NAME
proc – running processes

SYNOPSIS
bind #p /proc
/proc/trace
/proc/
n/args
/proc/
n/ctl
/proc/
n/fd
/proc/
n/fpregs
/proc/
n/kregs
/proc/
n/mem
/proc/
n/note
/proc/
n/noteid
/proc/
n/notepg
/proc/
n/ns
/proc/
n/proc
/proc/
n/profile
/proc/
n/regs
/proc/
n/segment
/proc/
n/status
/proc/
n/text
/proc/
n/wait
...

DESCRIPTION
The proc device serves a two–level directory structure. The first level contains the trace file (see below) and numbered directories corresponding to pids of live processes; each such directory contains a set of files representing the corresponding process.

The mem file contains the current memory image of the process. A read or write at offset o, which must be a valid virtual address, accesses bytes from address o up to the end of the memory segment containing o. Kernel virtual memory, including the kernel stack for the process and saved user registers (whose addresses are machine–dependent), can be accessed through mem. Writes are permitted only while the process is in the Stopped state and only to user addresses or registers.

The read–only proc file contains the kernel per–process structure. Its main use is to recover the kernel stack and program counter for kernel debugging.

The files regs, fpregs, and kregs hold representations of the user–level registers, floating–point registers, and kernel registers in machine–dependent form. The kregs file is read–only.

The read–only fd file lists the open file descriptors of the process. The first line of the file is its current directory; subsequent lines list, one per line, the open files, giving the decimal file descriptor number; whether the file is open for read (r), write, (w), or both (rw); the type, device number, and qid of the file; its I/O unit (the amount of data that may be transferred on the file as a contiguous piece; see iounit(2)), its I/O offset; and its name at the time it was opened.

The read–only ns file contains a textual representation of the process's file name space, in the format of namespace(6) accepted by newns (see auth(2)). The last line of the file identifies the current working directory of the process, in the form of a cd command (see rc(1)). The information in this file is based on the names files had when the name space was assembled, so the names it contains may be inaccessible if the files have been subsequently renamed or rearranged.

The read–only segment file contains a textual display of the memory segments attached to the process. Each line has multiple fields: the type of segment (Stack, Text, Data, Bss, etc.); one–letter flags such as R for read–only, if any; starting virtual address, in hexadecimal; ending virtual address, and reference count.

The read–only status file contains a string with twelve fields, each followed by a space. The fields are:
–     the process name and user name, each 27 characters left justified
–     the process state, 11 characters left justified (see ps(1))
–     the six 11–character numbers also held in the process's #c/cputime file
–     the amount of memory used by the process in units of 1024 bytes
–     the base and current scheduling priority, each 11 character numbers

The args file contains the arguments of the program when it was created by exec(2). Writing to the args file will overwrite its contents. If the program was not created by exec, such as by fork(2), its args file will be empty. The format of the file is a list of quoted strings suitable for tokenize; see getfields(2).

The text file is a pseudonym for the file from which the process was executed; its main use is to recover the symbol table of the process.

The wait file may be read to recover records from the exiting children of the process in the format of await (see wait(2)). If the process has no extant children, living or exited, a read of wait will block. It is an error for a process to attempt to read its own wait file when it has no children. When a process's wait file is being read, the process will draw an error if it attempts an await system call; similarly, if a process is in an await system call, its wait file cannot be read by any process.

The read–only profile file contains the instruction frequency count information used for multiprocess profiling; see tprof in prof(1). The information is gleaned by sampling the program's user–level program counter at interrupt time.

Strings written to the note file will be posted as a note to the process (see notify(2)). The note should be less than ERRLEN–1 characters long; the last character is reserved for a terminating NUL character. A read of at least ERRLEN characters will retrieve the oldest note posted to the process and prevent its delivery to the process. The notepg file is similar, but the note will be delivered to all the processes in the target process's note group (see fork(2)). However, if the process doing the write is in the group, it will not receive the note. The notepg file is write–only.

The textual noteid file may be read to recover an integer identifying the note group of the process (see RFNOTEG in fork(2)). The file may be written to cause the process to change to another note group, provided the group exists and is owned by the same user.

The file /proc/trace can be opened once and read to see trace events from processes that have had the string trace written to their ctl file. Each event produces, in native machine format, the pid, a type, and a time stamp (see /sys/include/trace.h and /sys/src/cmd/trace.c).

The watchpt file contains a list of the watchpoints set for the process. If supported by the hardware, watchpoints can be used to trap accesses to specific addresses. Each line in the file has the form "type address length", where type consists of the characters r (read), w (write), x (execute) or – (padding character). The watchpoint triggers on an access to the length bytes starting at address if the type of the access must match one of the characters in the type field.

Writing to the file either replaces (offset zero) or adds to (offset non–zero) the list of watchpoints. Each line written must be terminated by a newline. If and only if all lines written comply with the (usually rather idiosyncratic) hardware restrictions, the list is updated; otherwise all changes are discarded. Watchpoints can also be cleared by opening the file with OTRUNC (see open(2)).

A triggered watchpoint will deliver a sys: watchpoint note which includes a comma–separated list of the watchpoints that were triggered, where 0 corresponds to the first line in the watchpt file, 1 to the second and so forth.

Control messages
Textual messages written to the ctl file control the execution of the process. Some require that the process is in a particular state and return an error if it is not.
stop    Suspend execution of the process, putting it in the Stopped state.
start   Resume execution of a Stopped process.
waitstop
Do not affect the process directly but, like all other messages ending with stop, block the process writing the ctl file until the target process is in the Stopped state or exits. Also like other stop control messages, if the target process would receive a note while the message is pending, it is instead stopped and the debugging process is resumed.
startstop
Allow a Stopped process to resume, and then do a waitstop action.
hang    Set a bit in the process so that, when it completes an exec(2) system call, it will enter the Stopped state before returning to user mode. This bit is inherited across fork(2) and exec(2).
nohangClear the hang bit.
private
Make it impossible to read the process's user memory. This property is inherited on fork(2), cleared on exec(2), and is not otherwise resettable.
noswapDon't allow this process to be swapped out. This should be used carefully and sparingly or the system could run out of memory. It is meant for processes that can't be swapped, like the local fileserver implementing the swap device and for processes containing sensitive data. This property is inherited on
fork(2), cleared on exec(2), and is not otherwise resettable.
kill    Kill the process the next time it crosses the user/kernel boundary.
close n
Close file descriptor n in the process.
closefiles
Close all open file descriptors in the process.
pri n    Set the base priority for the process to the integer n.
wired n
Wire the process to processor n.
trace   Without an argument, toggle trace event generation for this process into /proc/trace (see below). With a zero argument, tracing for the proc is turned off, with a non–zero numeric argument, it is turned on.
interrupt
Interrupt a blocking system call. If no blocking call was in progress, the interrupt will be pending and the next attempt to block will be interrupted. This is similar to posting a note but, unlike notes, a pending interrupt is not cleared when crossing the user/kernel boundary.
nointerrupt
Clear a pending interrupt.
period nu
Set the real–time scheduling period of the process to nu, where n is an optionally signed number containing an optional decimal point and u is one of s, ms, us, µs, ns, or empty. The time is interpreted, respectively, as seconds, milliseconds, microseconds, microseconds, nanoseconds, or, in the case of an absent units specifier, as nanoseconds. If the time specifier is signed, it is interpreted as an increment or decrement from a previously set value. See also the admit command below.
deadline nu
Set the real–time deadline interval of the process to nu, where n and u are interpreted as for period above.
cost nu
Set the real–time cost (maximum CPU time per period) of the process to nu, where n and u are interpreted as for period above.
sporadic
Use sporadic scheduling for the real–time process. The description of the admit command below contains further details.
yieldonblock
Make the real–time process yield on blocking I/O.
The description of the admit command below contains further details.
admit   Given real–time period, deadline and cost are set (an unset deadline will set deadline to period), perform a schedulability test and start scheduling the process as a real–time process if the test succeeds. If the test fails, the write will fail with error set to the reason for failure.
event   Add a user event to the /proc/trace file.

Real–time scheduling
Real–time processes are periodically released, giving them a higher priority than non–real–time processes until they either give up the processor voluntarily, they exhaust their CPU allocation, or they reach their deadline. The moment of release is dictated by the period and whether the process is sporadic or not. Non– sporadic processes are called periodic and they are released precisely at intervals of their period (but periods can be skipped if the process blocks on I/O). Sporadic processes are released whenever they become runnable (after being blocked by sleep() or I/O), but always at least an interval of period after the previous release.

The deadline of a real–time process specifies that the process must complete within the first deadline seconds of its period. The dealine must be less than or equal to the period. If it is not specified, it is set to the period.

The cost of a real–time process describes the maximum CPU time the process may use per period.

A real–time process can give up the CPU before its deadline is reached or its allocation is exhausted. It does this by calling sleep(0). If yieldonblock is specified, it also does it by executing any blocking system call. Yieldonblock is assumed for sporadic processes.

Of the released processes, the one with the earliest deadline has the highest priority. Care should be taken using spin locks (see lock(2)) because a real–time process spinning on a lock will not give up the processor until its CPU allocation is exhausted; this is unlikely to be the desired behavior.

When a real–time process reaches its deadline or exhausts its CPU allocation, it remains schedulable, but at a very low priority.

The priority is interpreted by Plan 9's multilevel process scheduler. Priorities run from 0 to 19, with higher numbers representing higher priorities. A process has a base priority and a running priority which is less than or equal to the base priority. As a process uses up more of its allocated time, its priority is lowered. Unless explicitly set, user processes have base priority 10, kernel processes 13. Children inherit the parent's base priority.

FILES
/sys/src/9/*/mem.h
/sys/src/9/*/dat.h
/sys/include/trace.h

SEE ALSO
trace(1), debugger(2), mach(2), cons(3)

SOURCE
/sys/src/9/port/devproc.c