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).
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>
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>
46 #include <asm/uaccess.h>
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/printk.h>
52 * Architectures can override it:
54 void asmlinkage __attribute__((weak)) early_printk(const char *fmt, ...)
58 /* printk's without a loglevel use this.. */
59 #define DEFAULT_MESSAGE_LOGLEVEL CONFIG_DEFAULT_MESSAGE_LOGLEVEL
61 /* We show everything that is MORE important than this.. */
62 #define MINIMUM_CONSOLE_LOGLEVEL 1 /* Minimum loglevel we let people use */
63 #define DEFAULT_CONSOLE_LOGLEVEL 7 /* anything MORE serious than KERN_DEBUG */
65 DECLARE_WAIT_QUEUE_HEAD(log_wait);
67 int console_printk[4] = {
68 DEFAULT_CONSOLE_LOGLEVEL, /* console_loglevel */
69 DEFAULT_MESSAGE_LOGLEVEL, /* default_message_loglevel */
70 MINIMUM_CONSOLE_LOGLEVEL, /* minimum_console_loglevel */
71 DEFAULT_CONSOLE_LOGLEVEL, /* default_console_loglevel */
75 * Low level drivers may need that to know if they can schedule in
76 * their unblank() callback or not. So let's export it.
79 EXPORT_SYMBOL(oops_in_progress);
82 * console_sem protects the console_drivers list, and also
83 * provides serialisation for access to the entire console
86 static DEFINE_SEMAPHORE(console_sem);
87 struct console *console_drivers;
88 EXPORT_SYMBOL_GPL(console_drivers);
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 */
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
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
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
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"
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"
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.
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 */
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);
220 /* the next printk record to read by syslog(READ) or /proc/kmsg */
221 static u64 syslog_seq;
222 static u32 syslog_idx;
223 static enum log_flags syslog_prev;
224 static size_t syslog_partial;
226 /* index and sequence number of the first record stored in the buffer */
227 static u64 log_first_seq;
228 static u32 log_first_idx;
230 /* index and sequence number of the next record to store in the buffer */
231 static u64 log_next_seq;
232 static u32 log_next_idx;
234 /* the next printk record to write to the console */
235 static u64 console_seq;
236 static u32 console_idx;
237 static enum log_flags console_prev;
239 /* the next printk record to read after the last 'clear' command */
240 static u64 clear_seq;
241 static u32 clear_idx;
243 #define PREFIX_MAX 32
244 #define LOG_LINE_MAX 1024 - PREFIX_MAX
247 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
250 #define LOG_ALIGN __alignof__(struct log)
252 #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
253 static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
254 static char *log_buf = __log_buf;
255 static u32 log_buf_len = __LOG_BUF_LEN;
257 /* cpu currently holding logbuf_lock */
258 static volatile unsigned int logbuf_cpu = UINT_MAX;
260 /* human readable text of the record */
261 static char *log_text(const struct log *msg)
263 return (char *)msg + sizeof(struct log);
266 /* optional key/value pair dictionary attached to the record */
267 static char *log_dict(const struct log *msg)
269 return (char *)msg + sizeof(struct log) + msg->text_len;
272 /* get record by index; idx must point to valid msg */
273 static struct log *log_from_idx(u32 idx)
275 struct log *msg = (struct log *)(log_buf + idx);
278 * A length == 0 record is the end of buffer marker. Wrap around and
279 * read the message at the start of the buffer.
282 return (struct log *)log_buf;
286 /* get next record; idx must point to valid msg */
287 static u32 log_next(u32 idx)
289 struct log *msg = (struct log *)(log_buf + idx);
291 /* length == 0 indicates the end of the buffer; wrap */
293 * A length == 0 record is the end of buffer marker. Wrap around and
294 * read the message at the start of the buffer as *this* one, and
295 * return the one after that.
298 msg = (struct log *)log_buf;
301 return idx + msg->len;
304 /* insert record into the buffer, discard old ones, update heads */
305 static void log_store(int facility, int level,
306 enum log_flags flags, u64 ts_nsec,
307 const char *dict, u16 dict_len,
308 const char *text, u16 text_len)
313 /* number of '\0' padding bytes to next message */
314 size = sizeof(struct log) + text_len + dict_len;
315 pad_len = (-size) & (LOG_ALIGN - 1);
318 while (log_first_seq < log_next_seq) {
321 if (log_next_idx > log_first_idx)
322 free = max(log_buf_len - log_next_idx, log_first_idx);
324 free = log_first_idx - log_next_idx;
326 if (free > size + sizeof(struct log))
329 /* drop old messages until we have enough contiuous space */
330 log_first_idx = log_next(log_first_idx);
334 if (log_next_idx + size + sizeof(struct log) >= log_buf_len) {
336 * This message + an additional empty header does not fit
337 * at the end of the buffer. Add an empty header with len == 0
338 * to signify a wrap around.
340 memset(log_buf + log_next_idx, 0, sizeof(struct log));
345 msg = (struct log *)(log_buf + log_next_idx);
346 memcpy(log_text(msg), text, text_len);
347 msg->text_len = text_len;
348 memcpy(log_dict(msg), dict, dict_len);
349 msg->dict_len = dict_len;
350 msg->facility = facility;
351 msg->level = level & 7;
352 msg->flags = flags & 0x1f;
354 msg->ts_nsec = ts_nsec;
356 msg->ts_nsec = local_clock();
357 memset(log_dict(msg) + dict_len, 0, pad_len);
358 msg->len = sizeof(struct log) + text_len + dict_len + pad_len;
361 log_next_idx += msg->len;
365 /* /dev/kmsg - userspace message inject/listen interface */
366 struct devkmsg_user {
374 static ssize_t devkmsg_writev(struct kiocb *iocb, const struct iovec *iv,
375 unsigned long count, loff_t pos)
379 int level = default_message_loglevel;
380 int facility = 1; /* LOG_USER */
381 size_t len = iov_length(iv, count);
384 if (len > LOG_LINE_MAX)
386 buf = kmalloc(len+1, GFP_KERNEL);
391 for (i = 0; i < count; i++) {
392 if (copy_from_user(line, iv[i].iov_base, iv[i].iov_len))
394 line += iv[i].iov_len;
398 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
399 * the decimal value represents 32bit, the lower 3 bit are the log
400 * level, the rest are the log facility.
402 * If no prefix or no userspace facility is specified, we
403 * enforce LOG_USER, to be able to reliably distinguish
404 * kernel-generated messages from userspace-injected ones.
407 if (line[0] == '<') {
410 i = simple_strtoul(line+1, &endp, 10);
411 if (endp && endp[0] == '>') {
422 printk_emit(facility, level, NULL, 0, "%s", line);
428 static ssize_t devkmsg_read(struct file *file, char __user *buf,
429 size_t count, loff_t *ppos)
431 struct devkmsg_user *user = file->private_data;
442 ret = mutex_lock_interruptible(&user->lock);
445 raw_spin_lock_irq(&logbuf_lock);
446 while (user->seq == log_next_seq) {
447 if (file->f_flags & O_NONBLOCK) {
449 raw_spin_unlock_irq(&logbuf_lock);
453 raw_spin_unlock_irq(&logbuf_lock);
454 ret = wait_event_interruptible(log_wait,
455 user->seq != log_next_seq);
458 raw_spin_lock_irq(&logbuf_lock);
461 if (user->seq < log_first_seq) {
462 /* our last seen message is gone, return error and reset */
463 user->idx = log_first_idx;
464 user->seq = log_first_seq;
466 raw_spin_unlock_irq(&logbuf_lock);
470 msg = log_from_idx(user->idx);
471 ts_usec = msg->ts_nsec;
472 do_div(ts_usec, 1000);
475 * If we couldn't merge continuation line fragments during the print,
476 * export the stored flags to allow an optional external merge of the
477 * records. Merging the records isn't always neccessarily correct, like
478 * when we hit a race during printing. In most cases though, it produces
479 * better readable output. 'c' in the record flags mark the first
480 * fragment of a line, '+' the following.
482 if (msg->flags & LOG_CONT && !(user->prev & LOG_CONT))
484 else if ((msg->flags & LOG_CONT) ||
485 ((user->prev & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
488 len = sprintf(user->buf, "%u,%llu,%llu,%c;",
489 (msg->facility << 3) | msg->level,
490 user->seq, ts_usec, cont);
491 user->prev = msg->flags;
493 /* escape non-printable characters */
494 for (i = 0; i < msg->text_len; i++) {
495 unsigned char c = log_text(msg)[i];
497 if (c < ' ' || c >= 127 || c == '\\')
498 len += sprintf(user->buf + len, "\\x%02x", c);
500 user->buf[len++] = c;
502 user->buf[len++] = '\n';
507 for (i = 0; i < msg->dict_len; i++) {
508 unsigned char c = log_dict(msg)[i];
511 user->buf[len++] = ' ';
516 user->buf[len++] = '\n';
521 if (c < ' ' || c >= 127 || c == '\\') {
522 len += sprintf(user->buf + len, "\\x%02x", c);
526 user->buf[len++] = c;
528 user->buf[len++] = '\n';
531 user->idx = log_next(user->idx);
533 raw_spin_unlock_irq(&logbuf_lock);
540 if (copy_to_user(buf, user->buf, len)) {
546 mutex_unlock(&user->lock);
550 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
552 struct devkmsg_user *user = file->private_data;
560 raw_spin_lock_irq(&logbuf_lock);
563 /* the first record */
564 user->idx = log_first_idx;
565 user->seq = log_first_seq;
569 * The first record after the last SYSLOG_ACTION_CLEAR,
570 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
571 * changes no global state, and does not clear anything.
573 user->idx = clear_idx;
574 user->seq = clear_seq;
577 /* after the last record */
578 user->idx = log_next_idx;
579 user->seq = log_next_seq;
584 raw_spin_unlock_irq(&logbuf_lock);
588 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
590 struct devkmsg_user *user = file->private_data;
594 return POLLERR|POLLNVAL;
596 poll_wait(file, &log_wait, wait);
598 raw_spin_lock_irq(&logbuf_lock);
599 if (user->seq < log_next_seq) {
600 /* return error when data has vanished underneath us */
601 if (user->seq < log_first_seq)
602 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
603 ret = POLLIN|POLLRDNORM;
605 raw_spin_unlock_irq(&logbuf_lock);
610 static int devkmsg_open(struct inode *inode, struct file *file)
612 struct devkmsg_user *user;
615 /* write-only does not need any file context */
616 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
619 err = security_syslog(SYSLOG_ACTION_READ_ALL);
623 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
627 mutex_init(&user->lock);
629 raw_spin_lock_irq(&logbuf_lock);
630 user->idx = log_first_idx;
631 user->seq = log_first_seq;
632 raw_spin_unlock_irq(&logbuf_lock);
634 file->private_data = user;
638 static int devkmsg_release(struct inode *inode, struct file *file)
640 struct devkmsg_user *user = file->private_data;
645 mutex_destroy(&user->lock);
650 const struct file_operations kmsg_fops = {
651 .open = devkmsg_open,
652 .read = devkmsg_read,
653 .aio_write = devkmsg_writev,
654 .llseek = devkmsg_llseek,
655 .poll = devkmsg_poll,
656 .release = devkmsg_release,
661 * This appends the listed symbols to /proc/vmcoreinfo
663 * /proc/vmcoreinfo is used by various utiilties, like crash and makedumpfile to
664 * obtain access to symbols that are otherwise very difficult to locate. These
665 * symbols are specifically used so that utilities can access and extract the
666 * dmesg log from a vmcore file after a crash.
668 void log_buf_kexec_setup(void)
670 VMCOREINFO_SYMBOL(log_buf);
671 VMCOREINFO_SYMBOL(log_buf_len);
672 VMCOREINFO_SYMBOL(log_first_idx);
673 VMCOREINFO_SYMBOL(log_next_idx);
675 * Export struct log size and field offsets. User space tools can
676 * parse it and detect any changes to structure down the line.
678 VMCOREINFO_STRUCT_SIZE(log);
679 VMCOREINFO_OFFSET(log, ts_nsec);
680 VMCOREINFO_OFFSET(log, len);
681 VMCOREINFO_OFFSET(log, text_len);
682 VMCOREINFO_OFFSET(log, dict_len);
686 /* requested log_buf_len from kernel cmdline */
687 static unsigned long __initdata new_log_buf_len;
689 /* save requested log_buf_len since it's too early to process it */
690 static int __init log_buf_len_setup(char *str)
692 unsigned size = memparse(str, &str);
695 size = roundup_pow_of_two(size);
696 if (size > log_buf_len)
697 new_log_buf_len = size;
701 early_param("log_buf_len", log_buf_len_setup);
703 void __init setup_log_buf(int early)
709 if (!new_log_buf_len)
715 mem = memblock_alloc(new_log_buf_len, PAGE_SIZE);
718 new_log_buf = __va(mem);
720 new_log_buf = alloc_bootmem_nopanic(new_log_buf_len);
723 if (unlikely(!new_log_buf)) {
724 pr_err("log_buf_len: %ld bytes not available\n",
729 raw_spin_lock_irqsave(&logbuf_lock, flags);
730 log_buf_len = new_log_buf_len;
731 log_buf = new_log_buf;
733 free = __LOG_BUF_LEN - log_next_idx;
734 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
735 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
737 pr_info("log_buf_len: %d\n", log_buf_len);
738 pr_info("early log buf free: %d(%d%%)\n",
739 free, (free * 100) / __LOG_BUF_LEN);
742 #ifdef CONFIG_BOOT_PRINTK_DELAY
744 static int boot_delay; /* msecs delay after each printk during bootup */
745 static unsigned long long loops_per_msec; /* based on boot_delay */
747 static int __init boot_delay_setup(char *str)
751 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
752 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
754 get_option(&str, &boot_delay);
755 if (boot_delay > 10 * 1000)
758 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
759 "HZ: %d, loops_per_msec: %llu\n",
760 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
763 __setup("boot_delay=", boot_delay_setup);
765 static void boot_delay_msec(void)
767 unsigned long long k;
768 unsigned long timeout;
770 if (boot_delay == 0 || system_state != SYSTEM_BOOTING)
773 k = (unsigned long long)loops_per_msec * boot_delay;
775 timeout = jiffies + msecs_to_jiffies(boot_delay);
780 * use (volatile) jiffies to prevent
781 * compiler reduction; loop termination via jiffies
782 * is secondary and may or may not happen.
784 if (time_after(jiffies, timeout))
786 touch_nmi_watchdog();
790 static inline void boot_delay_msec(void)
795 #ifdef CONFIG_SECURITY_DMESG_RESTRICT
796 int dmesg_restrict = 1;
801 static int syslog_action_restricted(int type)
805 /* Unless restricted, we allow "read all" and "get buffer size" for everybody */
806 return type != SYSLOG_ACTION_READ_ALL && type != SYSLOG_ACTION_SIZE_BUFFER;
809 static int check_syslog_permissions(int type, bool from_file)
812 * If this is from /proc/kmsg and we've already opened it, then we've
813 * already done the capabilities checks at open time.
815 if (from_file && type != SYSLOG_ACTION_OPEN)
818 if (syslog_action_restricted(type)) {
819 if (capable(CAP_SYSLOG))
821 /* For historical reasons, accept CAP_SYS_ADMIN too, with a warning */
822 if (capable(CAP_SYS_ADMIN)) {
823 printk_once(KERN_WARNING "%s (%d): "
824 "Attempt to access syslog with CAP_SYS_ADMIN "
825 "but no CAP_SYSLOG (deprecated).\n",
826 current->comm, task_pid_nr(current));
834 #if defined(CONFIG_PRINTK_TIME)
835 static bool printk_time = 1;
837 static bool printk_time;
839 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
841 static size_t print_time(u64 ts, char *buf)
843 unsigned long rem_nsec;
851 rem_nsec = do_div(ts, 1000000000);
852 return sprintf(buf, "[%5lu.%06lu] ",
853 (unsigned long)ts, rem_nsec / 1000);
856 static size_t print_prefix(const struct log *msg, bool syslog, char *buf)
859 unsigned int prefix = (msg->facility << 3) | msg->level;
863 len += sprintf(buf, "<%u>", prefix);
868 else if (prefix > 99)
875 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
879 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
880 bool syslog, char *buf, size_t size)
882 const char *text = log_text(msg);
883 size_t text_size = msg->text_len;
888 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
891 if (msg->flags & LOG_CONT) {
892 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
895 if (!(msg->flags & LOG_NEWLINE))
900 const char *next = memchr(text, '\n', text_size);
904 text_len = next - text;
906 text_size -= next - text;
908 text_len = text_size;
912 if (print_prefix(msg, syslog, NULL) +
913 text_len + 1 >= size - len)
917 len += print_prefix(msg, syslog, buf + len);
918 memcpy(buf + len, text, text_len);
923 /* SYSLOG_ACTION_* buffer size only calculation */
925 len += print_prefix(msg, syslog, NULL);
938 static int syslog_print(char __user *buf, int size)
944 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
952 raw_spin_lock_irq(&logbuf_lock);
953 if (syslog_seq < log_first_seq) {
954 /* messages are gone, move to first one */
955 syslog_seq = log_first_seq;
956 syslog_idx = log_first_idx;
960 if (syslog_seq == log_next_seq) {
961 raw_spin_unlock_irq(&logbuf_lock);
965 skip = syslog_partial;
966 msg = log_from_idx(syslog_idx);
967 n = msg_print_text(msg, syslog_prev, true, text,
968 LOG_LINE_MAX + PREFIX_MAX);
969 if (n - syslog_partial <= size) {
970 /* message fits into buffer, move forward */
971 syslog_idx = log_next(syslog_idx);
973 syslog_prev = msg->flags;
977 /* partial read(), remember position */
982 raw_spin_unlock_irq(&logbuf_lock);
987 if (copy_to_user(buf, text + skip, n)) {
1002 static int syslog_print_all(char __user *buf, int size, bool clear)
1007 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1011 raw_spin_lock_irq(&logbuf_lock);
1016 enum log_flags prev;
1018 if (clear_seq < log_first_seq) {
1019 /* messages are gone, move to first available one */
1020 clear_seq = log_first_seq;
1021 clear_idx = log_first_idx;
1025 * Find first record that fits, including all following records,
1026 * into the user-provided buffer for this dump.
1031 while (seq < log_next_seq) {
1032 struct log *msg = log_from_idx(idx);
1034 len += msg_print_text(msg, prev, true, NULL, 0);
1035 idx = log_next(idx);
1039 /* move first record forward until length fits into the buffer */
1043 while (len > size && seq < log_next_seq) {
1044 struct log *msg = log_from_idx(idx);
1046 len -= msg_print_text(msg, prev, true, NULL, 0);
1047 idx = log_next(idx);
1051 /* last message fitting into this dump */
1052 next_seq = log_next_seq;
1056 while (len >= 0 && seq < next_seq) {
1057 struct log *msg = log_from_idx(idx);
1060 textlen = msg_print_text(msg, prev, true, text,
1061 LOG_LINE_MAX + PREFIX_MAX);
1066 idx = log_next(idx);
1070 raw_spin_unlock_irq(&logbuf_lock);
1071 if (copy_to_user(buf + len, text, textlen))
1075 raw_spin_lock_irq(&logbuf_lock);
1077 if (seq < log_first_seq) {
1078 /* messages are gone, move to next one */
1079 seq = log_first_seq;
1080 idx = log_first_idx;
1087 clear_seq = log_next_seq;
1088 clear_idx = log_next_idx;
1090 raw_spin_unlock_irq(&logbuf_lock);
1096 int do_syslog(int type, char __user *buf, int len, bool from_file)
1099 static int saved_console_loglevel = -1;
1102 error = check_syslog_permissions(type, from_file);
1106 error = security_syslog(type);
1111 case SYSLOG_ACTION_CLOSE: /* Close log */
1113 case SYSLOG_ACTION_OPEN: /* Open log */
1115 case SYSLOG_ACTION_READ: /* Read from log */
1117 if (!buf || len < 0)
1122 if (!access_ok(VERIFY_WRITE, buf, len)) {
1126 error = wait_event_interruptible(log_wait,
1127 syslog_seq != log_next_seq);
1130 error = syslog_print(buf, len);
1132 /* Read/clear last kernel messages */
1133 case SYSLOG_ACTION_READ_CLEAR:
1136 /* Read last kernel messages */
1137 case SYSLOG_ACTION_READ_ALL:
1139 if (!buf || len < 0)
1144 if (!access_ok(VERIFY_WRITE, buf, len)) {
1148 error = syslog_print_all(buf, len, clear);
1150 /* Clear ring buffer */
1151 case SYSLOG_ACTION_CLEAR:
1152 syslog_print_all(NULL, 0, true);
1154 /* Disable logging to console */
1155 case SYSLOG_ACTION_CONSOLE_OFF:
1156 if (saved_console_loglevel == -1)
1157 saved_console_loglevel = console_loglevel;
1158 console_loglevel = minimum_console_loglevel;
1160 /* Enable logging to console */
1161 case SYSLOG_ACTION_CONSOLE_ON:
1162 if (saved_console_loglevel != -1) {
1163 console_loglevel = saved_console_loglevel;
1164 saved_console_loglevel = -1;
1167 /* Set level of messages printed to console */
1168 case SYSLOG_ACTION_CONSOLE_LEVEL:
1170 if (len < 1 || len > 8)
1172 if (len < minimum_console_loglevel)
1173 len = minimum_console_loglevel;
1174 console_loglevel = len;
1175 /* Implicitly re-enable logging to console */
1176 saved_console_loglevel = -1;
1179 /* Number of chars in the log buffer */
1180 case SYSLOG_ACTION_SIZE_UNREAD:
1181 raw_spin_lock_irq(&logbuf_lock);
1182 if (syslog_seq < log_first_seq) {
1183 /* messages are gone, move to first one */
1184 syslog_seq = log_first_seq;
1185 syslog_idx = log_first_idx;
1191 * Short-cut for poll(/"proc/kmsg") which simply checks
1192 * for pending data, not the size; return the count of
1193 * records, not the length.
1195 error = log_next_idx - syslog_idx;
1197 u64 seq = syslog_seq;
1198 u32 idx = syslog_idx;
1199 enum log_flags prev = syslog_prev;
1202 while (seq < log_next_seq) {
1203 struct log *msg = log_from_idx(idx);
1205 error += msg_print_text(msg, prev, true, NULL, 0);
1206 idx = log_next(idx);
1210 error -= syslog_partial;
1212 raw_spin_unlock_irq(&logbuf_lock);
1214 /* Size of the log buffer */
1215 case SYSLOG_ACTION_SIZE_BUFFER:
1216 error = log_buf_len;
1226 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1228 return do_syslog(type, buf, len, SYSLOG_FROM_CALL);
1231 static bool __read_mostly ignore_loglevel;
1233 static int __init ignore_loglevel_setup(char *str)
1235 ignore_loglevel = 1;
1236 printk(KERN_INFO "debug: ignoring loglevel setting.\n");
1241 early_param("ignore_loglevel", ignore_loglevel_setup);
1242 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1243 MODULE_PARM_DESC(ignore_loglevel, "ignore loglevel setting, to"
1244 "print all kernel messages to the console.");
1247 * Call the console drivers, asking them to write out
1248 * log_buf[start] to log_buf[end - 1].
1249 * The console_lock must be held.
1251 static void call_console_drivers(int level, const char *text, size_t len)
1253 struct console *con;
1255 trace_console(text, 0, len, len);
1257 if (level >= console_loglevel && !ignore_loglevel)
1259 if (!console_drivers)
1262 for_each_console(con) {
1263 if (exclusive_console && con != exclusive_console)
1265 if (!(con->flags & CON_ENABLED))
1269 if (!cpu_online(smp_processor_id()) &&
1270 !(con->flags & CON_ANYTIME))
1272 con->write(con, text, len);
1277 * Zap console related locks when oopsing. Only zap at most once
1278 * every 10 seconds, to leave time for slow consoles to print a
1281 static void zap_locks(void)
1283 static unsigned long oops_timestamp;
1285 if (time_after_eq(jiffies, oops_timestamp) &&
1286 !time_after(jiffies, oops_timestamp + 30 * HZ))
1289 oops_timestamp = jiffies;
1292 /* If a crash is occurring, make sure we can't deadlock */
1293 raw_spin_lock_init(&logbuf_lock);
1294 /* And make sure that we print immediately */
1295 sema_init(&console_sem, 1);
1298 /* Check if we have any console registered that can be called early in boot. */
1299 static int have_callable_console(void)
1301 struct console *con;
1303 for_each_console(con)
1304 if (con->flags & CON_ANYTIME)
1311 * Can we actually use the console at this time on this cpu?
1313 * Console drivers may assume that per-cpu resources have
1314 * been allocated. So unless they're explicitly marked as
1315 * being able to cope (CON_ANYTIME) don't call them until
1316 * this CPU is officially up.
1318 static inline int can_use_console(unsigned int cpu)
1320 return cpu_online(cpu) || have_callable_console();
1324 * Try to get console ownership to actually show the kernel
1325 * messages from a 'printk'. Return true (and with the
1326 * console_lock held, and 'console_locked' set) if it
1327 * is successful, false otherwise.
1329 * This gets called with the 'logbuf_lock' spinlock held and
1330 * interrupts disabled. It should return with 'lockbuf_lock'
1331 * released but interrupts still disabled.
1333 static int console_trylock_for_printk(unsigned int cpu)
1334 __releases(&logbuf_lock)
1336 int retval = 0, wake = 0;
1338 if (console_trylock()) {
1342 * If we can't use the console, we need to release
1343 * the console semaphore by hand to avoid flushing
1344 * the buffer. We need to hold the console semaphore
1345 * in order to do this test safely.
1347 if (!can_use_console(cpu)) {
1353 logbuf_cpu = UINT_MAX;
1356 raw_spin_unlock(&logbuf_lock);
1360 int printk_delay_msec __read_mostly;
1362 static inline void printk_delay(void)
1364 if (unlikely(printk_delay_msec)) {
1365 int m = printk_delay_msec;
1369 touch_nmi_watchdog();
1375 * Continuation lines are buffered, and not committed to the record buffer
1376 * until the line is complete, or a race forces it. The line fragments
1377 * though, are printed immediately to the consoles to ensure everything has
1378 * reached the console in case of a kernel crash.
1380 static struct cont {
1381 char buf[LOG_LINE_MAX];
1382 size_t len; /* length == 0 means unused buffer */
1383 size_t cons; /* bytes written to console */
1384 struct task_struct *owner; /* task of first print*/
1385 u64 ts_nsec; /* time of first print */
1386 u8 level; /* log level of first message */
1387 u8 facility; /* log level of first message */
1388 enum log_flags flags; /* prefix, newline flags */
1389 bool flushed:1; /* buffer sealed and committed */
1392 static void cont_flush(enum log_flags flags)
1401 * If a fragment of this line was directly flushed to the
1402 * console; wait for the console to pick up the rest of the
1403 * line. LOG_NOCONS suppresses a duplicated output.
1405 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1406 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1408 cont.flushed = true;
1411 * If no fragment of this line ever reached the console,
1412 * just submit it to the store and free the buffer.
1414 log_store(cont.facility, cont.level, flags, 0,
1415 NULL, 0, cont.buf, cont.len);
1420 static bool cont_add(int facility, int level, const char *text, size_t len)
1422 if (cont.len && cont.flushed)
1425 if (cont.len + len > sizeof(cont.buf)) {
1426 /* the line gets too long, split it up in separate records */
1427 cont_flush(LOG_CONT);
1432 cont.facility = facility;
1434 cont.owner = current;
1435 cont.ts_nsec = local_clock();
1438 cont.flushed = false;
1441 memcpy(cont.buf + cont.len, text, len);
1444 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1445 cont_flush(LOG_CONT);
1450 static size_t cont_print_text(char *text, size_t size)
1455 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1456 textlen += print_time(cont.ts_nsec, text);
1460 len = cont.len - cont.cons;
1464 memcpy(text + textlen, cont.buf + cont.cons, len);
1466 cont.cons = cont.len;
1470 if (cont.flags & LOG_NEWLINE)
1471 text[textlen++] = '\n';
1472 /* got everything, release buffer */
1478 asmlinkage int vprintk_emit(int facility, int level,
1479 const char *dict, size_t dictlen,
1480 const char *fmt, va_list args)
1482 static int recursion_bug;
1483 static char textbuf[LOG_LINE_MAX];
1484 char *text = textbuf;
1486 enum log_flags lflags = 0;
1487 unsigned long flags;
1489 int printed_len = 0;
1494 /* This stops the holder of console_sem just where we want him */
1495 local_irq_save(flags);
1496 this_cpu = smp_processor_id();
1499 * Ouch, printk recursed into itself!
1501 if (unlikely(logbuf_cpu == this_cpu)) {
1503 * If a crash is occurring during printk() on this CPU,
1504 * then try to get the crash message out but make sure
1505 * we can't deadlock. Otherwise just return to avoid the
1506 * recursion and return - but flag the recursion so that
1507 * it can be printed at the next appropriate moment:
1509 if (!oops_in_progress && !lockdep_recursing(current)) {
1511 goto out_restore_irqs;
1517 raw_spin_lock(&logbuf_lock);
1518 logbuf_cpu = this_cpu;
1520 if (recursion_bug) {
1521 static const char recursion_msg[] =
1522 "BUG: recent printk recursion!";
1525 printed_len += strlen(recursion_msg);
1526 /* emit KERN_CRIT message */
1527 log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1528 NULL, 0, recursion_msg, printed_len);
1532 * The printf needs to come first; we need the syslog
1533 * prefix which might be passed-in as a parameter.
1535 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1537 /* mark and strip a trailing newline */
1538 if (text_len && text[text_len-1] == '\n') {
1540 lflags |= LOG_NEWLINE;
1543 /* strip syslog prefix and extract log level or control flags */
1544 if (text[0] == '<' && text[1] && text[2] == '>') {
1548 level = text[1] - '0';
1549 case 'd': /* KERN_DEFAULT */
1550 lflags |= LOG_PREFIX;
1551 case 'c': /* KERN_CONT */
1558 level = default_message_loglevel;
1561 lflags |= LOG_PREFIX|LOG_NEWLINE;
1563 if (!(lflags & LOG_NEWLINE)) {
1565 * Flush the conflicting buffer. An earlier newline was missing,
1566 * or another task also prints continuation lines.
1568 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1569 cont_flush(LOG_NEWLINE);
1571 /* buffer line if possible, otherwise store it right away */
1572 if (!cont_add(facility, level, text, text_len))
1573 log_store(facility, level, lflags | LOG_CONT, 0,
1574 dict, dictlen, text, text_len);
1576 bool stored = false;
1579 * If an earlier newline was missing and it was the same task,
1580 * either merge it with the current buffer and flush, or if
1581 * there was a race with interrupts (prefix == true) then just
1582 * flush it out and store this line separately.
1584 if (cont.len && cont.owner == current) {
1585 if (!(lflags & LOG_PREFIX))
1586 stored = cont_add(facility, level, text, text_len);
1587 cont_flush(LOG_NEWLINE);
1591 log_store(facility, level, lflags, 0,
1592 dict, dictlen, text, text_len);
1594 printed_len += text_len;
1597 * Try to acquire and then immediately release the console semaphore.
1598 * The release will print out buffers and wake up /dev/kmsg and syslog()
1601 * The console_trylock_for_printk() function will release 'logbuf_lock'
1602 * regardless of whether it actually gets the console semaphore or not.
1604 if (console_trylock_for_printk(this_cpu))
1609 local_irq_restore(flags);
1613 EXPORT_SYMBOL(vprintk_emit);
1615 asmlinkage int vprintk(const char *fmt, va_list args)
1617 return vprintk_emit(0, -1, NULL, 0, fmt, args);
1619 EXPORT_SYMBOL(vprintk);
1621 asmlinkage int printk_emit(int facility, int level,
1622 const char *dict, size_t dictlen,
1623 const char *fmt, ...)
1628 va_start(args, fmt);
1629 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1634 EXPORT_SYMBOL(printk_emit);
1637 * printk - print a kernel message
1638 * @fmt: format string
1640 * This is printk(). It can be called from any context. We want it to work.
1642 * We try to grab the console_lock. If we succeed, it's easy - we log the
1643 * output and call the console drivers. If we fail to get the semaphore, we
1644 * place the output into the log buffer and return. The current holder of
1645 * the console_sem will notice the new output in console_unlock(); and will
1646 * send it to the consoles before releasing the lock.
1648 * One effect of this deferred printing is that code which calls printk() and
1649 * then changes console_loglevel may break. This is because console_loglevel
1650 * is inspected when the actual printing occurs.
1655 * See the vsnprintf() documentation for format string extensions over C99.
1657 asmlinkage int printk(const char *fmt, ...)
1662 #ifdef CONFIG_KGDB_KDB
1663 if (unlikely(kdb_trap_printk)) {
1664 va_start(args, fmt);
1665 r = vkdb_printf(fmt, args);
1670 va_start(args, fmt);
1671 r = vprintk_emit(0, -1, NULL, 0, fmt, args);
1676 EXPORT_SYMBOL(printk);
1678 #else /* CONFIG_PRINTK */
1680 #define LOG_LINE_MAX 0
1681 #define PREFIX_MAX 0
1682 #define LOG_LINE_MAX 0
1683 static u64 syslog_seq;
1684 static u32 syslog_idx;
1685 static u64 console_seq;
1686 static u32 console_idx;
1687 static enum log_flags syslog_prev;
1688 static u64 log_first_seq;
1689 static u32 log_first_idx;
1690 static u64 log_next_seq;
1691 static enum log_flags console_prev;
1692 static struct cont {
1698 static struct log *log_from_idx(u32 idx) { return NULL; }
1699 static u32 log_next(u32 idx) { return 0; }
1700 static void call_console_drivers(int level, const char *text, size_t len) {}
1701 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
1702 bool syslog, char *buf, size_t size) { return 0; }
1703 static size_t cont_print_text(char *text, size_t size) { return 0; }
1705 #endif /* CONFIG_PRINTK */
1707 static int __add_preferred_console(char *name, int idx, char *options,
1710 struct console_cmdline *c;
1714 * See if this tty is not yet registered, and
1715 * if we have a slot free.
1717 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1718 if (strcmp(console_cmdline[i].name, name) == 0 &&
1719 console_cmdline[i].index == idx) {
1721 selected_console = i;
1724 if (i == MAX_CMDLINECONSOLES)
1727 selected_console = i;
1728 c = &console_cmdline[i];
1729 strlcpy(c->name, name, sizeof(c->name));
1730 c->options = options;
1731 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1732 c->brl_options = brl_options;
1738 * Set up a list of consoles. Called from init/main.c
1740 static int __init console_setup(char *str)
1742 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for index */
1743 char *s, *options, *brl_options = NULL;
1746 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1747 if (!memcmp(str, "brl,", 4)) {
1750 } else if (!memcmp(str, "brl=", 4)) {
1751 brl_options = str + 4;
1752 str = strchr(brl_options, ',');
1754 printk(KERN_ERR "need port name after brl=\n");
1762 * Decode str into name, index, options.
1764 if (str[0] >= '0' && str[0] <= '9') {
1765 strcpy(buf, "ttyS");
1766 strncpy(buf + 4, str, sizeof(buf) - 5);
1768 strncpy(buf, str, sizeof(buf) - 1);
1770 buf[sizeof(buf) - 1] = 0;
1771 if ((options = strchr(str, ',')) != NULL)
1774 if (!strcmp(str, "ttya"))
1775 strcpy(buf, "ttyS0");
1776 if (!strcmp(str, "ttyb"))
1777 strcpy(buf, "ttyS1");
1779 for (s = buf; *s; s++)
1780 if ((*s >= '0' && *s <= '9') || *s == ',')
1782 idx = simple_strtoul(s, NULL, 10);
1785 __add_preferred_console(buf, idx, options, brl_options);
1786 console_set_on_cmdline = 1;
1789 __setup("console=", console_setup);
1792 * add_preferred_console - add a device to the list of preferred consoles.
1793 * @name: device name
1794 * @idx: device index
1795 * @options: options for this console
1797 * The last preferred console added will be used for kernel messages
1798 * and stdin/out/err for init. Normally this is used by console_setup
1799 * above to handle user-supplied console arguments; however it can also
1800 * be used by arch-specific code either to override the user or more
1801 * commonly to provide a default console (ie from PROM variables) when
1802 * the user has not supplied one.
1804 int add_preferred_console(char *name, int idx, char *options)
1806 return __add_preferred_console(name, idx, options, NULL);
1809 int update_console_cmdline(char *name, int idx, char *name_new, int idx_new, char *options)
1811 struct console_cmdline *c;
1814 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1815 if (strcmp(console_cmdline[i].name, name) == 0 &&
1816 console_cmdline[i].index == idx) {
1817 c = &console_cmdline[i];
1818 strlcpy(c->name, name_new, sizeof(c->name));
1819 c->name[sizeof(c->name) - 1] = 0;
1820 c->options = options;
1828 bool console_suspend_enabled = 1;
1829 EXPORT_SYMBOL(console_suspend_enabled);
1831 static int __init console_suspend_disable(char *str)
1833 console_suspend_enabled = 0;
1836 __setup("no_console_suspend", console_suspend_disable);
1837 module_param_named(console_suspend, console_suspend_enabled,
1838 bool, S_IRUGO | S_IWUSR);
1839 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
1840 " and hibernate operations");
1843 * suspend_console - suspend the console subsystem
1845 * This disables printk() while we go into suspend states
1847 void suspend_console(void)
1849 if (!console_suspend_enabled)
1851 printk("Suspending console(s) (use no_console_suspend to debug)\n");
1853 console_suspended = 1;
1857 void resume_console(void)
1859 if (!console_suspend_enabled)
1862 console_suspended = 0;
1867 * console_cpu_notify - print deferred console messages after CPU hotplug
1868 * @self: notifier struct
1869 * @action: CPU hotplug event
1872 * If printk() is called from a CPU that is not online yet, the messages
1873 * will be spooled but will not show up on the console. This function is
1874 * called when a new CPU comes online (or fails to come up), and ensures
1875 * that any such output gets printed.
1877 static int __cpuinit console_cpu_notify(struct notifier_block *self,
1878 unsigned long action, void *hcpu)
1884 case CPU_DOWN_FAILED:
1885 case CPU_UP_CANCELED:
1893 * console_lock - lock the console system for exclusive use.
1895 * Acquires a lock which guarantees that the caller has
1896 * exclusive access to the console system and the console_drivers list.
1898 * Can sleep, returns nothing.
1900 void console_lock(void)
1902 BUG_ON(in_interrupt());
1904 if (console_suspended)
1907 console_may_schedule = 1;
1909 EXPORT_SYMBOL(console_lock);
1912 * console_trylock - try to lock the console system for exclusive use.
1914 * Tried to acquire a lock which guarantees that the caller has
1915 * exclusive access to the console system and the console_drivers list.
1917 * returns 1 on success, and 0 on failure to acquire the lock.
1919 int console_trylock(void)
1921 if (down_trylock(&console_sem))
1923 if (console_suspended) {
1928 console_may_schedule = 0;
1931 EXPORT_SYMBOL(console_trylock);
1933 int is_console_locked(void)
1935 return console_locked;
1939 * Delayed printk version, for scheduler-internal messages:
1941 #define PRINTK_BUF_SIZE 512
1943 #define PRINTK_PENDING_WAKEUP 0x01
1944 #define PRINTK_PENDING_SCHED 0x02
1946 static DEFINE_PER_CPU(int, printk_pending);
1947 static DEFINE_PER_CPU(char [PRINTK_BUF_SIZE], printk_sched_buf);
1949 void printk_tick(void)
1951 if (__this_cpu_read(printk_pending)) {
1952 int pending = __this_cpu_xchg(printk_pending, 0);
1953 if (pending & PRINTK_PENDING_SCHED) {
1954 char *buf = __get_cpu_var(printk_sched_buf);
1955 printk(KERN_WARNING "[sched_delayed] %s", buf);
1957 if (pending & PRINTK_PENDING_WAKEUP)
1958 wake_up_interruptible(&log_wait);
1962 int printk_needs_cpu(int cpu)
1964 if (cpu_is_offline(cpu))
1966 return __this_cpu_read(printk_pending);
1969 void wake_up_klogd(void)
1971 if (waitqueue_active(&log_wait))
1972 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
1975 static void console_cont_flush(char *text, size_t size)
1977 unsigned long flags;
1980 raw_spin_lock_irqsave(&logbuf_lock, flags);
1986 * We still queue earlier records, likely because the console was
1987 * busy. The earlier ones need to be printed before this one, we
1988 * did not flush any fragment so far, so just let it queue up.
1990 if (console_seq < log_next_seq && !cont.cons)
1993 len = cont_print_text(text, size);
1994 raw_spin_unlock(&logbuf_lock);
1995 stop_critical_timings();
1996 call_console_drivers(cont.level, text, len);
1997 start_critical_timings();
1998 local_irq_restore(flags);
2001 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2005 * console_unlock - unlock the console system
2007 * Releases the console_lock which the caller holds on the console system
2008 * and the console driver list.
2010 * While the console_lock was held, console output may have been buffered
2011 * by printk(). If this is the case, console_unlock(); emits
2012 * the output prior to releasing the lock.
2014 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2016 * console_unlock(); may be called from any context.
2018 void console_unlock(void)
2020 static char text[LOG_LINE_MAX + PREFIX_MAX];
2021 static u64 seen_seq;
2022 unsigned long flags;
2023 bool wake_klogd = false;
2026 if (console_suspended) {
2031 console_may_schedule = 0;
2033 /* flush buffered message fragment immediately to console */
2034 console_cont_flush(text, sizeof(text));
2041 raw_spin_lock_irqsave(&logbuf_lock, flags);
2042 if (seen_seq != log_next_seq) {
2044 seen_seq = log_next_seq;
2047 if (console_seq < log_first_seq) {
2048 /* messages are gone, move to first one */
2049 console_seq = log_first_seq;
2050 console_idx = log_first_idx;
2054 if (console_seq == log_next_seq)
2057 msg = log_from_idx(console_idx);
2058 if (msg->flags & LOG_NOCONS) {
2060 * Skip record we have buffered and already printed
2061 * directly to the console when we received it.
2063 console_idx = log_next(console_idx);
2066 * We will get here again when we register a new
2067 * CON_PRINTBUFFER console. Clear the flag so we
2068 * will properly dump everything later.
2070 msg->flags &= ~LOG_NOCONS;
2071 console_prev = msg->flags;
2076 len = msg_print_text(msg, console_prev, false,
2077 text, sizeof(text));
2078 console_idx = log_next(console_idx);
2080 console_prev = msg->flags;
2081 raw_spin_unlock(&logbuf_lock);
2083 stop_critical_timings(); /* don't trace print latency */
2084 call_console_drivers(level, text, len);
2085 start_critical_timings();
2086 local_irq_restore(flags);
2090 /* Release the exclusive_console once it is used */
2091 if (unlikely(exclusive_console))
2092 exclusive_console = NULL;
2094 raw_spin_unlock(&logbuf_lock);
2099 * Someone could have filled up the buffer again, so re-check if there's
2100 * something to flush. In case we cannot trylock the console_sem again,
2101 * there's a new owner and the console_unlock() from them will do the
2102 * flush, no worries.
2104 raw_spin_lock(&logbuf_lock);
2105 retry = console_seq != log_next_seq;
2106 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2108 if (retry && console_trylock())
2114 EXPORT_SYMBOL(console_unlock);
2117 * console_conditional_schedule - yield the CPU if required
2119 * If the console code is currently allowed to sleep, and
2120 * if this CPU should yield the CPU to another task, do
2123 * Must be called within console_lock();.
2125 void __sched console_conditional_schedule(void)
2127 if (console_may_schedule)
2130 EXPORT_SYMBOL(console_conditional_schedule);
2132 void console_unblank(void)
2137 * console_unblank can no longer be called in interrupt context unless
2138 * oops_in_progress is set to 1..
2140 if (oops_in_progress) {
2141 if (down_trylock(&console_sem) != 0)
2147 console_may_schedule = 0;
2149 if ((c->flags & CON_ENABLED) && c->unblank)
2155 * Return the console tty driver structure and its associated index
2157 struct tty_driver *console_device(int *index)
2160 struct tty_driver *driver = NULL;
2163 for_each_console(c) {
2166 driver = c->device(c, index);
2175 * Prevent further output on the passed console device so that (for example)
2176 * serial drivers can disable console output before suspending a port, and can
2177 * re-enable output afterwards.
2179 void console_stop(struct console *console)
2182 console->flags &= ~CON_ENABLED;
2185 EXPORT_SYMBOL(console_stop);
2187 void console_start(struct console *console)
2190 console->flags |= CON_ENABLED;
2193 EXPORT_SYMBOL(console_start);
2195 static int __read_mostly keep_bootcon;
2197 static int __init keep_bootcon_setup(char *str)
2200 printk(KERN_INFO "debug: skip boot console de-registration.\n");
2205 early_param("keep_bootcon", keep_bootcon_setup);
2208 * The console driver calls this routine during kernel initialization
2209 * to register the console printing procedure with printk() and to
2210 * print any messages that were printed by the kernel before the
2211 * console driver was initialized.
2213 * This can happen pretty early during the boot process (because of
2214 * early_printk) - sometimes before setup_arch() completes - be careful
2215 * of what kernel features are used - they may not be initialised yet.
2217 * There are two types of consoles - bootconsoles (early_printk) and
2218 * "real" consoles (everything which is not a bootconsole) which are
2219 * handled differently.
2220 * - Any number of bootconsoles can be registered at any time.
2221 * - As soon as a "real" console is registered, all bootconsoles
2222 * will be unregistered automatically.
2223 * - Once a "real" console is registered, any attempt to register a
2224 * bootconsoles will be rejected
2226 void register_console(struct console *newcon)
2229 unsigned long flags;
2230 struct console *bcon = NULL;
2233 * before we register a new CON_BOOT console, make sure we don't
2234 * already have a valid console
2236 if (console_drivers && newcon->flags & CON_BOOT) {
2237 /* find the last or real console */
2238 for_each_console(bcon) {
2239 if (!(bcon->flags & CON_BOOT)) {
2240 printk(KERN_INFO "Too late to register bootconsole %s%d\n",
2241 newcon->name, newcon->index);
2247 if (console_drivers && console_drivers->flags & CON_BOOT)
2248 bcon = console_drivers;
2250 if (preferred_console < 0 || bcon || !console_drivers)
2251 preferred_console = selected_console;
2253 if (newcon->early_setup)
2254 newcon->early_setup();
2257 * See if we want to use this console driver. If we
2258 * didn't select a console we take the first one
2259 * that registers here.
2261 if (preferred_console < 0) {
2262 if (newcon->index < 0)
2264 if (newcon->setup == NULL ||
2265 newcon->setup(newcon, NULL) == 0) {
2266 newcon->flags |= CON_ENABLED;
2267 if (newcon->device) {
2268 newcon->flags |= CON_CONSDEV;
2269 preferred_console = 0;
2275 * See if this console matches one we selected on
2278 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0];
2280 if (strcmp(console_cmdline[i].name, newcon->name) != 0)
2282 if (newcon->index >= 0 &&
2283 newcon->index != console_cmdline[i].index)
2285 if (newcon->index < 0)
2286 newcon->index = console_cmdline[i].index;
2287 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2288 if (console_cmdline[i].brl_options) {
2289 newcon->flags |= CON_BRL;
2290 braille_register_console(newcon,
2291 console_cmdline[i].index,
2292 console_cmdline[i].options,
2293 console_cmdline[i].brl_options);
2297 if (newcon->setup &&
2298 newcon->setup(newcon, console_cmdline[i].options) != 0)
2300 newcon->flags |= CON_ENABLED;
2301 newcon->index = console_cmdline[i].index;
2302 if (i == selected_console) {
2303 newcon->flags |= CON_CONSDEV;
2304 preferred_console = selected_console;
2309 if (!(newcon->flags & CON_ENABLED))
2313 * If we have a bootconsole, and are switching to a real console,
2314 * don't print everything out again, since when the boot console, and
2315 * the real console are the same physical device, it's annoying to
2316 * see the beginning boot messages twice
2318 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2319 newcon->flags &= ~CON_PRINTBUFFER;
2322 * Put this console in the list - keep the
2323 * preferred driver at the head of the list.
2326 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2327 newcon->next = console_drivers;
2328 console_drivers = newcon;
2330 newcon->next->flags &= ~CON_CONSDEV;
2332 newcon->next = console_drivers->next;
2333 console_drivers->next = newcon;
2335 if (newcon->flags & CON_PRINTBUFFER) {
2337 * console_unlock(); will print out the buffered messages
2340 raw_spin_lock_irqsave(&logbuf_lock, flags);
2341 console_seq = syslog_seq;
2342 console_idx = syslog_idx;
2343 console_prev = syslog_prev;
2344 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2346 * We're about to replay the log buffer. Only do this to the
2347 * just-registered console to avoid excessive message spam to
2348 * the already-registered consoles.
2350 exclusive_console = newcon;
2353 console_sysfs_notify();
2356 * By unregistering the bootconsoles after we enable the real console
2357 * we get the "console xxx enabled" message on all the consoles -
2358 * boot consoles, real consoles, etc - this is to ensure that end
2359 * users know there might be something in the kernel's log buffer that
2360 * went to the bootconsole (that they do not see on the real console)
2363 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2365 /* we need to iterate through twice, to make sure we print
2366 * everything out, before we unregister the console(s)
2368 printk(KERN_INFO "console [%s%d] enabled, bootconsole disabled\n",
2369 newcon->name, newcon->index);
2370 for_each_console(bcon)
2371 if (bcon->flags & CON_BOOT)
2372 unregister_console(bcon);
2374 printk(KERN_INFO "%sconsole [%s%d] enabled\n",
2375 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2376 newcon->name, newcon->index);
2379 EXPORT_SYMBOL(register_console);
2381 int unregister_console(struct console *console)
2383 struct console *a, *b;
2386 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2387 if (console->flags & CON_BRL)
2388 return braille_unregister_console(console);
2392 if (console_drivers == console) {
2393 console_drivers=console->next;
2395 } else if (console_drivers) {
2396 for (a=console_drivers->next, b=console_drivers ;
2397 a; b=a, a=b->next) {
2407 * If this isn't the last console and it has CON_CONSDEV set, we
2408 * need to set it on the next preferred console.
2410 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2411 console_drivers->flags |= CON_CONSDEV;
2414 console_sysfs_notify();
2417 EXPORT_SYMBOL(unregister_console);
2419 static int __init printk_late_init(void)
2421 struct console *con;
2423 for_each_console(con) {
2424 if (!keep_bootcon && con->flags & CON_BOOT) {
2425 printk(KERN_INFO "turn off boot console %s%d\n",
2426 con->name, con->index);
2427 unregister_console(con);
2430 hotcpu_notifier(console_cpu_notify, 0);
2433 late_initcall(printk_late_init);
2435 #if defined CONFIG_PRINTK
2437 int printk_sched(const char *fmt, ...)
2439 unsigned long flags;
2444 local_irq_save(flags);
2445 buf = __get_cpu_var(printk_sched_buf);
2447 va_start(args, fmt);
2448 r = vsnprintf(buf, PRINTK_BUF_SIZE, fmt, args);
2451 __this_cpu_or(printk_pending, PRINTK_PENDING_SCHED);
2452 local_irq_restore(flags);
2458 * printk rate limiting, lifted from the networking subsystem.
2460 * This enforces a rate limit: not more than 10 kernel messages
2461 * every 5s to make a denial-of-service attack impossible.
2463 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2465 int __printk_ratelimit(const char *func)
2467 return ___ratelimit(&printk_ratelimit_state, func);
2469 EXPORT_SYMBOL(__printk_ratelimit);
2472 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2473 * @caller_jiffies: pointer to caller's state
2474 * @interval_msecs: minimum interval between prints
2476 * printk_timed_ratelimit() returns true if more than @interval_msecs
2477 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2480 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2481 unsigned int interval_msecs)
2483 if (*caller_jiffies == 0
2484 || !time_in_range(jiffies, *caller_jiffies,
2486 + msecs_to_jiffies(interval_msecs))) {
2487 *caller_jiffies = jiffies;
2492 EXPORT_SYMBOL(printk_timed_ratelimit);
2494 static DEFINE_SPINLOCK(dump_list_lock);
2495 static LIST_HEAD(dump_list);
2498 * kmsg_dump_register - register a kernel log dumper.
2499 * @dumper: pointer to the kmsg_dumper structure
2501 * Adds a kernel log dumper to the system. The dump callback in the
2502 * structure will be called when the kernel oopses or panics and must be
2503 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2505 int kmsg_dump_register(struct kmsg_dumper *dumper)
2507 unsigned long flags;
2510 /* The dump callback needs to be set */
2514 spin_lock_irqsave(&dump_list_lock, flags);
2515 /* Don't allow registering multiple times */
2516 if (!dumper->registered) {
2517 dumper->registered = 1;
2518 list_add_tail_rcu(&dumper->list, &dump_list);
2521 spin_unlock_irqrestore(&dump_list_lock, flags);
2525 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2528 * kmsg_dump_unregister - unregister a kmsg dumper.
2529 * @dumper: pointer to the kmsg_dumper structure
2531 * Removes a dump device from the system. Returns zero on success and
2532 * %-EINVAL otherwise.
2534 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2536 unsigned long flags;
2539 spin_lock_irqsave(&dump_list_lock, flags);
2540 if (dumper->registered) {
2541 dumper->registered = 0;
2542 list_del_rcu(&dumper->list);
2545 spin_unlock_irqrestore(&dump_list_lock, flags);
2550 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2552 static bool always_kmsg_dump;
2553 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2556 * kmsg_dump - dump kernel log to kernel message dumpers.
2557 * @reason: the reason (oops, panic etc) for dumping
2559 * Call each of the registered dumper's dump() callback, which can
2560 * retrieve the kmsg records with kmsg_dump_get_line() or
2561 * kmsg_dump_get_buffer().
2563 void kmsg_dump(enum kmsg_dump_reason reason)
2565 struct kmsg_dumper *dumper;
2566 unsigned long flags;
2568 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2572 list_for_each_entry_rcu(dumper, &dump_list, list) {
2573 if (dumper->max_reason && reason > dumper->max_reason)
2576 /* initialize iterator with data about the stored records */
2577 dumper->active = true;
2579 raw_spin_lock_irqsave(&logbuf_lock, flags);
2580 dumper->cur_seq = clear_seq;
2581 dumper->cur_idx = clear_idx;
2582 dumper->next_seq = log_next_seq;
2583 dumper->next_idx = log_next_idx;
2584 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2586 /* invoke dumper which will iterate over records */
2587 dumper->dump(dumper, reason);
2589 /* reset iterator */
2590 dumper->active = false;
2596 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2597 * @dumper: registered kmsg dumper
2598 * @syslog: include the "<4>" prefixes
2599 * @line: buffer to copy the line to
2600 * @size: maximum size of the buffer
2601 * @len: length of line placed into buffer
2603 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2604 * record, and copy one record into the provided buffer.
2606 * Consecutive calls will return the next available record moving
2607 * towards the end of the buffer with the youngest messages.
2609 * A return value of FALSE indicates that there are no more records to
2612 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2614 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2615 char *line, size_t size, size_t *len)
2621 if (!dumper->active)
2624 if (dumper->cur_seq < log_first_seq) {
2625 /* messages are gone, move to first available one */
2626 dumper->cur_seq = log_first_seq;
2627 dumper->cur_idx = log_first_idx;
2631 if (dumper->cur_seq >= log_next_seq)
2634 msg = log_from_idx(dumper->cur_idx);
2635 l = msg_print_text(msg, 0, syslog, line, size);
2637 dumper->cur_idx = log_next(dumper->cur_idx);
2647 * kmsg_dump_get_line - retrieve one kmsg log line
2648 * @dumper: registered kmsg dumper
2649 * @syslog: include the "<4>" prefixes
2650 * @line: buffer to copy the line to
2651 * @size: maximum size of the buffer
2652 * @len: length of line placed into buffer
2654 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2655 * record, and copy one record into the provided buffer.
2657 * Consecutive calls will return the next available record moving
2658 * towards the end of the buffer with the youngest messages.
2660 * A return value of FALSE indicates that there are no more records to
2663 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2664 char *line, size_t size, size_t *len)
2666 unsigned long flags;
2669 raw_spin_lock_irqsave(&logbuf_lock, flags);
2670 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2671 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2675 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2678 * kmsg_dump_get_buffer - copy kmsg log lines
2679 * @dumper: registered kmsg dumper
2680 * @syslog: include the "<4>" prefixes
2681 * @buf: buffer to copy the line to
2682 * @size: maximum size of the buffer
2683 * @len: length of line placed into buffer
2685 * Start at the end of the kmsg buffer and fill the provided buffer
2686 * with as many of the the *youngest* kmsg records that fit into it.
2687 * If the buffer is large enough, all available kmsg records will be
2688 * copied with a single call.
2690 * Consecutive calls will fill the buffer with the next block of
2691 * available older records, not including the earlier retrieved ones.
2693 * A return value of FALSE indicates that there are no more records to
2696 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2697 char *buf, size_t size, size_t *len)
2699 unsigned long flags;
2704 enum log_flags prev;
2708 if (!dumper->active)
2711 raw_spin_lock_irqsave(&logbuf_lock, flags);
2712 if (dumper->cur_seq < log_first_seq) {
2713 /* messages are gone, move to first available one */
2714 dumper->cur_seq = log_first_seq;
2715 dumper->cur_idx = log_first_idx;
2719 if (dumper->cur_seq >= dumper->next_seq) {
2720 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2724 /* calculate length of entire buffer */
2725 seq = dumper->cur_seq;
2726 idx = dumper->cur_idx;
2728 while (seq < dumper->next_seq) {
2729 struct log *msg = log_from_idx(idx);
2731 l += msg_print_text(msg, prev, true, NULL, 0);
2732 idx = log_next(idx);
2737 /* move first record forward until length fits into the buffer */
2738 seq = dumper->cur_seq;
2739 idx = dumper->cur_idx;
2741 while (l > size && seq < dumper->next_seq) {
2742 struct log *msg = log_from_idx(idx);
2744 l -= msg_print_text(msg, prev, true, NULL, 0);
2745 idx = log_next(idx);
2750 /* last message in next interation */
2756 while (seq < dumper->next_seq) {
2757 struct log *msg = log_from_idx(idx);
2759 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
2760 idx = log_next(idx);
2765 dumper->next_seq = next_seq;
2766 dumper->next_idx = next_idx;
2768 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2774 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
2777 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
2778 * @dumper: registered kmsg dumper
2780 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2781 * kmsg_dump_get_buffer() can be called again and used multiple
2782 * times within the same dumper.dump() callback.
2784 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
2786 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
2788 dumper->cur_seq = clear_seq;
2789 dumper->cur_idx = clear_idx;
2790 dumper->next_seq = log_next_seq;
2791 dumper->next_idx = log_next_idx;
2795 * kmsg_dump_rewind - reset the interator
2796 * @dumper: registered kmsg dumper
2798 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2799 * kmsg_dump_get_buffer() can be called again and used multiple
2800 * times within the same dumper.dump() callback.
2802 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
2804 unsigned long flags;
2806 raw_spin_lock_irqsave(&logbuf_lock, flags);
2807 kmsg_dump_rewind_nolock(dumper);
2808 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2810 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);