message spam to
2361 * the already-registered consoles.
2363 exclusive_console = newcon;
2365 console_unlock();
2366 console_sysfs_notify();
2369 * By unregistering the bootconsoles after we enable the real console
2370 * we get the "console xxx enabled" message on all the consoles -
2371 * boot consoles, real consoles, etc - this is to ensure that end
2372 * users know there might be something in the kernel's log buffer that
2373 * went to the bootconsole (that they do not see on the real console)
2375 if (bcon &&
2376 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2377 !keep_bootcon) {
2378 /* we need to iterate through twice, to make sure we print
2379 * everything out, before we unregister the console(s)
2381 printk(KERN_INFO "console [%s%d] enabled, bootconsole disabled\n",
2382 newcon->name, newcon->index);
2383 for_each_console(bcon)
2384 if (bcon->flags & CON_BOOT)
2385 unregister_console(bcon);
2386 } else {
2387 printk(KERN_INFO "%sconsole [%s%d] enabled\n",
2388 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2389 newcon->name, newcon->index);
2392 EXPORT_SYMBOL(register_console);
2394 int unregister_console(struct console *console)
2396 struct console *a, *b;
2397 int res = 1;
2399 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2400 if (console->flags & CON_BRL)
2401 return braille_unregister_console(console);
2402 #endif
2404 console_lock();
2405 if (console_drivers == console) {
2406 console_drivers=console->next;
2407 res = 0;
2408 } else if (console_drivers) {
2409 for (a=console_drivers->next, b=console_drivers ;
2410 a; b=a, a=b->next) {
2411 if (a == console) {
2412 b->next = a->next;
2413 res = 0;
2414 break;
2420 * If this isn't the last console and it has CON_CONSDEV set, we
2421 * need to set it on the next preferred console.
2423 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2424 console_drivers->flags |= CON_CONSDEV;
2426 console_unlock();
2427 console_sysfs_notify();
2428 return res;
2430 EXPORT_SYMBOL(unregister_console);
2432 static int __init printk_late_init(void)
2434 struct console *con;
2436 for_each_console(con) {
2437 if (!keep_bootcon && con->flags & CON_BOOT) {
2438 printk(KERN_INFO "turn off boot console %s%d\n",
2439 con->name, con->index);
2440 unregister_console(con);
2443 hotcpu_notifier(console_cpu_notify, 0);
2444 return 0;
2446 late_initcall(printk_late_init);
2448 #if defined CONFIG_PRINTK
2450 * Delayed printk version, for scheduler-internal messages:
2452 #define PRINTK_BUF_SIZE 512
2454 #define PRINTK_PENDING_WAKEUP 0x01
2455 #define PRINTK_PENDING_SCHED 0x02
2457 static DEFINE_PER_CPU(int, printk_pending);
2458 static DEFINE_PER_CPU(char [PRINTK_BUF_SIZE], printk_sched_buf);
2460 static void wake_up_klogd_work_func(struct irq_work *irq_work)
2462 int pending = __this_cpu_xchg(printk_pending, 0);
2464 if (pending & PRINTK_PENDING_SCHED) {
2465 char *buf = __get_cpu_var(printk_sched_buf);
2466 printk(KERN_WARNING "[sched_delayed] %s", buf);
2469 if (pending & PRINTK_PENDING_WAKEUP)
2470 wake_up_interruptible(&log_wait);
2473 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2474 .func = wake_up_klogd_work_func,
2475 .flags = IRQ_WORK_LAZY,
2478 void wake_up_klogd(void)
2480 preempt_disable();
2481 if (waitqueue_active(&log_wait)) {
2482 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2483 irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
2485 preempt_enable();
2488 int printk_sched(const char *fmt, ...)
2490 unsigned long flags;
2491 va_list args;
2492 char *buf;
2493 int r;
2495 local_irq_save(flags);
2496 buf = __get_cpu_var(printk_sched_buf);
2498 va_start(args, fmt);
2499 r = vsnprintf(buf, PRINTK_BUF_SIZE, fmt, args);
2500 va_end(args);
2502 __this_cpu_or(printk_pending, PRINTK_PENDING_SCHED);
2503 irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
2504 local_irq_restore(flags);
2506 return r;
2510 * printk rate limiting, lifted from the networking subsystem.
2512 * This enforces a rate limit: not more than 10 kernel messages
2513 * every 5s to make a denial-of-service attack impossible.
2515 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2517 int __printk_ratelimit(const char *func)
2519 return ___ratelimit(&printk_ratelimit_state, func);
2521 EXPORT_SYMBOL(__printk_ratelimit);
2524 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2525 * @caller_jiffies: pointer to caller's state
2526 * @interval_msecs: minimum interval between prints
2528 * printk_timed_ratelimit() returns true if more than @interval_msecs
2529 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2530 * returned true.
2532 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2533 unsigned int interval_msecs)
2535 if (*caller_jiffies == 0
2536 || !time_in_range(jiffies, *caller_jiffies,
2537 *caller_jiffies
2538 + msecs_to_jiffies(interval_msecs))) {
2539 *caller_jiffies = jiffies;
2540 return true;
2542 return false;
2544 EXPORT_SYMBOL(printk_timed_ratelimit);
2546 static DEFINE_SPINLOCK(dump_list_lock);
2547 static LIST_HEAD(dump_list);
2550 * kmsg_dump_register - register a kernel log dumper.
2551 * @dumper: pointer to the kmsg_dumper structure
2553 * Adds a kernel log dumper to the system. The dump callback in the
2554 * structure will be called when the kernel oopses or panics and must be
2555 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2557 int kmsg_dump_register(struct kmsg_dumper *dumper)
2559 unsigned long flags;
2560 int err = -EBUSY;
2562 /* The dump callback needs to be set */
2563 if (!dumper->dump)
2564 return -EINVAL;
2566 spin_lock_irqsave(&dump_list_lock, flags);
2567 /* Don't allow registering multiple times */
2568 if (!dumper->registered) {
2569 dumper->registered = 1;
2570 list_add_tail_rcu(&dumper->list, &dump_list);
2571 err = 0;
2573 spin_unlock_irqrestore(&dump_list_lock, flags);
2575 return err;
2577 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2580 * kmsg_dump_unregister - unregister a kmsg dumper.
2581 * @dumper: pointer to the kmsg_dumper structure
2583 * Removes a dump device from the system. Returns zero on success and
2584 * %-EINVAL otherwise.
2586 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2588 unsigned long flags;
2589 int err = -EINVAL;
2591 spin_lock_irqsave(&dump_list_lock, flags);
2592 if (dumper->registered) {
2593 dumper->registered = 0;
2594 list_del_rcu(&dumper->list);
2595 err = 0;
2597 spin_unlock_irqrestore(&dump_list_lock, flags);
2598 synchronize_rcu();
2600 return err;
2602 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2604 static bool always_kmsg_dump;
2605 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2608 * kmsg_dump - dump kernel log to kernel message dumpers.
2609 * @reason: the reason (oops, panic etc) for dumping
2611 * Call each of the registered dumper's dump() callback, which can
2612 * retrieve the kmsg records with kmsg_dump_get_line() or
2613 * kmsg_dump_get_buffer().
2615 void kmsg_dump(enum kmsg_dump_reason reason)
2617 struct kmsg_dumper *dumper;
2618 unsigned long flags;
2620 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2621 return;
2623 rcu_read_lock();
2624 list_for_each_entry_rcu(dumper, &dump_list, list) {
2625 if (dumper->max_reason && reason > dumper->max_reason)
2626 continue;
2628 /* initialize iterator with data about the stored records */
2629 dumper->active = true;
2631 raw_spin_lock_irqsave(&logbuf_lock, flags);
2632 dumper->cur_seq = clear_seq;
2633 dumper->cur_idx = clear_idx;
2634 dumper->next_seq = log_next_seq;
2635 dumper->next_idx = log_next_idx;
2636 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2638 /* invoke dumper which will iterate over records */
2639 dumper->dump(dumper, reason);
2641 /* reset iterator */
2642 dumper->active = false;
2644 rcu_read_unlock();
2648 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2649 * @dumper: registered kmsg dumper
2650 * @syslog: include the "<4>" prefixes
2651 * @line: buffer to copy the line to
2652 * @size: maximum size of the buffer
2653 * @len: length of line placed into buffer
2655 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2656 * record, and copy one record into the provided buffer.
2658 * Consecutive calls will return the next available record moving
2659 * towards the end of the buffer with the youngest messages.
2661 * A return value of FALSE indicates that there are no more records to
2662 * read.
2664 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2666 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2667 char *line, size_t size, size_t *len)
2669 struct log *msg;
2670 size_t l = 0;
2671 bool ret = false;
2673 if (!dumper->active)
2674 goto out;
2676 if (dumper->cur_seq < log_first_seq) {
2677 /* messages are gone, move to first available one */
2678 dumper->cur_seq = log_first_seq;
2679 dumper->cur_idx = log_first_idx;
2682 /* last entry */
2683 if (dumper->cur_seq >= log_next_seq)
2684 goto out;
2686 msg = log_from_idx(dumper->cur_idx);
2687 l = msg_print_text(msg, 0, syslog, line, size);
2689 dumper->cur_idx = log_next(dumper->cur_idx);
2690 dumper->cur_seq++;
2691 ret = true;
2692 out:
2693 if (len)
2694 *len = l;
2695 return ret;
2699 * kmsg_dump_get_line - retrieve one kmsg log line
2700 * @dumper: registered kmsg dumper
2701 * @syslog: include the "<4>" prefixes
2702 * @line: buffer to copy the line to
2703 * @size: maximum size of the buffer
2704 * @len: length of line placed into buffer
2706 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2707 * record, and copy one record into the provided buffer.
2709 * Consecutive calls will return the next available record moving
2710 * towards the end of the buffer with the youngest messages.
2712 * A return value of FALSE indicates that there are no more records to
2713 * read.
2715 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2716 char *line, size_t size, size_t *len)
2718 unsigned long flags;
2719 bool ret;
2721 raw_spin_lock_irqsave(&logbuf_lock, flags);
2722 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2723 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2725 return ret;
2727 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2730 * kmsg_dump_get_buffer - copy kmsg log lines
2731 * @dumper: registered kmsg dumper
2732 * @syslog: include the "<4>" prefixes
2733 * @buf: buffer to copy the line to
2734 * @size: maximum size of the buffer
2735 * @len: length of line placed into buffer
2737 * Start at the end of the kmsg buffer and fill the provided buffer
2738 * with as many of the the *youngest* kmsg records that fit into it.
2739 * If the buffer is large enough, all available kmsg records will be
2740 * copied with a single call.
2742 * Consecutive calls will fill the buffer with the next block of
2743 * available older records, not including the earlier retrieved ones.
2745 * A return value of FALSE indicates that there are no more records to
2746 * read.
2748 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2749 char *buf, size_t size, size_t *len)
2751 unsigned long flags;
2752 u64 seq;
2753 u32 idx;
2754 u64 next_seq;
2755 u32 next_idx;
2756 enum log_flags prev;
2757 size_t l = 0;
2758 bool ret = false;
2760 if (!dumper->active)
2761 goto out;
2763 raw_spin_lock_irqsave(&logbuf_lock, flags);
2764 if (dumper->cur_seq < log_first_seq) {
2765 /* messages are gone, move to first available one */
2766 dumper->cur_seq = log_first_seq;
2767 dumper->cur_idx = log_first_idx;
2770 /* last entry */
2771 if (dumper->cur_seq >= dumper->next_seq) {
2772 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2773 goto out;
2776 /* calculate length of entire buffer */
2777 seq = dumper->cur_seq;
2778 idx = dumper->cur_idx;
2779 prev = 0;
2780 while (seq < dumper->next_seq) {
2781 struct log *msg = log_from_idx(idx);
2783 l += msg_print_text(msg, prev, true, NULL, 0);
2784 idx = log_next(idx);
2785 seq++;
2786 prev = msg->flags;
2789 /* move first record forward until length fits into the buffer */
2790 seq = dumper->cur_seq;
2791 idx = dumper->cur_idx;
2792 prev = 0;
2793 while (l > size && seq < dumper->next_seq) {
2794 struct log *msg = log_from_idx(idx);
2796 l -= msg_print_text(msg, prev, true, NULL, 0);
2797 idx = log_next(idx);
2798 seq++;
2799 prev = msg->flags;
2802 /* last message in next interation */
2803 next_seq = seq;
2804 next_idx = idx;
2806 l = 0;
2807 prev = 0;
2808 while (seq < dumper->next_seq) {
2809 struct log *msg = log_from_idx(idx);
2811 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
2812 idx = log_next(idx);
2813 seq++;
2814 prev = msg->flags;
2817 dumper->next_seq = next_seq;
2818 dumper->next_idx = next_idx;
2819 ret = true;
2820 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2821 out:
2822 if (len)
2823 *len = l;
2824 return ret;
2826 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
2829 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
2830 * @dumper: registered kmsg dumper
2832 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2833 * kmsg_dump_get_buffer() can be called again and used multiple
2834 * times within the same dumper.dump() callback.
2836 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
2838 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
2840 dumper->cur_seq = clear_seq;
2841 dumper->cur_idx = clear_idx;
2842 dumper->next_seq = log_next_seq;
2843 dumper->next_idx = log_next_idx;
2847 * kmsg_dump_rewind - reset the interator
2848 * @dumper: registered kmsg dumper
2850 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2851 * kmsg_dump_get_buffer() can be called again and used multiple
2852 * times within the same dumper.dump() callback.
2854 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
2856 unsigned long flags;
2858 raw_spin_lock_irqsave(&logbuf_lock, flags);
2859 kmsg_dump_rewind_nolock(dumper);
2860 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2862 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
2864 static char dump_stack_arch_desc_str[128];
2867 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
2868 * @fmt: printf-style format string
2869 * @...: arguments for the format string
2871 * The configured string will be printed right after utsname during task
2872 * dumps. Usually used to add arch-specific system identifiers. If an
2873 * arch wants to make use of such an ID string, it should initialize this
2874 * as soon as possible during boot.
2876 void __init dump_stack_set_arch_desc(const char *fmt, ...)
2878 va_list args;
2880 va_start(args, fmt);
2881 vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
2882 fmt, args);
2883 va_end(args);
2887 * dump_stack_print_info - print generic debug info for dump_stack()
2888 * @log_lvl: log level
2890 * Arch-specific dump_stack() implementations can use this function to
2891 * print out the same debug information as the generic dump_stack().
2893 void dump_stack_print_info(const char *log_lvl)
2895 printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
2896 log_lvl, raw_smp_processor_id(), current->pid, current->comm,
2897 print_tainted(), init_utsname()->release,
2898 (int)strcspn(init_utsname()->version, " "),
2899 init_utsname()->version);
2901 if (dump_stack_arch_desc_str[0] != '\0')
2902 printk("%sHardware name: %s\n",
2903 log_lvl, dump_stack_arch_desc_str);
2905 print_worker_info(log_lvl, current);
2909 * show_regs_print_info - print generic debug info for show_regs()
2910 * @log_lvl: log level
2912 * show_regs() implementations can use this function to print out generic
2913 * debug information.
2915 void show_regs_print_info(const char *log_lvl)
2917 dump_stack_print_info(log_lvl);
2919 printk("%stask: %p ti: %p task.ti: %p\n",
2920 log_lvl, current, current_thread_info(),
2921 task_thread_info(current));
2924 #endif