Section 10.3. Process Information


10.3. Process Information

The kernel makes available quite a bit of information about each process, and some information is passed to new programs when they are loaded. All of this information forms the execution environment for a process.

10.3.1. Program Arguments

There are two types of values passed to new programs when they are run: command-line arguments and environment variables. Various conventions have been set for their usage, but the system itself does not enforce any of these conventions. It is a good idea to stay within the conventions, however, to help your program fit into the Unix world.

Command-line arguments are a set of strings passed to the program. Usually, these are the text typed after the command name in a shell (hence the name), with optional arguments beginning with a dash (-)character.

Environment variables are a set of name/value pairs. Each pair is represented as a single string of the form NAME=VALUE, and the set of these strings makes up the program's environment. For example, the current user's home directory is normally contained in the HOME environment variable, so programs Joe runs often have HOME=/home/joe in their environment.

Both the command-line arguments and environment are made available to a program at startup. The command-line arguments are passed as parameters to the program's main() function, whereas a pointer to the environment is stored in the environ global variable, which is defined in <unistd.h>.[3]

[3] Most systems pass the environment as a parameter to main(), as well, but this method is not standardized by POSIX. The environ variable is the POSIX-approved method.

Here is the complete prototype of main() in the Linux, Unix, and ANSI/ISO C world:

 int main(int argc, char * argv[]); 

You may be surprised to see that main() returns a value (other than void). The value main() returns is given to the process's parent after the process has exited. By convention, 0 indicates that the process completed successfully, and non-0 values indicate failure. Only the lower eight bits of a process's exit code are considered significant. The negative values between -1 and -128 are reserved for processes that are terminated abnormally by another process or by the kernel. An exit code of 0 indicates successful completion, and exit codes between 1 and 127 indicate the program exited because of an error.

The first parameter, argc, contains the number of command-line arguments passed to the program, whereas argv is an array of pointers to the argument strings. The first item in the array is argv[0], which contains the name of the program being invoked (although not necessarily the complete path to the program). The next-to-last item in the array argv[argc - 1] points to the final command-line argument, and argv[argc] contains NULL.

To access the environment directly, use the following global variable:

 extern char * environ[]; 

This provides environ as an array of pointers to each element in the program's environment (remember, each element is a NAME=VALUE pair), and the final item in the array is NULL. This declaration appears in <unistd.h>, so you do not need to declare it yourself. The most common way of checking for elements in the environment is through getenv, which eliminates the need for directly accessing environ.

 const char * getenv(const char * name); 

The sole parameter to getenv() is the name of the environment variable whose value you are interested in. If the variable exists, getenv() returns a pointer to its value. If the variable does not exist in the current environment (the environment pointed to by environ), it returns NULL.

Linux provides two ways of adding strings to the program's environment, setenv() and putenv(). POSIX defines only putenv(), making it the more portable of the two.

 int putenv(const char * string); 

The string passed must be of the form NAME=VALUE.putenv() adds a variable named NAME to the current environment and gives it the value of VALUE. If the environment already contains a variable named NAME, its value is replaced with VALUE.

BSD defines setenv(), which Linux also provides, a more flexible and easier-to-use method for adding items to the environment.

 int setenv(const char * name, const char * value, int overwrite); 

Here, the name and the new value of the environment variable to manipulate are passed separately, which is usually easier for a program to do. If overwrite is 0, the environment is not modified if it already contains a variable called name. Otherwise, the variable's value is modified, as in putenv().

Here is a short example of both of these functions. Both calls accomplish exactly the same thing, changing the current PATH environment variable for the running program.

 putenv("PATH=/bin:/usr/bin"); setenv("PATH", "/bin:/usr/bin", 1); 

10.3.2. Resource Usage

The Linux kernel tracks how many resources each process is using. Although only a few resources are tracked, their measurement can be useful for developers, administrators, and users. Table 10.1 lists the process resources usage currently tracked by the Linux kernel, as of version 2.6.7.

Table 10.1. Process Resources Tracked by Linux






Total time spent executing user mode code. This includes all the time spent running instructions in the application, but not the time the kernel spends fulfilling application requests.



Total time the kernel spent executing requests from the process. This does not include time spent while the process was blocked inside a system call.



The number of minor faults that this process caused. Minor faults are memory accesses that force the processor into kernel mode but do not result in a disk access. These occur when a process tries to write past the end of its stack, forcing the kernel to allocate more stack space before continuing the process, for example.



The number of major faults that this process caused. Major faults are memory accesses that force the kernel to access the disk before the program can continue to run. One common cause of major faults is a process accessing a part of its executable memory that has not yet been loaded into RAM from the disk or has been swapped out at some point.



The number of memory pages that have been paged in from disk due to memory accesses from the process.

A process may examine the resource usage of itself, the accumulated usages of all its children, or the sum of the two. The getrusage() system call returns a struct rusage (which is defined in <sys/resource.h>) that contains the current resources used.

 int getrusage(int who, struct rusage * usage); 

The first parameter, who, tells which of the three available resource counts should be returned. RUSAGE_SELF returns the usage for the current process, RUSAGE_CHILDREN returns the total usage for all of the current process's children, and RUSAGE_BOTH yields the total resources used by this process and all its children. The second parameter to getrusage() is a pointer to a struct rusage, which gets filled in with the appropriate resource utilizations. Although struct rusage contains quite a few members (the list is derived from BSD), most of those members are not yet used by Linux. Here is the complete definition of struct rusage. Table 10.1 describes the members currently used by Linux.

 #include <sys/resource.h> struct rusage {   struct timeval ru_utime;   struct timeval ru_stime;   long int ru_maxrss;   long int ru_ixrss;   long int ru_idrss;   long int ru_isrss;   long int ru_minflt;   long int ru_majflt;   long int ru_nswap;   long int ru_inblock;   long int ru_oublock;   long int ru_msgsnd;   long int ru_msgrcv;   long int ru_nsignals;   long int ru_nvcsw;   long int ru_nivcsw; }; 

10.3.3. Establishing Usage Limits

To help prevent runaway processes from destroying a system's performance, Unix keeps track of many of the resources a process can use and allows the system administrator and the users themselves to place limits on the resources a process may consume.

There are two classes of limits available: hard limits and soft limits. Hard limits may be lowered by any process but may be raised only by the super user. Hard limits are usually set to RLIM_INFINITY on system startup, which means no limit is enforced. The only exception to this is RLIMIT_CORE (the maximum size of a core dump), which Linux initializes to 0 to prevent unexpected core dumps. Many distributions reset this limit on startup, however, as most technical users expect core dumps under some conditions (see page 130 for information on what a core dump is). Soft limits are the limits that the kernel actually enforces. Any process may set the soft limit for a resource to any value less than or equal to that process's hard limit for the resource.

The various limits that may be set are listed in Table 10.2 and are defined in <sys/resource.h>. The geTRlimit() and setrlimit() system calls get and set the limit for a single resource.

Table 10.2. Resource Limits




Maximum amount of memory available to the process. This includes memory used for the stack, global variables, and dynamically allocated memory.


Maximum size of a core file generated by the kernel (if the core file would be too large, none is created).


Total CPU time used (in seconds). For more information on this limit, see the description of SIGXCPU on page 221


Maximum size of data memory (in bytes). This doesn't include memory dynamically allocated by the program.


Maximum size for an open file (checked on writes). For more information on this limit, see the description of SIGXFSZ on page 221.


Maximum amount of memory that may be locked with mlock() (mlock() is decsribed on page 275).


Maximum number of open files.


Maximum number of child processes the process may spawn. This limits only how many children the process may have at one time. It does not limit how many descendants it may have each child may have up to RLIMIT_NPROC children of its own.


Maximum amount of RAM used at any time (any memory usage exceeding this causes paging). This is known as a process's resident set size.


Maximum size of stack memory (in bytes), including all local variables.

 int getrlimit(int resource, struct rlimit *rlim); int setrlimit(int resource, const struct rlimit *rlim); 

Both of these functions use struct rlimit, which is defined as follows:

 #include <sys/resource.h> struct rlimit {   long int rlim_cur;              /* the soft limit */   long int rlim_max;              /* the hard limit */ }; 

The second member, rlim_max, indicates the hard limit for the limit indicated by the resource parameter, and rlim_cur contains the soft limit. These are the same sets of limits manipulated by the ulimit and limit commands, one or the other of which is built into most shells.


    Linux Application Development
    Linux Application Development (paperback) (2nd Edition)
    ISBN: 0321563220
    EAN: 2147483647
    Year: 2003
    Pages: 168

    Similar book on Amazon © 2008-2017.
    If you may any questions please contact us: