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>
45 #include <linux/irq_work.h>
47 #include <asm/uaccess.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/printk.h>
53 * Architectures can override it:
55 void asmlinkage __attribute__((weak)) early_printk(const char *fmt, ...)
59 /* printk's without a loglevel use this.. */
60 #define DEFAULT_MESSAGE_LOGLEVEL CONFIG_DEFAULT_MESSAGE_LOGLEVEL
62 /* We show everything that is MORE important than this.. */
63 #define MINIMUM_CONSOLE_LOGLEVEL 1 /* Minimum loglevel we let people use */
64 #define DEFAULT_CONSOLE_LOGLEVEL 7 /* anything MORE serious than KERN_DEBUG */
66 DECLARE_WAIT_QUEUE_HEAD(log_wait);
68 int console_printk[4] = {
69 DEFAULT_CONSOLE_LOGLEVEL, /* console_loglevel */
70 DEFAULT_MESSAGE_LOGLEVEL, /* default_message_loglevel */
71 MINIMUM_CONSOLE_LOGLEVEL, /* minimum_console_loglevel */
72 DEFAULT_CONSOLE_LOGLEVEL, /* default_console_loglevel */
76 * Low level drivers may need that to know if they can schedule in
77 * their unblank() callback or not. So let's export it.
80 EXPORT_SYMBOL(oops_in_progress);
83 * console_sem protects the console_drivers list, and also
84 * provides serialisation for access to the entire console
87 static DEFINE_SEMAPHORE(console_sem);
88 struct console *console_drivers;
89 EXPORT_SYMBOL_GPL(console_drivers);
92 * This is used for debugging the mess that is the VT code by
93 * keeping track if we have the console semaphore held. It's
94 * definitely not the perfect debug tool (we don't know if _WE_
95 * hold it are racing, but it helps tracking those weird code
96 * path in the console code where we end up in places I want
97 * locked without the console sempahore held
99 static int console_locked, console_suspended;
102 * If exclusive_console is non-NULL then only this console is to be printed to.
104 static struct console *exclusive_console;
107 * Array of consoles built from command line options (console=)
109 struct console_cmdline
111 char name[8]; /* Name of the driver */
112 int index; /* Minor dev. to use */
113 char *options; /* Options for the driver */
114 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
115 char *brl_options; /* Options for braille driver */
119 #define MAX_CMDLINECONSOLES 8
121 static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
122 static int selected_console = -1;
123 static int preferred_console = -1;
124 int console_set_on_cmdline;
125 EXPORT_SYMBOL(console_set_on_cmdline);
127 /* Flag: console code may call schedule() */
128 static int console_may_schedule;
131 * The printk log buffer consists of a chain of concatenated variable
132 * length records. Every record starts with a record header, containing
133 * the overall length of the record.
135 * The heads to the first and last entry in the buffer, as well as the
136 * sequence numbers of these both entries are maintained when messages
139 * If the heads indicate available messages, the length in the header
140 * tells the start next message. A length == 0 for the next message
141 * indicates a wrap-around to the beginning of the buffer.
143 * Every record carries the monotonic timestamp in microseconds, as well as
144 * the standard userspace syslog level and syslog facility. The usual
145 * kernel messages use LOG_KERN; userspace-injected messages always carry
146 * a matching syslog facility, by default LOG_USER. The origin of every
147 * message can be reliably determined that way.
149 * The human readable log message directly follows the message header. The
150 * length of the message text is stored in the header, the stored message
153 * Optionally, a message can carry a dictionary of properties (key/value pairs),
154 * to provide userspace with a machine-readable message context.
156 * Examples for well-defined, commonly used property names are:
157 * DEVICE=b12:8 device identifier
161 * +sound:card0 subsystem:devname
162 * SUBSYSTEM=pci driver-core subsystem name
164 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
165 * follows directly after a '=' character. Every property is terminated by
166 * a '\0' character. The last property is not terminated.
168 * Example of a message structure:
169 * 0000 ff 8f 00 00 00 00 00 00 monotonic time in nsec
170 * 0008 34 00 record is 52 bytes long
171 * 000a 0b 00 text is 11 bytes long
172 * 000c 1f 00 dictionary is 23 bytes long
173 * 000e 03 00 LOG_KERN (facility) LOG_ERR (level)
174 * 0010 69 74 27 73 20 61 20 6c "it's a l"
176 * 001b 44 45 56 49 43 "DEVIC"
177 * 45 3d 62 38 3a 32 00 44 "E=b8:2\0D"
178 * 52 49 56 45 52 3d 62 75 "RIVER=bu"
180 * 0032 00 00 00 padding to next message header
182 * The 'struct log' buffer header must never be directly exported to
183 * userspace, it is a kernel-private implementation detail that might
184 * need to be changed in the future, when the requirements change.
186 * /dev/kmsg exports the structured data in the following line format:
187 * "level,sequnum,timestamp;<message text>\n"
189 * The optional key/value pairs are attached as continuation lines starting
190 * with a space character and terminated by a newline. All possible
191 * non-prinatable characters are escaped in the "\xff" notation.
193 * Users of the export format should ignore possible additional values
194 * separated by ',', and find the message after the ';' character.
198 LOG_NOCONS = 1, /* already flushed, do not print to console */
199 LOG_NEWLINE = 2, /* text ended with a newline */
200 LOG_PREFIX = 4, /* text started with a prefix */
201 LOG_CONT = 8, /* text is a fragment of a continuation line */
205 u64 ts_nsec; /* timestamp in nanoseconds */
206 u16 len; /* length of entire record */
207 u16 text_len; /* length of text buffer */
208 u16 dict_len; /* length of dictionary buffer */
209 u8 facility; /* syslog facility */
210 u8 flags:5; /* internal record flags */
211 u8 level:3; /* syslog level */
215 * The logbuf_lock protects kmsg buffer, indices, counters. It is also
216 * used in interesting ways to provide interlocking in console_unlock();
218 static DEFINE_RAW_SPINLOCK(logbuf_lock);
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
248 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)
251 #define LOG_ALIGN __alignof__(struct log)
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.
283 return (struct log *)log_buf;
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.
299 msg = (struct log *)log_buf;
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)
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);
319 while (log_first_seq < log_next_seq) {
322 if (log_next_idx > log_first_idx)
323 free = max(log_buf_len - log_next_idx, log_first_idx);
325 free = log_first_idx - log_next_idx;
327 if (free > size + sizeof(struct log))
330 /* drop old messages until we have enough contiuous space */
331 log_first_idx = log_next(log_first_idx);
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));
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;
355 msg->ts_nsec = ts_nsec;
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;
362 log_next_idx += msg->len;
366 /* /dev/kmsg - userspace message inject/listen interface */
367 struct devkmsg_user {
375 static ssize_t devkmsg_writev(struct kiocb *iocb, const struct iovec *iv,
376 unsigned long count, loff_t pos)
380 int level = default_message_loglevel;
381 int facility = 1; /* LOG_USER */
382 size_t len = iov_length(iv, count);
385 if (len > LOG_LINE_MAX)
387 buf = kmalloc(len+1, GFP_KERNEL);
392 for (i = 0; i < count; i++) {
393 if (copy_from_user(line, iv[i].iov_base, iv[i].iov_len)) {
397 line += iv[i].iov_len;
401 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
402 * the decimal value represents 32bit, the lower 3 bit are the log
403 * level, the rest are the log facility.
405 * If no prefix or no userspace facility is specified, we
406 * enforce LOG_USER, to be able to reliably distinguish
407 * kernel-generated messages from userspace-injected ones.
410 if (line[0] == '<') {
413 i = simple_strtoul(line+1, &endp, 10);
414 if (endp && endp[0] == '>') {
425 printk_emit(facility, level, NULL, 0, "%s", line);
431 static ssize_t devkmsg_read(struct file *file, char __user *buf,
432 size_t count, loff_t *ppos)
434 struct devkmsg_user *user = file->private_data;
445 ret = mutex_lock_interruptible(&user->lock);
448 raw_spin_lock_irq(&logbuf_lock);
449 while (user->seq == log_next_seq) {
450 if (file->f_flags & O_NONBLOCK) {
452 raw_spin_unlock_irq(&logbuf_lock);
456 raw_spin_unlock_irq(&logbuf_lock);
457 ret = wait_event_interruptible(log_wait,
458 user->seq != log_next_seq);
461 raw_spin_lock_irq(&logbuf_lock);
464 if (user->seq < log_first_seq) {
465 /* our last seen message is gone, return error and reset */
466 user->idx = log_first_idx;
467 user->seq = log_first_seq;
469 raw_spin_unlock_irq(&logbuf_lock);
473 msg = log_from_idx(user->idx);
474 ts_usec = msg->ts_nsec;
475 do_div(ts_usec, 1000);
478 * If we couldn't merge continuation line fragments during the print,
479 * export the stored flags to allow an optional external merge of the
480 * records. Merging the records isn't always neccessarily correct, like
481 * when we hit a race during printing. In most cases though, it produces
482 * better readable output. 'c' in the record flags mark the first
483 * fragment of a line, '+' the following.
485 if (msg->flags & LOG_CONT && !(user->prev & LOG_CONT))
487 else if ((msg->flags & LOG_CONT) ||
488 ((user->prev & LOG_CONT) && !(msg->flags & LOG_PREFIX)))
491 len = sprintf(user->buf, "%u,%llu,%llu,%c;",
492 (msg->facility << 3) | msg->level,
493 user->seq, ts_usec, cont);
494 user->prev = msg->flags;
496 /* escape non-printable characters */
497 for (i = 0; i < msg->text_len; i++) {
498 unsigned char c = log_text(msg)[i];
500 if (c < ' ' || c >= 127 || c == '\\')
501 len += sprintf(user->buf + len, "\\x%02x", c);
503 user->buf[len++] = c;
505 user->buf[len++] = '\n';
510 for (i = 0; i < msg->dict_len; i++) {
511 unsigned char c = log_dict(msg)[i];
514 user->buf[len++] = ' ';
519 user->buf[len++] = '\n';
524 if (c < ' ' || c >= 127 || c == '\\') {
525 len += sprintf(user->buf + len, "\\x%02x", c);
529 user->buf[len++] = c;
531 user->buf[len++] = '\n';
534 user->idx = log_next(user->idx);
536 raw_spin_unlock_irq(&logbuf_lock);
543 if (copy_to_user(buf, user->buf, len)) {
549 mutex_unlock(&user->lock);
553 static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
555 struct devkmsg_user *user = file->private_data;
563 raw_spin_lock_irq(&logbuf_lock);
566 /* the first record */
567 user->idx = log_first_idx;
568 user->seq = log_first_seq;
572 * The first record after the last SYSLOG_ACTION_CLEAR,
573 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
574 * changes no global state, and does not clear anything.
576 user->idx = clear_idx;
577 user->seq = clear_seq;
580 /* after the last record */
581 user->idx = log_next_idx;
582 user->seq = log_next_seq;
587 raw_spin_unlock_irq(&logbuf_lock);
591 static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
593 struct devkmsg_user *user = file->private_data;
597 return POLLERR|POLLNVAL;
599 poll_wait(file, &log_wait, wait);
601 raw_spin_lock_irq(&logbuf_lock);
602 if (user->seq < log_next_seq) {
603 /* return error when data has vanished underneath us */
604 if (user->seq < log_first_seq)
605 ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
606 ret = POLLIN|POLLRDNORM;
608 raw_spin_unlock_irq(&logbuf_lock);
613 static int devkmsg_open(struct inode *inode, struct file *file)
615 struct devkmsg_user *user;
618 /* write-only does not need any file context */
619 if ((file->f_flags & O_ACCMODE) == O_WRONLY)
622 err = security_syslog(SYSLOG_ACTION_READ_ALL);
626 user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
630 mutex_init(&user->lock);
632 raw_spin_lock_irq(&logbuf_lock);
633 user->idx = log_first_idx;
634 user->seq = log_first_seq;
635 raw_spin_unlock_irq(&logbuf_lock);
637 file->private_data = user;
641 static int devkmsg_release(struct inode *inode, struct file *file)
643 struct devkmsg_user *user = file->private_data;
648 mutex_destroy(&user->lock);
653 const struct file_operations kmsg_fops = {
654 .open = devkmsg_open,
655 .read = devkmsg_read,
656 .aio_write = devkmsg_writev,
657 .llseek = devkmsg_llseek,
658 .poll = devkmsg_poll,
659 .release = devkmsg_release,
664 * This appends the listed symbols to /proc/vmcoreinfo
666 * /proc/vmcoreinfo is used by various utiilties, like crash and makedumpfile to
667 * obtain access to symbols that are otherwise very difficult to locate. These
668 * symbols are specifically used so that utilities can access and extract the
669 * dmesg log from a vmcore file after a crash.
671 void log_buf_kexec_setup(void)
673 VMCOREINFO_SYMBOL(log_buf);
674 VMCOREINFO_SYMBOL(log_buf_len);
675 VMCOREINFO_SYMBOL(log_first_idx);
676 VMCOREINFO_SYMBOL(log_next_idx);
678 * Export struct log size and field offsets. User space tools can
679 * parse it and detect any changes to structure down the line.
681 VMCOREINFO_STRUCT_SIZE(log);
682 VMCOREINFO_OFFSET(log, ts_nsec);
683 VMCOREINFO_OFFSET(log, len);
684 VMCOREINFO_OFFSET(log, text_len);
685 VMCOREINFO_OFFSET(log, dict_len);
689 /* requested log_buf_len from kernel cmdline */
690 static unsigned long __initdata new_log_buf_len;
692 /* save requested log_buf_len since it's too early to process it */
693 static int __init log_buf_len_setup(char *str)
695 unsigned size = memparse(str, &str);
698 size = roundup_pow_of_two(size);
699 if (size > log_buf_len)
700 new_log_buf_len = size;
704 early_param("log_buf_len", log_buf_len_setup);
706 void __init setup_log_buf(int early)
712 if (!new_log_buf_len)
718 mem = memblock_alloc(new_log_buf_len, PAGE_SIZE);
721 new_log_buf = __va(mem);
723 new_log_buf = alloc_bootmem_nopanic(new_log_buf_len);
726 if (unlikely(!new_log_buf)) {
727 pr_err("log_buf_len: %ld bytes not available\n",
732 raw_spin_lock_irqsave(&logbuf_lock, flags);
733 log_buf_len = new_log_buf_len;
734 log_buf = new_log_buf;
736 free = __LOG_BUF_LEN - log_next_idx;
737 memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
738 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
740 pr_info("log_buf_len: %d\n", log_buf_len);
741 pr_info("early log buf free: %d(%d%%)\n",
742 free, (free * 100) / __LOG_BUF_LEN);
745 static bool __read_mostly ignore_loglevel;
747 static int __init ignore_loglevel_setup(char *str)
750 printk(KERN_INFO "debug: ignoring loglevel setting.\n");
755 early_param("ignore_loglevel", ignore_loglevel_setup);
756 module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
757 MODULE_PARM_DESC(ignore_loglevel, "ignore loglevel setting, to"
758 "print all kernel messages to the console.");
760 #ifdef CONFIG_BOOT_PRINTK_DELAY
762 static int boot_delay; /* msecs delay after each printk during bootup */
763 static unsigned long long loops_per_msec; /* based on boot_delay */
765 static int __init boot_delay_setup(char *str)
769 lpj = preset_lpj ? preset_lpj : 1000000; /* some guess */
770 loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
772 get_option(&str, &boot_delay);
773 if (boot_delay > 10 * 1000)
776 pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
777 "HZ: %d, loops_per_msec: %llu\n",
778 boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
781 __setup("boot_delay=", boot_delay_setup);
783 static void boot_delay_msec(int level)
785 unsigned long long k;
786 unsigned long timeout;
788 if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
789 || (level >= console_loglevel && !ignore_loglevel)) {
793 k = (unsigned long long)loops_per_msec * boot_delay;
795 timeout = jiffies + msecs_to_jiffies(boot_delay);
800 * use (volatile) jiffies to prevent
801 * compiler reduction; loop termination via jiffies
802 * is secondary and may or may not happen.
804 if (time_after(jiffies, timeout))
806 touch_nmi_watchdog();
810 static inline void boot_delay_msec(int level)
815 #ifdef CONFIG_SECURITY_DMESG_RESTRICT
816 int dmesg_restrict = 1;
821 static int syslog_action_restricted(int type)
825 /* Unless restricted, we allow "read all" and "get buffer size" for everybody */
826 return type != SYSLOG_ACTION_READ_ALL && type != SYSLOG_ACTION_SIZE_BUFFER;
829 static int check_syslog_permissions(int type, bool from_file)
832 * If this is from /proc/kmsg and we've already opened it, then we've
833 * already done the capabilities checks at open time.
835 if (from_file && type != SYSLOG_ACTION_OPEN)
838 if (syslog_action_restricted(type)) {
839 if (capable(CAP_SYSLOG))
841 /* For historical reasons, accept CAP_SYS_ADMIN too, with a warning */
842 if (capable(CAP_SYS_ADMIN)) {
843 printk_once(KERN_WARNING "%s (%d): "
844 "Attempt to access syslog with CAP_SYS_ADMIN "
845 "but no CAP_SYSLOG (deprecated).\n",
846 current->comm, task_pid_nr(current));
854 #if defined(CONFIG_PRINTK_TIME)
855 static bool printk_time = 1;
857 static bool printk_time;
859 module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
861 static size_t print_time(u64 ts, char *buf)
863 unsigned long rem_nsec;
868 rem_nsec = do_div(ts, 1000000000);
871 return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
873 return sprintf(buf, "[%5lu.%06lu] ",
874 (unsigned long)ts, rem_nsec / 1000);
877 static size_t print_prefix(const struct log *msg, bool syslog, char *buf)
880 unsigned int prefix = (msg->facility << 3) | msg->level;
884 len += sprintf(buf, "<%u>", prefix);
889 else if (prefix > 99)
896 len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
900 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
901 bool syslog, char *buf, size_t size)
903 const char *text = log_text(msg);
904 size_t text_size = msg->text_len;
909 if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
912 if (msg->flags & LOG_CONT) {
913 if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
916 if (!(msg->flags & LOG_NEWLINE))
921 const char *next = memchr(text, '\n', text_size);
925 text_len = next - text;
927 text_size -= next - text;
929 text_len = text_size;
933 if (print_prefix(msg, syslog, NULL) +
934 text_len + 1 >= size - len)
938 len += print_prefix(msg, syslog, buf + len);
939 memcpy(buf + len, text, text_len);
944 /* SYSLOG_ACTION_* buffer size only calculation */
946 len += print_prefix(msg, syslog, NULL);
959 static int syslog_print(char __user *buf, int size)
965 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
973 raw_spin_lock_irq(&logbuf_lock);
974 if (syslog_seq < log_first_seq) {
975 /* messages are gone, move to first one */
976 syslog_seq = log_first_seq;
977 syslog_idx = log_first_idx;
981 if (syslog_seq == log_next_seq) {
982 raw_spin_unlock_irq(&logbuf_lock);
986 skip = syslog_partial;
987 msg = log_from_idx(syslog_idx);
988 n = msg_print_text(msg, syslog_prev, true, text,
989 LOG_LINE_MAX + PREFIX_MAX);
990 if (n - syslog_partial <= size) {
991 /* message fits into buffer, move forward */
992 syslog_idx = log_next(syslog_idx);
994 syslog_prev = msg->flags;
998 /* partial read(), remember position */
1000 syslog_partial += n;
1003 raw_spin_unlock_irq(&logbuf_lock);
1008 if (copy_to_user(buf, text + skip, n)) {
1023 static int syslog_print_all(char __user *buf, int size, bool clear)
1028 text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1032 raw_spin_lock_irq(&logbuf_lock);
1037 enum log_flags prev;
1039 if (clear_seq < log_first_seq) {
1040 /* messages are gone, move to first available one */
1041 clear_seq = log_first_seq;
1042 clear_idx = log_first_idx;
1046 * Find first record that fits, including all following records,
1047 * into the user-provided buffer for this dump.
1052 while (seq < log_next_seq) {
1053 struct log *msg = log_from_idx(idx);
1055 len += msg_print_text(msg, prev, true, NULL, 0);
1057 idx = log_next(idx);
1061 /* move first record forward until length fits into the buffer */
1065 while (len > size && seq < log_next_seq) {
1066 struct log *msg = log_from_idx(idx);
1068 len -= msg_print_text(msg, prev, true, NULL, 0);
1070 idx = log_next(idx);
1074 /* last message fitting into this dump */
1075 next_seq = log_next_seq;
1079 while (len >= 0 && seq < next_seq) {
1080 struct log *msg = log_from_idx(idx);
1083 textlen = msg_print_text(msg, prev, true, text,
1084 LOG_LINE_MAX + PREFIX_MAX);
1089 idx = log_next(idx);
1093 raw_spin_unlock_irq(&logbuf_lock);
1094 if (copy_to_user(buf + len, text, textlen))
1098 raw_spin_lock_irq(&logbuf_lock);
1100 if (seq < log_first_seq) {
1101 /* messages are gone, move to next one */
1102 seq = log_first_seq;
1103 idx = log_first_idx;
1110 clear_seq = log_next_seq;
1111 clear_idx = log_next_idx;
1113 raw_spin_unlock_irq(&logbuf_lock);
1119 int do_syslog(int type, char __user *buf, int len, bool from_file)
1122 static int saved_console_loglevel = -1;
1125 error = check_syslog_permissions(type, from_file);
1129 error = security_syslog(type);
1134 case SYSLOG_ACTION_CLOSE: /* Close log */
1136 case SYSLOG_ACTION_OPEN: /* Open log */
1138 case SYSLOG_ACTION_READ: /* Read from log */
1140 if (!buf || len < 0)
1145 if (!access_ok(VERIFY_WRITE, buf, len)) {
1149 error = wait_event_interruptible(log_wait,
1150 syslog_seq != log_next_seq);
1153 error = syslog_print(buf, len);
1155 /* Read/clear last kernel messages */
1156 case SYSLOG_ACTION_READ_CLEAR:
1159 /* Read last kernel messages */
1160 case SYSLOG_ACTION_READ_ALL:
1162 if (!buf || len < 0)
1167 if (!access_ok(VERIFY_WRITE, buf, len)) {
1171 error = syslog_print_all(buf, len, clear);
1173 /* Clear ring buffer */
1174 case SYSLOG_ACTION_CLEAR:
1175 syslog_print_all(NULL, 0, true);
1177 /* Disable logging to console */
1178 case SYSLOG_ACTION_CONSOLE_OFF:
1179 if (saved_console_loglevel == -1)
1180 saved_console_loglevel = console_loglevel;
1181 console_loglevel = minimum_console_loglevel;
1183 /* Enable logging to console */
1184 case SYSLOG_ACTION_CONSOLE_ON:
1185 if (saved_console_loglevel != -1) {
1186 console_loglevel = saved_console_loglevel;
1187 saved_console_loglevel = -1;
1190 /* Set level of messages printed to console */
1191 case SYSLOG_ACTION_CONSOLE_LEVEL:
1193 if (len < 1 || len > 8)
1195 if (len < minimum_console_loglevel)
1196 len = minimum_console_loglevel;
1197 console_loglevel = len;
1198 /* Implicitly re-enable logging to console */
1199 saved_console_loglevel = -1;
1202 /* Number of chars in the log buffer */
1203 case SYSLOG_ACTION_SIZE_UNREAD:
1204 raw_spin_lock_irq(&logbuf_lock);
1205 if (syslog_seq < log_first_seq) {
1206 /* messages are gone, move to first one */
1207 syslog_seq = log_first_seq;
1208 syslog_idx = log_first_idx;
1214 * Short-cut for poll(/"proc/kmsg") which simply checks
1215 * for pending data, not the size; return the count of
1216 * records, not the length.
1218 error = log_next_idx - syslog_idx;
1220 u64 seq = syslog_seq;
1221 u32 idx = syslog_idx;
1222 enum log_flags prev = syslog_prev;
1225 while (seq < log_next_seq) {
1226 struct log *msg = log_from_idx(idx);
1228 error += msg_print_text(msg, prev, true, NULL, 0);
1229 idx = log_next(idx);
1233 error -= syslog_partial;
1235 raw_spin_unlock_irq(&logbuf_lock);
1237 /* Size of the log buffer */
1238 case SYSLOG_ACTION_SIZE_BUFFER:
1239 error = log_buf_len;
1249 SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1251 return do_syslog(type, buf, len, SYSLOG_FROM_CALL);
1255 * Call the console drivers, asking them to write out
1256 * log_buf[start] to log_buf[end - 1].
1257 * The console_lock must be held.
1259 static void call_console_drivers(int level, const char *text, size_t len)
1261 struct console *con;
1263 trace_console(text, 0, len, len);
1265 if (level >= console_loglevel && !ignore_loglevel)
1267 if (!console_drivers)
1270 for_each_console(con) {
1271 if (exclusive_console && con != exclusive_console)
1273 if (!(con->flags & CON_ENABLED))
1277 if (!cpu_online(smp_processor_id()) &&
1278 !(con->flags & CON_ANYTIME))
1280 con->write(con, text, len);
1285 * Zap console related locks when oopsing. Only zap at most once
1286 * every 10 seconds, to leave time for slow consoles to print a
1289 static void zap_locks(void)
1291 static unsigned long oops_timestamp;
1293 if (time_after_eq(jiffies, oops_timestamp) &&
1294 !time_after(jiffies, oops_timestamp + 30 * HZ))
1297 oops_timestamp = jiffies;
1300 /* If a crash is occurring, make sure we can't deadlock */
1301 raw_spin_lock_init(&logbuf_lock);
1302 /* And make sure that we print immediately */
1303 sema_init(&console_sem, 1);
1306 /* Check if we have any console registered that can be called early in boot. */
1307 static int have_callable_console(void)
1309 struct console *con;
1311 for_each_console(con)
1312 if (con->flags & CON_ANYTIME)
1319 * Can we actually use the console at this time on this cpu?
1321 * Console drivers may assume that per-cpu resources have
1322 * been allocated. So unless they're explicitly marked as
1323 * being able to cope (CON_ANYTIME) don't call them until
1324 * this CPU is officially up.
1326 static inline int can_use_console(unsigned int cpu)
1328 return cpu_online(cpu) || have_callable_console();
1332 * Try to get console ownership to actually show the kernel
1333 * messages from a 'printk'. Return true (and with the
1334 * console_lock held, and 'console_locked' set) if it
1335 * is successful, false otherwise.
1337 * This gets called with the 'logbuf_lock' spinlock held and
1338 * interrupts disabled. It should return with 'lockbuf_lock'
1339 * released but interrupts still disabled.
1341 static int console_trylock_for_printk(unsigned int cpu)
1342 __releases(&logbuf_lock)
1344 int retval = 0, wake = 0;
1346 if (console_trylock()) {
1350 * If we can't use the console, we need to release
1351 * the console semaphore by hand to avoid flushing
1352 * the buffer. We need to hold the console semaphore
1353 * in order to do this test safely.
1355 if (!can_use_console(cpu)) {
1361 logbuf_cpu = UINT_MAX;
1364 raw_spin_unlock(&logbuf_lock);
1368 int printk_delay_msec __read_mostly;
1370 static inline void printk_delay(void)
1372 if (unlikely(printk_delay_msec)) {
1373 int m = printk_delay_msec;
1377 touch_nmi_watchdog();
1383 * Continuation lines are buffered, and not committed to the record buffer
1384 * until the line is complete, or a race forces it. The line fragments
1385 * though, are printed immediately to the consoles to ensure everything has
1386 * reached the console in case of a kernel crash.
1388 static struct cont {
1389 char buf[LOG_LINE_MAX];
1390 size_t len; /* length == 0 means unused buffer */
1391 size_t cons; /* bytes written to console */
1392 struct task_struct *owner; /* task of first print*/
1393 u64 ts_nsec; /* time of first print */
1394 u8 level; /* log level of first message */
1395 u8 facility; /* log level of first message */
1396 enum log_flags flags; /* prefix, newline flags */
1397 bool flushed:1; /* buffer sealed and committed */
1400 static void cont_flush(enum log_flags flags)
1409 * If a fragment of this line was directly flushed to the
1410 * console; wait for the console to pick up the rest of the
1411 * line. LOG_NOCONS suppresses a duplicated output.
1413 log_store(cont.facility, cont.level, flags | LOG_NOCONS,
1414 cont.ts_nsec, NULL, 0, cont.buf, cont.len);
1416 cont.flushed = true;
1419 * If no fragment of this line ever reached the console,
1420 * just submit it to the store and free the buffer.
1422 log_store(cont.facility, cont.level, flags, 0,
1423 NULL, 0, cont.buf, cont.len);
1428 static bool cont_add(int facility, int level, const char *text, size_t len)
1430 if (cont.len && cont.flushed)
1433 if (cont.len + len > sizeof(cont.buf)) {
1434 /* the line gets too long, split it up in separate records */
1435 cont_flush(LOG_CONT);
1440 cont.facility = facility;
1442 cont.owner = current;
1443 cont.ts_nsec = local_clock();
1446 cont.flushed = false;
1449 memcpy(cont.buf + cont.len, text, len);
1452 if (cont.len > (sizeof(cont.buf) * 80) / 100)
1453 cont_flush(LOG_CONT);
1458 static size_t cont_print_text(char *text, size_t size)
1463 if (cont.cons == 0 && (console_prev & LOG_NEWLINE)) {
1464 textlen += print_time(cont.ts_nsec, text);
1468 len = cont.len - cont.cons;
1472 memcpy(text + textlen, cont.buf + cont.cons, len);
1474 cont.cons = cont.len;
1478 if (cont.flags & LOG_NEWLINE)
1479 text[textlen++] = '\n';
1480 /* got everything, release buffer */
1486 asmlinkage int vprintk_emit(int facility, int level,
1487 const char *dict, size_t dictlen,
1488 const char *fmt, va_list args)
1490 static int recursion_bug;
1491 static char textbuf[LOG_LINE_MAX];
1492 char *text = textbuf;
1494 enum log_flags lflags = 0;
1495 unsigned long flags;
1497 int printed_len = 0;
1499 boot_delay_msec(level);
1502 /* This stops the holder of console_sem just where we want him */
1503 local_irq_save(flags);
1504 this_cpu = smp_processor_id();
1507 * Ouch, printk recursed into itself!
1509 if (unlikely(logbuf_cpu == this_cpu)) {
1511 * If a crash is occurring during printk() on this CPU,
1512 * then try to get the crash message out but make sure
1513 * we can't deadlock. Otherwise just return to avoid the
1514 * recursion and return - but flag the recursion so that
1515 * it can be printed at the next appropriate moment:
1517 if (!oops_in_progress && !lockdep_recursing(current)) {
1519 goto out_restore_irqs;
1525 raw_spin_lock(&logbuf_lock);
1526 logbuf_cpu = this_cpu;
1528 if (recursion_bug) {
1529 static const char recursion_msg[] =
1530 "BUG: recent printk recursion!";
1533 printed_len += strlen(recursion_msg);
1534 /* emit KERN_CRIT message */
1535 log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1536 NULL, 0, recursion_msg, printed_len);
1540 * The printf needs to come first; we need the syslog
1541 * prefix which might be passed-in as a parameter.
1543 text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1545 /* mark and strip a trailing newline */
1546 if (text_len && text[text_len-1] == '\n') {
1548 lflags |= LOG_NEWLINE;
1551 /* strip kernel syslog prefix and extract log level or control flags */
1552 if (facility == 0) {
1553 int kern_level = printk_get_level(text);
1556 const char *end_of_header = printk_skip_level(text);
1557 switch (kern_level) {
1560 level = kern_level - '0';
1561 case 'd': /* KERN_DEFAULT */
1562 lflags |= LOG_PREFIX;
1563 case 'c': /* KERN_CONT */
1566 text_len -= end_of_header - text;
1567 text = (char *)end_of_header;
1572 level = default_message_loglevel;
1575 lflags |= LOG_PREFIX|LOG_NEWLINE;
1577 if (!(lflags & LOG_NEWLINE)) {
1579 * Flush the conflicting buffer. An earlier newline was missing,
1580 * or another task also prints continuation lines.
1582 if (cont.len && (lflags & LOG_PREFIX || cont.owner != current))
1583 cont_flush(LOG_NEWLINE);
1585 /* buffer line if possible, otherwise store it right away */
1586 if (!cont_add(facility, level, text, text_len))
1587 log_store(facility, level, lflags | LOG_CONT, 0,
1588 dict, dictlen, text, text_len);
1590 bool stored = false;
1593 * If an earlier newline was missing and it was the same task,
1594 * either merge it with the current buffer and flush, or if
1595 * there was a race with interrupts (prefix == true) then just
1596 * flush it out and store this line separately.
1598 if (cont.len && cont.owner == current) {
1599 if (!(lflags & LOG_PREFIX))
1600 stored = cont_add(facility, level, text, text_len);
1601 cont_flush(LOG_NEWLINE);
1605 log_store(facility, level, lflags, 0,
1606 dict, dictlen, text, text_len);
1608 printed_len += text_len;
1611 * Try to acquire and then immediately release the console semaphore.
1612 * The release will print out buffers and wake up /dev/kmsg and syslog()
1615 * The console_trylock_for_printk() function will release 'logbuf_lock'
1616 * regardless of whether it actually gets the console semaphore or not.
1618 if (console_trylock_for_printk(this_cpu))
1623 local_irq_restore(flags);
1627 EXPORT_SYMBOL(vprintk_emit);
1629 asmlinkage int vprintk(const char *fmt, va_list args)
1631 return vprintk_emit(0, -1, NULL, 0, fmt, args);
1633 EXPORT_SYMBOL(vprintk);
1635 asmlinkage int printk_emit(int facility, int level,
1636 const char *dict, size_t dictlen,
1637 const char *fmt, ...)
1642 va_start(args, fmt);
1643 r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1648 EXPORT_SYMBOL(printk_emit);
1651 * printk - print a kernel message
1652 * @fmt: format string
1654 * This is printk(). It can be called from any context. We want it to work.
1656 * We try to grab the console_lock. If we succeed, it's easy - we log the
1657 * output and call the console drivers. If we fail to get the semaphore, we
1658 * place the output into the log buffer and return. The current holder of
1659 * the console_sem will notice the new output in console_unlock(); and will
1660 * send it to the consoles before releasing the lock.
1662 * One effect of this deferred printing is that code which calls printk() and
1663 * then changes console_loglevel may break. This is because console_loglevel
1664 * is inspected when the actual printing occurs.
1669 * See the vsnprintf() documentation for format string extensions over C99.
1671 asmlinkage int printk(const char *fmt, ...)
1676 #ifdef CONFIG_KGDB_KDB
1677 if (unlikely(kdb_trap_printk)) {
1678 va_start(args, fmt);
1679 r = vkdb_printf(fmt, args);
1684 va_start(args, fmt);
1685 r = vprintk_emit(0, -1, NULL, 0, fmt, args);
1690 EXPORT_SYMBOL(printk);
1692 #else /* CONFIG_PRINTK */
1694 #define LOG_LINE_MAX 0
1695 #define PREFIX_MAX 0
1696 #define LOG_LINE_MAX 0
1697 static u64 syslog_seq;
1698 static u32 syslog_idx;
1699 static u64 console_seq;
1700 static u32 console_idx;
1701 static enum log_flags syslog_prev;
1702 static u64 log_first_seq;
1703 static u32 log_first_idx;
1704 static u64 log_next_seq;
1705 static enum log_flags console_prev;
1706 static struct cont {
1712 static struct log *log_from_idx(u32 idx) { return NULL; }
1713 static u32 log_next(u32 idx) { return 0; }
1714 static void call_console_drivers(int level, const char *text, size_t len) {}
1715 static size_t msg_print_text(const struct log *msg, enum log_flags prev,
1716 bool syslog, char *buf, size_t size) { return 0; }
1717 static size_t cont_print_text(char *text, size_t size) { return 0; }
1719 #endif /* CONFIG_PRINTK */
1721 static int __add_preferred_console(char *name, int idx, char *options,
1724 struct console_cmdline *c;
1728 * See if this tty is not yet registered, and
1729 * if we have a slot free.
1731 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1732 if (strcmp(console_cmdline[i].name, name) == 0 &&
1733 console_cmdline[i].index == idx) {
1735 selected_console = i;
1738 if (i == MAX_CMDLINECONSOLES)
1741 selected_console = i;
1742 c = &console_cmdline[i];
1743 strlcpy(c->name, name, sizeof(c->name));
1744 c->options = options;
1745 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1746 c->brl_options = brl_options;
1752 * Set up a list of consoles. Called from init/main.c
1754 static int __init console_setup(char *str)
1756 char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for index */
1757 char *s, *options, *brl_options = NULL;
1760 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
1761 if (!memcmp(str, "brl,", 4)) {
1764 } else if (!memcmp(str, "brl=", 4)) {
1765 brl_options = str + 4;
1766 str = strchr(brl_options, ',');
1768 printk(KERN_ERR "need port name after brl=\n");
1776 * Decode str into name, index, options.
1778 if (str[0] >= '0' && str[0] <= '9') {
1779 strcpy(buf, "ttyS");
1780 strncpy(buf + 4, str, sizeof(buf) - 5);
1782 strncpy(buf, str, sizeof(buf) - 1);
1784 buf[sizeof(buf) - 1] = 0;
1785 if ((options = strchr(str, ',')) != NULL)
1788 if (!strcmp(str, "ttya"))
1789 strcpy(buf, "ttyS0");
1790 if (!strcmp(str, "ttyb"))
1791 strcpy(buf, "ttyS1");
1793 for (s = buf; *s; s++)
1794 if ((*s >= '0' && *s <= '9') || *s == ',')
1796 idx = simple_strtoul(s, NULL, 10);
1799 __add_preferred_console(buf, idx, options, brl_options);
1800 console_set_on_cmdline = 1;
1803 __setup("console=", console_setup);
1806 * add_preferred_console - add a device to the list of preferred consoles.
1807 * @name: device name
1808 * @idx: device index
1809 * @options: options for this console
1811 * The last preferred console added will be used for kernel messages
1812 * and stdin/out/err for init. Normally this is used by console_setup
1813 * above to handle user-supplied console arguments; however it can also
1814 * be used by arch-specific code either to override the user or more
1815 * commonly to provide a default console (ie from PROM variables) when
1816 * the user has not supplied one.
1818 int add_preferred_console(char *name, int idx, char *options)
1820 return __add_preferred_console(name, idx, options, NULL);
1823 int update_console_cmdline(char *name, int idx, char *name_new, int idx_new, char *options)
1825 struct console_cmdline *c;
1828 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0]; i++)
1829 if (strcmp(console_cmdline[i].name, name) == 0 &&
1830 console_cmdline[i].index == idx) {
1831 c = &console_cmdline[i];
1832 strlcpy(c->name, name_new, sizeof(c->name));
1833 c->name[sizeof(c->name) - 1] = 0;
1834 c->options = options;
1842 bool console_suspend_enabled = 1;
1843 EXPORT_SYMBOL(console_suspend_enabled);
1845 static int __init console_suspend_disable(char *str)
1847 console_suspend_enabled = 0;
1850 __setup("no_console_suspend", console_suspend_disable);
1851 module_param_named(console_suspend, console_suspend_enabled,
1852 bool, S_IRUGO | S_IWUSR);
1853 MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
1854 " and hibernate operations");
1857 * suspend_console - suspend the console subsystem
1859 * This disables printk() while we go into suspend states
1861 void suspend_console(void)
1863 if (!console_suspend_enabled)
1865 printk("Suspending console(s) (use no_console_suspend to debug)\n");
1867 console_suspended = 1;
1871 void resume_console(void)
1873 if (!console_suspend_enabled)
1876 console_suspended = 0;
1881 * console_cpu_notify - print deferred console messages after CPU hotplug
1882 * @self: notifier struct
1883 * @action: CPU hotplug event
1886 * If printk() is called from a CPU that is not online yet, the messages
1887 * will be spooled but will not show up on the console. This function is
1888 * called when a new CPU comes online (or fails to come up), and ensures
1889 * that any such output gets printed.
1891 static int __cpuinit console_cpu_notify(struct notifier_block *self,
1892 unsigned long action, void *hcpu)
1897 case CPU_DOWN_FAILED:
1898 case CPU_UP_CANCELED:
1906 * console_lock - lock the console system for exclusive use.
1908 * Acquires a lock which guarantees that the caller has
1909 * exclusive access to the console system and the console_drivers list.
1911 * Can sleep, returns nothing.
1913 void console_lock(void)
1918 if (console_suspended)
1921 console_may_schedule = 1;
1923 EXPORT_SYMBOL(console_lock);
1926 * console_trylock - try to lock the console system for exclusive use.
1928 * Tried to acquire a lock which guarantees that the caller has
1929 * exclusive access to the console system and the console_drivers list.
1931 * returns 1 on success, and 0 on failure to acquire the lock.
1933 int console_trylock(void)
1935 if (down_trylock(&console_sem))
1937 if (console_suspended) {
1942 console_may_schedule = 0;
1945 EXPORT_SYMBOL(console_trylock);
1947 int is_console_locked(void)
1949 return console_locked;
1953 * Delayed printk version, for scheduler-internal messages:
1955 #define PRINTK_BUF_SIZE 512
1957 #define PRINTK_PENDING_WAKEUP 0x01
1958 #define PRINTK_PENDING_SCHED 0x02
1960 static DEFINE_PER_CPU(int, printk_pending);
1961 static DEFINE_PER_CPU(char [PRINTK_BUF_SIZE], printk_sched_buf);
1963 static void wake_up_klogd_work_func(struct irq_work *irq_work)
1965 int pending = __this_cpu_xchg(printk_pending, 0);
1967 if (pending & PRINTK_PENDING_SCHED) {
1968 char *buf = __get_cpu_var(printk_sched_buf);
1969 printk(KERN_WARNING "[sched_delayed] %s", buf);
1972 if (pending & PRINTK_PENDING_WAKEUP)
1973 wake_up_interruptible(&log_wait);
1976 static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
1977 .func = wake_up_klogd_work_func,
1978 .flags = IRQ_WORK_LAZY,
1981 void wake_up_klogd(void)
1984 if (waitqueue_active(&log_wait)) {
1985 this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
1986 irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
1991 static void console_cont_flush(char *text, size_t size)
1993 unsigned long flags;
1996 raw_spin_lock_irqsave(&logbuf_lock, flags);
2002 * We still queue earlier records, likely because the console was
2003 * busy. The earlier ones need to be printed before this one, we
2004 * did not flush any fragment so far, so just let it queue up.
2006 if (console_seq < log_next_seq && !cont.cons)
2009 len = cont_print_text(text, size);
2010 raw_spin_unlock(&logbuf_lock);
2011 stop_critical_timings();
2012 call_console_drivers(cont.level, text, len);
2013 start_critical_timings();
2014 local_irq_restore(flags);
2017 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2021 * console_unlock - unlock the console system
2023 * Releases the console_lock which the caller holds on the console system
2024 * and the console driver list.
2026 * While the console_lock was held, console output may have been buffered
2027 * by printk(). If this is the case, console_unlock(); emits
2028 * the output prior to releasing the lock.
2030 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2032 * console_unlock(); may be called from any context.
2034 void console_unlock(void)
2036 static char text[LOG_LINE_MAX + PREFIX_MAX];
2037 static u64 seen_seq;
2038 unsigned long flags;
2039 bool wake_klogd = false;
2042 if (console_suspended) {
2047 console_may_schedule = 0;
2049 /* flush buffered message fragment immediately to console */
2050 console_cont_flush(text, sizeof(text));
2057 raw_spin_lock_irqsave(&logbuf_lock, flags);
2058 if (seen_seq != log_next_seq) {
2060 seen_seq = log_next_seq;
2063 if (console_seq < log_first_seq) {
2064 /* messages are gone, move to first one */
2065 console_seq = log_first_seq;
2066 console_idx = log_first_idx;
2070 if (console_seq == log_next_seq)
2073 msg = log_from_idx(console_idx);
2074 if (msg->flags & LOG_NOCONS) {
2076 * Skip record we have buffered and already printed
2077 * directly to the console when we received it.
2079 console_idx = log_next(console_idx);
2082 * We will get here again when we register a new
2083 * CON_PRINTBUFFER console. Clear the flag so we
2084 * will properly dump everything later.
2086 msg->flags &= ~LOG_NOCONS;
2087 console_prev = msg->flags;
2092 len = msg_print_text(msg, console_prev, false,
2093 text, sizeof(text));
2094 console_idx = log_next(console_idx);
2096 console_prev = msg->flags;
2097 raw_spin_unlock(&logbuf_lock);
2099 stop_critical_timings(); /* don't trace print latency */
2100 call_console_drivers(level, text, len);
2101 start_critical_timings();
2102 local_irq_restore(flags);
2106 /* Release the exclusive_console once it is used */
2107 if (unlikely(exclusive_console))
2108 exclusive_console = NULL;
2110 raw_spin_unlock(&logbuf_lock);
2115 * Someone could have filled up the buffer again, so re-check if there's
2116 * something to flush. In case we cannot trylock the console_sem again,
2117 * there's a new owner and the console_unlock() from them will do the
2118 * flush, no worries.
2120 raw_spin_lock(&logbuf_lock);
2121 retry = console_seq != log_next_seq;
2122 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2124 if (retry && console_trylock())
2130 EXPORT_SYMBOL(console_unlock);
2133 * console_conditional_schedule - yield the CPU if required
2135 * If the console code is currently allowed to sleep, and
2136 * if this CPU should yield the CPU to another task, do
2139 * Must be called within console_lock();.
2141 void __sched console_conditional_schedule(void)
2143 if (console_may_schedule)
2146 EXPORT_SYMBOL(console_conditional_schedule);
2148 void console_unblank(void)
2153 * console_unblank can no longer be called in interrupt context unless
2154 * oops_in_progress is set to 1..
2156 if (oops_in_progress) {
2157 if (down_trylock(&console_sem) != 0)
2163 console_may_schedule = 0;
2165 if ((c->flags & CON_ENABLED) && c->unblank)
2171 * Return the console tty driver structure and its associated index
2173 struct tty_driver *console_device(int *index)
2176 struct tty_driver *driver = NULL;
2179 for_each_console(c) {
2182 driver = c->device(c, index);
2191 * Prevent further output on the passed console device so that (for example)
2192 * serial drivers can disable console output before suspending a port, and can
2193 * re-enable output afterwards.
2195 void console_stop(struct console *console)
2198 console->flags &= ~CON_ENABLED;
2201 EXPORT_SYMBOL(console_stop);
2203 void console_start(struct console *console)
2206 console->flags |= CON_ENABLED;
2209 EXPORT_SYMBOL(console_start);
2211 static int __read_mostly keep_bootcon;
2213 static int __init keep_bootcon_setup(char *str)
2216 printk(KERN_INFO "debug: skip boot console de-registration.\n");
2221 early_param("keep_bootcon", keep_bootcon_setup);
2224 * The console driver calls this routine during kernel initialization
2225 * to register the console printing procedure with printk() and to
2226 * print any messages that were printed by the kernel before the
2227 * console driver was initialized.
2229 * This can happen pretty early during the boot process (because of
2230 * early_printk) - sometimes before setup_arch() completes - be careful
2231 * of what kernel features are used - they may not be initialised yet.
2233 * There are two types of consoles - bootconsoles (early_printk) and
2234 * "real" consoles (everything which is not a bootconsole) which are
2235 * handled differently.
2236 * - Any number of bootconsoles can be registered at any time.
2237 * - As soon as a "real" console is registered, all bootconsoles
2238 * will be unregistered automatically.
2239 * - Once a "real" console is registered, any attempt to register a
2240 * bootconsoles will be rejected
2242 void register_console(struct console *newcon)
2245 unsigned long flags;
2246 struct console *bcon = NULL;
2249 * before we register a new CON_BOOT console, make sure we don't
2250 * already have a valid console
2252 if (console_drivers && newcon->flags & CON_BOOT) {
2253 /* find the last or real console */
2254 for_each_console(bcon) {
2255 if (!(bcon->flags & CON_BOOT)) {
2256 printk(KERN_INFO "Too late to register bootconsole %s%d\n",
2257 newcon->name, newcon->index);
2263 if (console_drivers && console_drivers->flags & CON_BOOT)
2264 bcon = console_drivers;
2266 if (preferred_console < 0 || bcon || !console_drivers)
2267 preferred_console = selected_console;
2269 if (newcon->early_setup)
2270 newcon->early_setup();
2273 * See if we want to use this console driver. If we
2274 * didn't select a console we take the first one
2275 * that registers here.
2277 if (preferred_console < 0) {
2278 if (newcon->index < 0)
2280 if (newcon->setup == NULL ||
2281 newcon->setup(newcon, NULL) == 0) {
2282 newcon->flags |= CON_ENABLED;
2283 if (newcon->device) {
2284 newcon->flags |= CON_CONSDEV;
2285 preferred_console = 0;
2291 * See if this console matches one we selected on
2294 for (i = 0; i < MAX_CMDLINECONSOLES && console_cmdline[i].name[0];
2296 if (strcmp(console_cmdline[i].name, newcon->name) != 0)
2298 if (newcon->index >= 0 &&
2299 newcon->index != console_cmdline[i].index)
2301 if (newcon->index < 0)
2302 newcon->index = console_cmdline[i].index;
2303 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2304 if (console_cmdline[i].brl_options) {
2305 newcon->flags |= CON_BRL;
2306 braille_register_console(newcon,
2307 console_cmdline[i].index,
2308 console_cmdline[i].options,
2309 console_cmdline[i].brl_options);
2313 if (newcon->setup &&
2314 newcon->setup(newcon, console_cmdline[i].options) != 0)
2316 newcon->flags |= CON_ENABLED;
2317 newcon->index = console_cmdline[i].index;
2318 if (i == selected_console) {
2319 newcon->flags |= CON_CONSDEV;
2320 preferred_console = selected_console;
2325 if (!(newcon->flags & CON_ENABLED))
2329 * If we have a bootconsole, and are switching to a real console,
2330 * don't print everything out again, since when the boot console, and
2331 * the real console are the same physical device, it's annoying to
2332 * see the beginning boot messages twice
2334 if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2335 newcon->flags &= ~CON_PRINTBUFFER;
2338 * Put this console in the list - keep the
2339 * preferred driver at the head of the list.
2342 if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2343 newcon->next = console_drivers;
2344 console_drivers = newcon;
2346 newcon->next->flags &= ~CON_CONSDEV;
2348 newcon->next = console_drivers->next;
2349 console_drivers->next = newcon;
2351 if (newcon->flags & CON_PRINTBUFFER) {
2353 * console_unlock(); will print out the buffered messages
2356 raw_spin_lock_irqsave(&logbuf_lock, flags);
2357 console_seq = syslog_seq;
2358 console_idx = syslog_idx;
2359 console_prev = syslog_prev;
2360 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2362 * We're about to replay the log buffer. Only do this to the
2363 * just-registered console to avoid excessive message spam to
2364 * the already-registered consoles.
2366 exclusive_console = newcon;
2369 console_sysfs_notify();
2372 * By unregistering the bootconsoles after we enable the real console
2373 * we get the "console xxx enabled" message on all the consoles -
2374 * boot consoles, real consoles, etc - this is to ensure that end
2375 * users know there might be something in the kernel's log buffer that
2376 * went to the bootconsole (that they do not see on the real console)
2379 ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2381 /* we need to iterate through twice, to make sure we print
2382 * everything out, before we unregister the console(s)
2384 printk(KERN_INFO "console [%s%d] enabled, bootconsole disabled\n",
2385 newcon->name, newcon->index);
2386 for_each_console(bcon)
2387 if (bcon->flags & CON_BOOT)
2388 unregister_console(bcon);
2390 printk(KERN_INFO "%sconsole [%s%d] enabled\n",
2391 (newcon->flags & CON_BOOT) ? "boot" : "" ,
2392 newcon->name, newcon->index);
2395 EXPORT_SYMBOL(register_console);
2397 int unregister_console(struct console *console)
2399 struct console *a, *b;
2402 #ifdef CONFIG_A11Y_BRAILLE_CONSOLE
2403 if (console->flags & CON_BRL)
2404 return braille_unregister_console(console);
2408 if (console_drivers == console) {
2409 console_drivers=console->next;
2411 } else if (console_drivers) {
2412 for (a=console_drivers->next, b=console_drivers ;
2413 a; b=a, a=b->next) {
2423 * If this isn't the last console and it has CON_CONSDEV set, we
2424 * need to set it on the next preferred console.
2426 if (console_drivers != NULL && console->flags & CON_CONSDEV)
2427 console_drivers->flags |= CON_CONSDEV;
2430 console_sysfs_notify();
2433 EXPORT_SYMBOL(unregister_console);
2435 static int __init printk_late_init(void)
2437 struct console *con;
2439 for_each_console(con) {
2440 if (!keep_bootcon && con->flags & CON_BOOT) {
2441 printk(KERN_INFO "turn off boot console %s%d\n",
2442 con->name, con->index);
2443 unregister_console(con);
2446 hotcpu_notifier(console_cpu_notify, 0);
2449 late_initcall(printk_late_init);
2451 #if defined CONFIG_PRINTK
2453 int printk_sched(const char *fmt, ...)
2455 unsigned long flags;
2460 local_irq_save(flags);
2461 buf = __get_cpu_var(printk_sched_buf);
2463 va_start(args, fmt);
2464 r = vsnprintf(buf, PRINTK_BUF_SIZE, fmt, args);
2467 __this_cpu_or(printk_pending, PRINTK_PENDING_SCHED);
2468 irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
2469 local_irq_restore(flags);
2475 * printk rate limiting, lifted from the networking subsystem.
2477 * This enforces a rate limit: not more than 10 kernel messages
2478 * every 5s to make a denial-of-service attack impossible.
2480 DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2482 int __printk_ratelimit(const char *func)
2484 return ___ratelimit(&printk_ratelimit_state, func);
2486 EXPORT_SYMBOL(__printk_ratelimit);
2489 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2490 * @caller_jiffies: pointer to caller's state
2491 * @interval_msecs: minimum interval between prints
2493 * printk_timed_ratelimit() returns true if more than @interval_msecs
2494 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2497 bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2498 unsigned int interval_msecs)
2500 if (*caller_jiffies == 0
2501 || !time_in_range(jiffies, *caller_jiffies,
2503 + msecs_to_jiffies(interval_msecs))) {
2504 *caller_jiffies = jiffies;
2509 EXPORT_SYMBOL(printk_timed_ratelimit);
2511 static DEFINE_SPINLOCK(dump_list_lock);
2512 static LIST_HEAD(dump_list);
2515 * kmsg_dump_register - register a kernel log dumper.
2516 * @dumper: pointer to the kmsg_dumper structure
2518 * Adds a kernel log dumper to the system. The dump callback in the
2519 * structure will be called when the kernel oopses or panics and must be
2520 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2522 int kmsg_dump_register(struct kmsg_dumper *dumper)
2524 unsigned long flags;
2527 /* The dump callback needs to be set */
2531 spin_lock_irqsave(&dump_list_lock, flags);
2532 /* Don't allow registering multiple times */
2533 if (!dumper->registered) {
2534 dumper->registered = 1;
2535 list_add_tail_rcu(&dumper->list, &dump_list);
2538 spin_unlock_irqrestore(&dump_list_lock, flags);
2542 EXPORT_SYMBOL_GPL(kmsg_dump_register);
2545 * kmsg_dump_unregister - unregister a kmsg dumper.
2546 * @dumper: pointer to the kmsg_dumper structure
2548 * Removes a dump device from the system. Returns zero on success and
2549 * %-EINVAL otherwise.
2551 int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2553 unsigned long flags;
2556 spin_lock_irqsave(&dump_list_lock, flags);
2557 if (dumper->registered) {
2558 dumper->registered = 0;
2559 list_del_rcu(&dumper->list);
2562 spin_unlock_irqrestore(&dump_list_lock, flags);
2567 EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2569 static bool always_kmsg_dump;
2570 module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2573 * kmsg_dump - dump kernel log to kernel message dumpers.
2574 * @reason: the reason (oops, panic etc) for dumping
2576 * Call each of the registered dumper's dump() callback, which can
2577 * retrieve the kmsg records with kmsg_dump_get_line() or
2578 * kmsg_dump_get_buffer().
2580 void kmsg_dump(enum kmsg_dump_reason reason)
2582 struct kmsg_dumper *dumper;
2583 unsigned long flags;
2585 if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2589 list_for_each_entry_rcu(dumper, &dump_list, list) {
2590 if (dumper->max_reason && reason > dumper->max_reason)
2593 /* initialize iterator with data about the stored records */
2594 dumper->active = true;
2596 raw_spin_lock_irqsave(&logbuf_lock, flags);
2597 dumper->cur_seq = clear_seq;
2598 dumper->cur_idx = clear_idx;
2599 dumper->next_seq = log_next_seq;
2600 dumper->next_idx = log_next_idx;
2601 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2603 /* invoke dumper which will iterate over records */
2604 dumper->dump(dumper, reason);
2606 /* reset iterator */
2607 dumper->active = false;
2613 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2614 * @dumper: registered kmsg dumper
2615 * @syslog: include the "<4>" prefixes
2616 * @line: buffer to copy the line to
2617 * @size: maximum size of the buffer
2618 * @len: length of line placed into buffer
2620 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2621 * record, and copy one record into the provided buffer.
2623 * Consecutive calls will return the next available record moving
2624 * towards the end of the buffer with the youngest messages.
2626 * A return value of FALSE indicates that there are no more records to
2629 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2631 bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2632 char *line, size_t size, size_t *len)
2638 if (!dumper->active)
2641 if (dumper->cur_seq < log_first_seq) {
2642 /* messages are gone, move to first available one */
2643 dumper->cur_seq = log_first_seq;
2644 dumper->cur_idx = log_first_idx;
2648 if (dumper->cur_seq >= log_next_seq)
2651 msg = log_from_idx(dumper->cur_idx);
2652 l = msg_print_text(msg, 0, syslog, line, size);
2654 dumper->cur_idx = log_next(dumper->cur_idx);
2664 * kmsg_dump_get_line - retrieve one kmsg log line
2665 * @dumper: registered kmsg dumper
2666 * @syslog: include the "<4>" prefixes
2667 * @line: buffer to copy the line to
2668 * @size: maximum size of the buffer
2669 * @len: length of line placed into buffer
2671 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2672 * record, and copy one record into the provided buffer.
2674 * Consecutive calls will return the next available record moving
2675 * towards the end of the buffer with the youngest messages.
2677 * A return value of FALSE indicates that there are no more records to
2680 bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2681 char *line, size_t size, size_t *len)
2683 unsigned long flags;
2686 raw_spin_lock_irqsave(&logbuf_lock, flags);
2687 ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2688 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2692 EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2695 * kmsg_dump_get_buffer - copy kmsg log lines
2696 * @dumper: registered kmsg dumper
2697 * @syslog: include the "<4>" prefixes
2698 * @buf: buffer to copy the line to
2699 * @size: maximum size of the buffer
2700 * @len: length of line placed into buffer
2702 * Start at the end of the kmsg buffer and fill the provided buffer
2703 * with as many of the the *youngest* kmsg records that fit into it.
2704 * If the buffer is large enough, all available kmsg records will be
2705 * copied with a single call.
2707 * Consecutive calls will fill the buffer with the next block of
2708 * available older records, not including the earlier retrieved ones.
2710 * A return value of FALSE indicates that there are no more records to
2713 bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2714 char *buf, size_t size, size_t *len)
2716 unsigned long flags;
2721 enum log_flags prev;
2725 if (!dumper->active)
2728 raw_spin_lock_irqsave(&logbuf_lock, flags);
2729 if (dumper->cur_seq < log_first_seq) {
2730 /* messages are gone, move to first available one */
2731 dumper->cur_seq = log_first_seq;
2732 dumper->cur_idx = log_first_idx;
2736 if (dumper->cur_seq >= dumper->next_seq) {
2737 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2741 /* calculate length of entire buffer */
2742 seq = dumper->cur_seq;
2743 idx = dumper->cur_idx;
2745 while (seq < dumper->next_seq) {
2746 struct log *msg = log_from_idx(idx);
2748 l += msg_print_text(msg, prev, true, NULL, 0);
2749 idx = log_next(idx);
2754 /* move first record forward until length fits into the buffer */
2755 seq = dumper->cur_seq;
2756 idx = dumper->cur_idx;
2758 while (l > size && seq < dumper->next_seq) {
2759 struct log *msg = log_from_idx(idx);
2761 l -= msg_print_text(msg, prev, true, NULL, 0);
2762 idx = log_next(idx);
2767 /* last message in next interation */
2773 while (seq < dumper->next_seq) {
2774 struct log *msg = log_from_idx(idx);
2776 l += msg_print_text(msg, prev, syslog, buf + l, size - l);
2777 idx = log_next(idx);
2782 dumper->next_seq = next_seq;
2783 dumper->next_idx = next_idx;
2785 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2791 EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
2794 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
2795 * @dumper: registered kmsg dumper
2797 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2798 * kmsg_dump_get_buffer() can be called again and used multiple
2799 * times within the same dumper.dump() callback.
2801 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
2803 void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
2805 dumper->cur_seq = clear_seq;
2806 dumper->cur_idx = clear_idx;
2807 dumper->next_seq = log_next_seq;
2808 dumper->next_idx = log_next_idx;
2812 * kmsg_dump_rewind - reset the interator
2813 * @dumper: registered kmsg dumper
2815 * Reset the dumper's iterator so that kmsg_dump_get_line() and
2816 * kmsg_dump_get_buffer() can be called again and used multiple
2817 * times within the same dumper.dump() callback.
2819 void kmsg_dump_rewind(struct kmsg_dumper *dumper)
2821 unsigned long flags;
2823 raw_spin_lock_irqsave(&logbuf_lock, flags);
2824 kmsg_dump_rewind_nolock(dumper);
2825 raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2827 EXPORT_SYMBOL_GPL(kmsg_dump_rewind);