Kernel logging: APIs and implementation

The use of logs for debugging is as old as computing itself. Logs are useful not only for understanding the internal operation of a system but also the timing and relationships of activities within the system through the time-ordered messages within a time-stamped log.

This article begins its exploration of logging in the kernel by looking at the application programming interfaces (APIs) used to configure and collect the log information (see Figure 1 for a view of the overall framework and components). It then looks at movement of log data from the kernel into user space. Finally, the article explores the target of kernel-based log data: the log management framework in user space with rsyslog.

Figure 1. Kernel logging ecosystem and major components
Kernel logging ecosystem and major components

Kernel API

Logging within the kernel is performed using the printk function, which shares similarity with its user space counterpart, printf (print formatted). The printf command has a long history in languages, most recently in the C language but going much farther back into the 1950s and 1960s in the Fortran (PRINT and FORMAT statements), BCPL (writf function; BCPL was a precursor of C), and ALGOL 68 languages (printf, putf).

Within the kernel, printk (print kernel) is used to write formatted messages into a buffer using a format almost identical to the printf function. You can find the format of printk in ./linux/include/linux/kernel.h (and its implementation in ./linux/kernel/printk.c):

int printk( const char ∗ fmt, ... );

This format indicates that a string is used to define the text and format (just like printf) and is accompanied by a variable set of arguments (identified by the ellipsis [...]).

One of the first differences you’ll see in the use of printk is more about protocol and less about function. This feature uses an obscure aspect of the C language to simplify the specification of message level or priority. The kernel allows each message to be classified with a log level (one of eight that define the severity of the particular message). These levels communicate whether the system has become unusable (an emergency message), a critical condition has occurred (a critical message), or the message is simply informational. The kernel code simply defines the log level as the first argument of the message, as illustrated in the following example for a critical message:

printk( KERN_CRIT "Error code %08x.\n", val );

Note that the first argument is not an argument at all, as no comma (,) separates the level (KERN_CRIT) from the format string. The KERN_CRIT is nothing more than a string itself (in fact, it represents the string "<2>"; see Table 1 for a full list of log levels). As part of the preprocessor, C automatically combines those two strings in a capability called string literal concatenation. The result is a single string that incorporates the log level and user-specified format string as a single string. Note that if a caller does not provide a log level within printk, a default of KERN_WARNING is automatically used (meaning that only log messages of KERN_WARNING and higher priority will be logged).

Table 1. Log levels, symbolics, and uses
Symbolic String Usage
KERN_EMERG (less-than0> Emergency messages (precede a crash)
KERN_ALERT (less-than1> Error requiring immediate attention
KERN_CRIT (less-than2> Critical error (hardware or software)
KERN_ERR (less-than3> Error conditions (common in drivers)
KERN_WARNING (less-than4> Warning conditions (could lead to errors)
KERN_NOTICE (less-than5> Not an error but a significant condition
KERN_INFO (less-than6> Informational message
KERN_DEBUG (less-than7> Used only for debug messages
KERN_DEFAULT (less-thand> Default kernel logging level
KERN_CONT (less-thanc> Continuation of a log line (avoid adding new time stamp)

The printk call can be called from any context in the kernel. The call begins in ./linux/kernel/printk.c in the printk function, which calls vprintk (in the same source file) after resolving the variable-length arguments using va_start.

The vprintk function performs a number of management-level checks (looking for recursion), and then grabs the lock for the log buffer (__log_buf). Next, the incoming string is checked for the log-level string; if found, the log level is set accordingly. Finally, vprintk grabs the current time (using the function cpu_clock) and converts it into a string using sprintf (not the standard library version but an internal kernel version implemented in ./linux/lib/vsprintf.c). The string passed into printk is then copied into the kernel log buffer using a special function that manages the bounds of the ring (emit_log_char). At the end of this function, a gratuitous acquisition and release of the console semaphore is performed that emits the next log message to the console (performed within release_console_sem). The size of the kernel ring buffer was originally 4KB but in recent kernels is sized at 16KB (and up to 1MB, depending on the architecture).

At this point, you’ve explored the API used to insert log messages into the kernel ring buffer. Now, let’s look at the method used to migrate data from the kernel into the host.

Kernel logging and interface

Access to the log buffer is provided at the core through the multi-purpose syslog system call. This single call implements a variety of actions that can all be performed from user space but only one action for non-root users. The prototype for the syslog system call is defined in ./linux/include/linux/syslog.h; its implementation is in ./linux/kernel/printk.c.

The syslog call serves as the input/output (I/O) and control interface to the kernel’s log message ring buffer. From the syslog call, an application can read log messages (partial, in their entirety, or only new messages) as well as control the behavior of the ring buffer (clear contents, set the level of messages to be logged, enable or disable console, and so on). Figure 2 provides a graphical illustration of the logging stack with some of the major components discussed.

Figure 2. Kernel logging stack identifying the major components
Kernel logging stack identifying the major components

The syslog call (called do_syslog within the kernel in ./linux/kernel/printk.c) is a relatively small function that provides the ability to read and control the kernel ring buffer. Note that in glibc 2.0, this function is called klogctl because of overuse of the term syslog, which refers to a variety of calls and applications. The prototype function (in user space) for syslog and klogctl is defined as:

int syslog( int type, char ∗bufp, int len );
int klogctl( int type, char ∗bufp, int len );

The type argument communicates the command to perform and is associated with an optional buffer with its length. Some commands (such as clearing the ring buffer) ignore the bufp and len arguments. Although the first two command types perform no action within the kernel, the rest are used to read log messages or control aspects of logging. Three commands are used to read log messages. The SYSLOG_ACTION_READ command is used to block until log messages are available, and then return them in the provided buffer. This command consumes the messages (older messages will not appear in subsequent calls to this command). The SYSLOG_ACTION_READ_ALL command reads the last n characters from the log (where n is defined as the ‘len’ parameter passed to klogctl). The SYSLOG_ACTION_READ_CLEAR command performs the SYSLOG_ACTION_READ_ALL action followed by a SYSLOG_ACTION_CLEAR command (clear the ring buffer). SYSLOG_ACTION_CONSOLE ON and OFF manipulate the log level to enable or disable log messages to the console, where SYSLOG_CONSOLE_LEVEL allows the caller to define the level of log messages for the console to accept. Finally, SYSLOG_ACTION_SIZE_BUFFER returns the size of the kernel ring buffer, and SYSLOG_ACTION_SIZE_UNREAD returns the number of characters currently available to be read in the kernel ring buffer. The complete list of SYSLOG commands is shown in Table 2.

Table 2. Commands implemented with the syslog/klogctl system call
Command/opcode Purpose
SYSLOG_ACTION_CLOSE (0) Close the log (unimplemented)
SYSLOG_ACTION_OPEN (1) Open the log (unimplemented)
SYSLOG_ACTION_READ (2) Read from the log
SYSLOG_ACTION_READ_ALL (3) Read all messages from the log (non-destructively)
SYSLOG_ACTION_READ_CLEAR (4) Read and clear all messages from the log
SYSLOG_ACTION_CLEAR (5) Clear the ring buffer
SYSLOG_ACTION_CONSOLE_OFF (6) Disable printks to the console
SYSLOG_ACTION_CONSOLE_ON (7) Enable printks to the console
SYSLOG_ACTION_CONSOLE_LEVEL (8) Set level of messages the console accepts
SYSLOG_ACTION_SIZE_UNREAD (9) Return the number of unread characters in the log
SYSLOG_ACTION_SIZE_BUFFER (10) Return the size of the kernel ring buffer

Implemented above the syslog/klogctl layer, the kmsg proc file system is a I/O path (implemented in ./linux/fs/proc/kmsg.c) that provides a binary interface for reading log messages from the kernel buffer. This is commonly read by a daemon (klogd or rsyslogd) that consumes the messages and passes them to rsyslog for routing to the appropriate log file (based on its configuration).

The file /proc/kmsg implements a small number of file operations that equate to internal do_syslog operations. Internally, the open call relates to the SYSLOG_ACTION_OPEN and the release call to SYSLOG_ACTION_CLOSE (each of which is implemented as a No Operation Performed [NOP]). The poll operation allows the wait for activity on the file, and then invokes SYSLOG_ACTION_SIZE_UNREAD to identify the number of characters available to read. Finally, the read operation maps to SYSLOG_ACTION_READ to consume the available log messages. Note that the /proc/kmsg file is not useful to users: It is used by a single daemon to grab log messages and route them to the necessary log file in the /var space.

User space applications

User space provides a number of access points for reading and managing kernel logging. Let’s begin with the lower-level interfaces (such as the /proc file system configuration elements), and then expand to the higher-level applications.

The /proc file system exports more than just a binary interface for accessing log messages (kmsg). It also presents a number of configuration elements both related and independent of those discussed through the syslog/klogctl. Listing 1 shows an exploration of these parameters.

Listing 1. Exploring the printk configuration parameters in /proc
mtj@ubuntu:~$ cat /proc/sys/kernel/printk
4    4    1    7
mtj@ubuntu:~$ cat /proc/sys/kernel/printk_delay
mtj@ubuntu:~$ cat /proc/sys/kernel/printk_ratelimit
mtj@ubuntu:~$ cat /proc/sys/kernel/printk_ratelimit_burst

From Listing 1, the first entry defines the log levels currently used in the printk API. These log levels represent the console log level, default message log level, minimum console log level, and default console log level. The printk_delay value represents the number of milliseconds to delay between printk messages (to add readability in some scenarios). Note here that it’s set to zero, and it cannot be set through /proc. The printk_ratelimit defines the minimum length of time allowed between messages (currently defined as some number of kernel messages every 5 seconds). The number of messages is defined by printk_ratelimit_burst (currently defined as 10). This is particularly useful if you have a chatty kernel but a bandwidth-constrained console device (such as over a serial port). Note that within the kernel, rate limiting is caller controlled and is not implemented within printk. A printk user who desires rate limiting calls the printk_ratelimit function.

The dmesg command can also be used to print and control the kernel ring buffer. This command uses the klogctl system call to read the kernel ring buffer and emit it to standard output (stdout). The command can also be used to clear the kernel ring buffer (using the -c option), set the level for logging to the console (the -n option), and define the size of the buffer used to read the kernel log messages (the -s option). Note that if the buffer size is not specified, dmesg identifies the proper buffer size using the SYSLOG_ACTION_SIZE_BUFFER operation to klogctl.

Finally, the mother of all logging applications is syslog, a standardized logging framework that is implemented in major operating systems (including Linux® and Berkeley Software Distribution [BSD]). syslog has its own protocol used to convey event notification messages over a variety of transport protocols (dividing components into originators, relays, and collectors). In many cases, all three are implemented in a single host. In addition to syslog‘s many interesting features, it specifies how logging information is collected and filtered as well as where to store it. syslog has gone through numerous changes and evolved. You’ve probably heard of syslog, klog, or sysklogd. In more recent distributions of Ubuntu, a new version of syslog called rsyslog is used (based upon the original syslog), which refers to the reliable and extended syslogd.

The rsyslogd daemon, through its configuration file in /etc/rsyslog.conf), understands the /proc file system kmsg interface and uses it to extract kernel logging messages. Note that internally, all log levels are written through /proc/kmsg so that instead of the kernel defining which log levels to transport, the task is left to rsyslog itself. The kernel log messages are then stored in /var/log/kern.log (among other configured files). In /var/log, you’ll find a plethora of log files that include general message and system-related calls (/var/log/messages), system boot log (/var/log/boot.log), authentication logs (/var/log/auth.log), and others.

Although the logs are available for your review, you can also use them for automated audits and forensics. A variety of log file analyzers exists for troubleshooting or compliance with security regulations and automatically look for problems using techniques such as pattern recognition or correlation analysis (even across systems).

Going farther

This article gave a glimpse into kernel logging and applications—from kernel log message creation in the kernel to its storage within the kernel’s ring buffer to its transport into user space through syslog/klogctl or /proc/kmsg to its routing through the rsyslog logging framework to its final resting place in the /var/log subtree. Linux provides a rich and flexible framework for logging (both in the kernel and external).