PROC(4) PROC(4)
NAME
/proc - process file system
SYNOPSIS
#include <sys/proc.h>
#include <sys/pioctl.h>
DESCRIPTION
Proc is a file-system mount point that provides access to
the image of each running process in the system. The name
of each entry in the /proc directory is a five-digit decimal
number corresponding to the process id. The owner of each
`file' is determined by the process's user-id; only the
owner is granted access permissions (and only if the pro-
cess' text file is readable). The size of a file is the
total virtual memory size of the process.
The standard system-call interface is used to access proc.
Open(2) and close(2) behave as usual. The object process is
unaffected, except that setuid bits will be ignored if an
opened process exec's. (Setuid bits are also ignored if the
exec'ing process has traced signals, or stops on exec; see
the description of PIOCSMASK and PIOCSEXEC below.) Data may
be transferred from or to any locations in the object's
address space through lseek(2), read(2), and write(2). The
text segment begins at virtual address 0; the data segment
starts above the text. The user area extends downward below
virtual address 0x80000000, and is UPAGES*NBPG bytes long;
the stack segment grows downward below the user area.
Between the end of the data and the beginning of the stack
lies no-man's land. The text, data, and stack sizes may be
determined from the process' proc structure (see below).
There are two differences from reading and writing ordinary
files: (1) no i/o transfer may span a segment boundary; (2)
the user area is writable only in the locations of saved
user registers.
Several process control actions are available through an
ioctl(2) of the form
union { struct proc p; long i; } buffer;
retval = ioctl(fildes, code, &buffer);
The possible codes are as follows:
PIOCGETPR copies the object's proc structure from the ker-
nel process table into buffer.p. Since this
information resides in system space, it is not
accessible via a normal read.
PROC(4) PROC(4)
PIOCSTOP sends the signal SIGSTOP to the object, and waits
for it to enter the stopped state.
PIOCWSTOP simply waits for the object to stop.
PIOCRUN makes the object runnable again after a stop.
PIOCSMASK defines (via the bit mask buffer.i) a set of sig-
nals to be traced; i.e., the arrival of such a
signal will cause the object to stop. (The sig-
nal numbered n is specified by the bit 1<<(n-1).)
A mask of zeroes turns off the trace. The traced
state and mask bits are inherited by the child of
a fork(2). When the object is closed, the mask
bits are lost, but the traced state is retained
for side effects.
PIOCSEXEC causes the object to stop after exec'ing. This
condition is inherited by children and is
retained when the object is closed.
PIOCREXEC reverses the effect of PIOCSEXEC.
PIOCCSIG clears the object's currently pending signal (if
any).
PIOCKILL sends a signal to the process.
PIOCOPENT provides, in retval, a read-only file descriptor
for the object process' text file. This allows a
debugger to find the symbol table without having
to know any path names.
PIOCNICE increments the object's nice(2) priority by the
amount buffer.i. Only the super user may better a
process' priority in this way, but any user may
make the priority worse.
All system calls are interruptible by signals, so that, for
example, an alarm(2) may be set to avoid waiting forever for
a process that may never stop. Any system call is guaran-
teed to be atomic with respect to the object, but, as with
ordinary files, there is nothing to prevent more than one
process from trying to control the same object.
The following header files are useful in analyzing proc
files:
<signal.h> list of signal numbers
<sys/param.h> size parameters (e.g., UPAGES, NBPG)
<sys/types.h> special system types
<sys/user.h> user structure
PROC(4) PROC(4)
<sys/proc.h> proc structure
<sys/reg.h> locations of saved user registers
<sys/pioctl.h> ioctl codes
As with any file system, /proc must be mounted in order to
be used. The mount point should be an empty directory cre-
ated with mode 0555; /etc/procmount should then be run at
boot time. (The file system can be unmounted by
`/etc/procmount -u'.)
FILES
/proc directory (list of active processes)
/proc/nnnnn process image
SEE ALSO
ps(1), hang(1), signal(2), pi(9.1)
DIAGNOSTICS
This is a list of errors which can occur in addition to the
errors normally associated with the file system; see
intro(2):
ENOENT is returned if the object process has exited after
being opened.
EIO is returned if I/O is attempted at an illegal
address in the object.
EBUSY is returned if the object is in the midst of chang-
ing virtual memory attributes, or has pages locked
for physical I/O.
ENOSPC is returned if a write is attempted on a shared text
segment, but there is no room on the swap space to
make a copy.
EPERM is returned if someone other than the super user
attempts to better a process' priority by issuing a
PIOCNICE.
BUGS
A process must be swapped in for reading and writing (but
not ioctl); this consumes minimal system resources, but may
involve a noticeable delay.
The spectrum of states which result in the EBUSY error is
too conservative.
A process loaded from a text file with magic number 0407
does not have a read-only text segment; in this (presumably
rare) case PIOCOPENT does not work, and the process is
accessible even if the text file is read-only.