Access the Linux kernel using the /proc files...

Access the Linux kernel using the /proc filesystem
This virtual filesystem opens a window of communication between the kernel and user space


Level: Introductory
M. Tim Jones (mtj@mtjones.com), Consultant Engineer, Emulex
14 Mar 2006
The /proc filesystem is a virtual filesystem that permits a novel approach for communication between the Linux® kernel and user space. In the /proc filesystem, virtual files can be read from or written to as a means of communicating with entities in the kernel, but unlike regular files, the content of these virtual files is dynamically created. This article introduces you to the /proc virtual filesystem and demonstrates its use.
The /proc filesystem was originally developed to provide information on the processes in a system. But given the filesystem‘s usefulness, many elements of the kernel use it both to report information and to enable dynamic runtime configuration.
The /proc filesystem contains directories (as a way of organizing information) and virtual files. A virtual file can present information from the kernel to the user and also serve as a means of sending information from the user to the kernel. It‘s not actually required to do both, but this article show you how to configure the filesystem for input and output.
A short article like this can‘t detail all the uses of /proc, but it does demonstrate a couple of uses to give you an idea of how powerful /proc can be. Listing 1 is an interactive tour of some of the /proc elements. It shows the root level of the /proc filesystem. Note the series of numbered files on the left. Each of these is a directory representing a process in the system. Because the first process created in GNU/Linux is the init process, it has a process-id of 1. Next, performing an ls on the directory shows a list of files. Each file provides details on the particular process. For example, to see the command-line entry for init, simply cat the cmdline file.
Some of the other interesting files in /proc are cpuinfo, which identifies the type of processor and its speed; pci, which shows the devices found on the PCI buses; and modules, which identifies the modules that are currently loaded into the kernel.
[root@plato]# ls /proc 1 2040 2347 2874 474 fb mdstat sys 104 2061 2356 2930 9 filesystems meminfo sysrq-trigger 113 2073 2375 2933 acpi fs misc sysvipc 1375 21 2409 2934 buddyinfo ide modules tty 1395 2189 2445 2935 bus interrupts mounts uptime 1706 2201 2514 2938 cmdline iomem mtrr version 179 2211 2515 2947 cpuinfo ioports net vmstat 180 2223 2607 3 crypto irq partitions 181 2278 2608 3004 devices kallsyms pci 182 2291 2609 3008 diskstats kcore self 2 2301 263 3056 dma kmsg slabinfo 2015 2311 2805 394 driver loadavg stat 2019 2337 2821 4 execdomains locks swaps [root@plato 1]# ls /proc/1 auxv cwd exe loginuid mem oom_adj root statm task cmdline environ fd maps mounts oom_score stat status wchan [root@plato]# cat /proc/1/cmdline init [5] [root@plato]#
Listing 2 illustrates reading from and then writing to a virtual file in /proc. This example checks and then enables IP forwarding within the kernel‘s TCP/IP stack.
[root@plato]# cat /proc/sys/net/ipv4/ip_forward 0 [root@plato]# echo "1" > /proc/sys/net/ipv4/ip_forward [root@plato]# cat /proc/sys/net/ipv4/ip_forward 1 [root@plato]#
Alternatively, you could use sysctl to configure these kernel items. See theResources section for more information on that.
By the way, the /proc filesystem isn‘t the only virtual filesystem in GNU/Linux. One such system, sysfs, is similar to /proc but a bit more organized (having learned lessons from /proc). However, /proc is entrenched and therefore, even though sysfs has some advantages over it, /proc is here to stay. There‘s also the debugfs filesystem, but it tends to be (as the name implies) more of a debugging interface. An advantage to debugfs is that it‘s extremely simple to export a single value to user space (in fact, it‘s a single call).
Loadable Kernel Modules (LKM) are an easy way to demonstrate the /proc filesystem, because they‘re a novel way to dynamically add or remove code from the Linux kernel. LKMs are also a popular mechanism for device drivers and filesystems in the Linux kernel.
If you‘ve ever recompiled the Linux kernel, you probably found that in the kernel configuration process, many device drivers and other kernel elements are compiled as modules. If a driver is compiled directly into the kernel, its code and static data occupy space even if they‘re not used. But if the driver is compiled as a module, it requires memory only if memory is needed and subsequently loaded, into the kernel. Interestingly, you won‘t notice a performance hit for LKMs, so they‘re a powerful means of creating a lean kernel that adapts to its environment based upon the available hardware and attached devices.
Here‘s a simple LKM to help you understand how it differs from standard (non-dynamically loadable) code that you‘ll find in the Linux kernel. Listing 3 presents the simplest LKM. (You can download the sample code for this article from theDownloads section, below.)
Listing 3 includes the necessary module header (which defines the module APIs, types, and macros). It then defines the license for the module using MODULE_LICENSE. Here, it specifies GPL to avoid tainting the kernel.
Listing 3 then defines the module init and cleanup functions. The my_module_init function is called when the module is loaded and the function can be used for initialization purposes. The my_module_cleanup function is called when the module is being unloaded and is used to free memory and generally remove traces of the module. Note the use of printk here: this is the kernel printf function. The KERN_INFO symbol is a string that you can use to filter information from entering the kernel ring buffer (much like syslog).
Finally, Listing 3 declares the entry and exit functions using the module_init and module_exit macros. This allows you to name the module init and cleanup functions the way you want but then tell the kernel which functions are the maintenance functions.
#include /* Defines the license for this LKM */ MODULE_LICENSE("GPL"); /* Init function called on module entry */ int my_module_init( void ) { printk(KERN_INFO "my_module_init called. Module is now loaded.\n"); return 0; } /* Cleanup function called on module exit */ void my_module_cleanup( void ) { printk(KERN_INFO "my_module_cleanup called. Module is now unloaded.\n"); return; } /* Declare entry and exit functions */ module_init( my_module_init ); module_exit( my_module_cleanup );
Listing 3 is a real LKM, albeit a simple one. Now, let‘s build it and test it out on a 2.6 kernel. The 2.6 kernel introduces a new method for kernel module building that I find simpler than the older methods. With the file simple-lkm.c, create a makefile whose sole content is:
obj-m += simple-lkm.o
To build the LKM, use the make command as shown in Listing 4.
[root@plato]# make -C /usr/src/linux-`uname -r` SUBDIRS=$PWD modules make: Entering directory `/usr/src/linux-2.6.11‘ CC [M] /root/projects/misc/module2.6/simple/simple-lkm.o Building modules, stage 2. MODPOST CC /root/projects/misc/module2.6/simple/simple-lkm.mod.o LD [M] /root/projects/misc/module2.6/simple/simple-lkm.ko make: Leaving directory `/usr/src/linux-2.6.11‘ [root@plato]#
The result is simple-lkm.ko. The new naming convention helps to distinguish kernel objects (LKMs) from standard objects. You can now load and unload the module and then view its output. To load the module, use the insmod command; conversely, to unload the module, use the rmmod command. lsmod shows the currently loaded LKMs (see Listing 5).
[root@plato]# insmod simple-lkm.ko [root@plato]# lsmod Module Size Used by simple_lkm 1536 0 autofs4 26244 0 video 13956 0 button 5264 0 battery 7684 0 ac 3716 0 yenta_socket 18952 3 rsrc_nonstatic 9472 1 yenta_socket uhci_hcd 32144 0 i2c_piix4 7824 0 dm_mod 56468 3 [root@plato]# rmmod simple-lkm [root@plato]#
Note that kernel output goes to the kernel ring buffer and not to stdout, because stdout is process specific. To inspect messages on the kernel ring buffer, you can use the dmesg utility (or work through /proc itself with the command cat /proc/kmsg). Listing 6 shows the output of the last few messages from dmesg.
[root@plato]# dmesg | tail -5 cs: IO port probe 0xa00-0xaff: clean. eth0: Link is down eth0: Link is up, running at 100Mbit half-duplex my_module_init called. Module is now loaded. my_module_cleanup called. Module is now unloaded. [root@plato]#
You can see the module‘s messages in the kernel output. Now let‘s move beyond this simple example and look at some of the kernel APIs that allow you to develop useful LKMs.


Back to top
The standard APIs that are available to kernel programmers are also available to LKM programmers. It‘s even possible for an LKM to export new variables and functions that the kernel can use. A complete treatment of the APIs is beyond the scope of this article, so I simply present some of the elements that I use later to demonstrate a more useful LKM.
To create a virtual file in the /proc filesystem, use the create_proc_entry function. This function accepts a file name, a set of permissions, and a location in the /proc filesystem in which the file is to reside. The return value of create_proc_entry is a proc_dir_entry pointer (or NULL, indicating an error in create). You can then use the return pointer to configure other aspects of the virtual file, such as the function to call when a read is performed on the file. The prototype for create_proc_entry and a portion of the proc_dir_entry structure are shown in Listing 7.
struct proc_dir_entry *create_proc_entry( const char *name, mode_t mode, struct proc_dir_entry *parent ); struct proc_dir_entry { const char *name; // virtual file name mode_t mode; // mode permissions uid_t uid; // File‘s user id gid_t gid; // File‘s group id struct inode_operations *proc_iops; // Inode operations functions struct file_operations *proc_fops; // File operations functions struct proc_dir_entry *parent; // Parent directory ... read_proc_t *read_proc; // /proc read function write_proc_t *write_proc; // /proc write function void *data; // Pointer to private data atomic_t count; // use count ... }; void remove_proc_entry( const char *name, struct proc_dir_entry *parent );
Later you see how to use the read_proc and write_proc commands to plug in functions for reading and writing the virtual file.
To remove a file from /proc, use the remove_proc_entry function. To use this function, provide the file name string as well as the location of the file in the /proc filesystem (its parent). The function prototype is also shown in Listing 7.
The parent argument can be NULL for the /proc root or a number of other values, depending upon where you want the file to be placed. Table 1 lists some of the other parent proc_dir_entrys that you can use, along with their location in the filesystem.
proc_dir_entry Filesystem location
proc_root_fs /proc
proc_net /proc/net
proc_bus /proc/bus
proc_root_driver /proc/driver
You can write to a /proc entry (from the user to the kernel) by using a write_proc function. This function has this prototype:
int mod_write( struct file *filp, const char __user *buff, unsigned long len, void *data );
The filp argument is essentially an open file structure (we‘ll ignore this). The buff argument is the string data being passed to you. The buffer address is actually a user-space buffer, so you won‘t be able to read it directly. The len argument defines how much data in buff is being written. The data argument is a pointer to the private data (seeListing 7). In the module, I declare a function of this type to deal with the incoming data.
Linux provides a set of APIs to move data between user space and kernel space. For the write_proc case, I use the copy_from_user functions to manipulate the user-space data.
You can read data from a /proc entry (from the kernel to the user) by using the read_proc function. This function has the following prototype:
int mod_read( char *page, char **start, off_t off, int count, int *eof, void *data );
The page argument is the location into which you write the data intended for the user, where count defines the maximum number of characters that can be written. Use the start and off arguments when returning more than a page of data (typically 4KB). When all the data have been written, set the eof (end-of-file) argument. As with write, data represents private data. The page buffer provided here is in kernel space. Therefore, you can write to it without having to invoke copy_to_user.
You can also create directories within the /proc filesystem using proc_mkdir as well as symlinks with proc_symlink. For simple /proc entries that require only a read function, use create_proc_read_entry, which creates the /proc entry and initializes the read_proc function in one call. The prototypes for these functions are shown in Listing 8.
/* Create a directory in the proc filesystem */ struct proc_dir_entry *proc_mkdir( const char *name, struct proc_dir_entry *parent ); /* Create a symlink in the proc filesystem */ struct proc_dir_entry *proc_symlink( const char *name, struct proc_dir_entry *parent, const char *dest ); /* Create a proc_dir_entry with a read_proc_t in one call */ struct proc_dir_entry *create_proc_read_entry( const char *name, mode_t mode, struct proc_dir_entry *base, read_proc_t *read_proc, void *data ); /* Copy buffer to user-space from kernel-space */ unsigned long copy_to_user( void __user *to, const void *from, unsigned long n ); /* Copy buffer to kernel-space from user-space */ unsigned long copy_from_user( void *to, const void __user *from, unsigned long n ); /* Allocate a ‘virtually‘ contiguous block of memory */ void *vmalloc( unsigned long size ); /* Free a vmalloc‘d block of memory */ void vfree( void *addr ); /* Export a symbol to the kernel (make it visible to the kernel) */ EXPORT_SYMBOL( symbol ); /* Export all symbols in a file to the kernel (declare before module.h) */ EXPORT_SYMTAB


Back to top
Here‘s an LKM that supports both reading and writing. This simple application provides a fortune cookie dispenser. After the module is loaded, the user can load text fortunes into it using the echo command and then read them back out individually using the cat command.
Listing 9 presents the basic module functions and variables. The init function (init_fortune_module) allocates space for the cookie pot with vmalloc and then clears it out with memset. With the cookie_pot allocated and empty, I create my proc_dir_entry next in the /proc root called fortune. With proc_entry successfully created, I initialize my local variables and the proc_entry structure. I load my /proc read and write functions (shown in Listings 9 and 10) and identify the owner of the module. The cleanup function simply removes the entry from the /proc filesystem and then frees the memory that cookie_pot occupies.
The cookie_pot is a page in length (4KB) and is managed by two indexes. The first, cookie_index, identifies where the next cookie will be written. The variable next_fortune identifies where the next cookie will be read for output. I simply wrap next_fortune to the beginning when all fortunes have been read.
#include #include #include #include #include #include MODULE_LICENSE("GPL"); MODULE_DESCRIPTION("Fortune Cookie Kernel Module"); MODULE_AUTHOR("M. Tim Jones"); #define MAX_COOKIE_LENGTH PAGE_SIZE static struct proc_dir_entry *proc_entry; static char *cookie_pot; // Space for fortune strings static int cookie_index; // Index to write next fortune static int next_fortune; // Index to read next fortune int init_fortune_module( void ) { int ret = 0; cookie_pot = (char *)vmalloc( MAX_COOKIE_LENGTH ); if (!cookie_pot) { ret = -ENOMEM; } else { memset( cookie_pot, 0, MAX_COOKIE_LENGTH ); proc_entry = create_proc_entry( "fortune", 0644, NULL ); if (proc_entry == NULL) { ret = -ENOMEM; vfree(cookie_pot); printk(KERN_INFO "fortune: Couldn‘t create proc entry\n"); } else { cookie_index = 0; next_fortune = 0; proc_entry->read_proc = fortune_read; proc_entry->write_proc = fortune_write; proc_entry->owner = THIS_MODULE; printk(KERN_INFO "fortune: Module loaded.\n"); } } return ret; } void cleanup_fortune_module( void ) { remove_proc_entry("fortune", &proc_root); vfree(cookie_pot); printk(KERN_INFO "fortune: Module unloaded.\n"); } module_init( init_fortune_module ); module_exit( cleanup_fortune_module );
Writing a new cookie to the pot is a simple process (shown in Listing 10). With the length of the cookie being written, I check to see that space is available for it. If not, I return -ENOSPC, which is communicated to the user process. Otherwise, the space exists, and I use copy_from_user to copy the user buffer directly into the cookie_pot. I then increment the cookie_index (based upon the length of the user buffer) and NULL terminate the string. Finally, I return the number of characters actually written into the cookie_pot that is propagated to the user process.
ssize_t fortune_write( struct file *filp, const char __user *buff, unsigned long len, void *data ) { int space_available = (MAX_COOKIE_LENGTH-cookie_index)+1; if (len > space_available) { printk(KERN_INFO "fortune: cookie pot is full!\n"); return -ENOSPC; } if (copy_from_user( &cookie_pot[cookie_index], buff, len )) { return -EFAULT; } cookie_index += len; cookie_pot[cookie_index-1] = 0; return len; }
Reading a fortune is just as simple, as shown in Listing 11. Because the buffer that I‘ll write to (page) is already in kernel space, I can manipulate it directly and use sprintf to write the next fortune. If the next_fortune index is greater than the cookie_index (next position to write), I wrap next_fortune back to zero, which is the index of the first fortune. After the fortune is written to the user buffer, I increment the next_fortune index by the length of the last fortune written. This places me at the index of the next available fortune. The length of the fortune is returned and propagated to the user.
int fortune_read( char *page, char **start, off_t off, int count, int *eof, void *data ) { int len; if (off > 0) { *eof = 1; return 0; } /* Wrap-around */ if (next_fortune >= cookie_index) next_fortune = 0; len = sprintf(page, "%s\n", &cookie_pot[next_fortune]); next_fortune += len; return len; }
You can see from this simple example that communicating with the kernel through the /proc filesystem is a trivial effort. Now take a look at the fortune module in action (Listing 12).
[root@plato]# insmod fortune.ko [root@plato]# echo "Success is an individual proposition. Thomas Watson" > /proc/fortune [root@plato]# echo "If a man does his best, what else is there? Gen. Patton" > /proc/fortune [root@plato]# echo "Cats: All your base are belong to us. Zero Wing" > /proc/fortune [root@plato]# cat /proc/fortune Success is an individual proposition. Thomas Watson [root@plato]# cat /proc/fortune If a man does his best, what else is there? Gen. Patton [root@plato]#
The /proc virtual filesystem is widely used to report kernel information and also for dynamic configuration. You‘ll find it integral to both driver and module programming. You can learn more about it in theResources below.
Learn
"Administer Linux on the fly" (developerWorks, May 2003) gives you a thorough grounding in /proc, including how you can administer many details of the operating system without ever having to shut down and reboot the machine.
Explore thefiles and subdirectories in the /proc filesystem.
This article ondriver porting to the 2.6 Linux kernel discusses kernel modules in detail.
LinuxHQ is a great site for information on the Linux kernel.
Thedebugfs filesystem is a debugging alternative to /proc.
"Kernel comparison: Improvements in kernel development from 2.4 to 2.6" (developerWorks, February 2004) takes a look behind the scenes at the tools, tests, and techniques that make up kernel 2.6.
"Kernel debugging with Kprobes" (developerWorks, August 2004) shows how in combination with 2.6 kernels, Kprobes provides a lightweight, non-disruptive, and powerful mechanism to insert the printk function dynamically.
The printk function and dmesg methods are common means for kernel debugging. Allessando Rubini‘s book Linux Device Drivers provides anonline chapter about kernel debugging techniques.
Thesysctl command is another option for dynamic kernel configuration.
In thedeveloperWorks Linux zone, find more resources for Linux developers.
Stay current withdeveloperWorks technical events and Webcasts.
Get products and technologies
kernel.org has the latest Linux kernel.
TheGNU make utility documentation is at thegnu.org site.
TheModutils package provides a number of utilities for kernel modules.
Order the SEK for Linux, a two-DVD set containing the latest IBM trial software for Linux from DB2®, Lotus®, Rational®, Tivoli®, and WebSphere®.
WithIBM trial software, available for download directly from developerWorks, build your next development project on Linux.
Discuss
Check outdeveloperWorks blogs and get involved in thedeveloperWorks community.



M. Tim Jones is an embedded software engineer and the author of GNU/Linux Application Programming, AI Application Programming, and BSD Sockets Programming from a Multilanguage Perspective. His engineering background ranges from the development of kernels for geosynchronous spacecraft to embedded systems architecture and networking protocols development. Tim is a senior principal engineer at Emulex Corp.