1 | =head1 NAME |
---|
2 | |
---|
3 | perlfork - Perl's fork() emulation |
---|
4 | |
---|
5 | =head1 SYNOPSIS |
---|
6 | |
---|
7 | Perl provides a fork() keyword that corresponds to the Unix system call |
---|
8 | of the same name. On most Unix-like platforms where the fork() system |
---|
9 | call is available, Perl's fork() simply calls it. |
---|
10 | |
---|
11 | On some platforms such as Windows where the fork() system call is not |
---|
12 | available, Perl can be built to emulate fork() at the interpreter level. |
---|
13 | While the emulation is designed to be as compatible as possible with the |
---|
14 | real fork() at the the level of the Perl program, there are certain |
---|
15 | important differences that stem from the fact that all the pseudo child |
---|
16 | "processes" created this way live in the same real process as far as the |
---|
17 | operating system is concerned. |
---|
18 | |
---|
19 | This document provides a general overview of the capabilities and |
---|
20 | limitations of the fork() emulation. Note that the issues discussed here |
---|
21 | are not applicable to platforms where a real fork() is available and Perl |
---|
22 | has been configured to use it. |
---|
23 | |
---|
24 | =head1 DESCRIPTION |
---|
25 | |
---|
26 | The fork() emulation is implemented at the level of the Perl interpreter. |
---|
27 | What this means in general is that running fork() will actually clone the |
---|
28 | running interpreter and all its state, and run the cloned interpreter in |
---|
29 | a separate thread, beginning execution in the new thread just after the |
---|
30 | point where the fork() was called in the parent. We will refer to the |
---|
31 | thread that implements this child "process" as the pseudo-process. |
---|
32 | |
---|
33 | To the Perl program that called fork(), all this is designed to be |
---|
34 | transparent. The parent returns from the fork() with a pseudo-process |
---|
35 | ID that can be subsequently used in any process manipulation functions; |
---|
36 | the child returns from the fork() with a value of C<0> to signify that |
---|
37 | it is the child pseudo-process. |
---|
38 | |
---|
39 | =head2 Behavior of other Perl features in forked pseudo-processes |
---|
40 | |
---|
41 | Most Perl features behave in a natural way within pseudo-processes. |
---|
42 | |
---|
43 | =over 8 |
---|
44 | |
---|
45 | =item $$ or $PROCESS_ID |
---|
46 | |
---|
47 | This special variable is correctly set to the pseudo-process ID. |
---|
48 | It can be used to identify pseudo-processes within a particular |
---|
49 | session. Note that this value is subject to recycling if any |
---|
50 | pseudo-processes are launched after others have been wait()-ed on. |
---|
51 | |
---|
52 | =item %ENV |
---|
53 | |
---|
54 | Each pseudo-process maintains its own virtual enviroment. Modifications |
---|
55 | to %ENV affect the virtual environment, and are only visible within that |
---|
56 | pseudo-process, and in any processes (or pseudo-processes) launched from |
---|
57 | it. |
---|
58 | |
---|
59 | =item chdir() and all other builtins that accept filenames |
---|
60 | |
---|
61 | Each pseudo-process maintains its own virtual idea of the current directory. |
---|
62 | Modifications to the current directory using chdir() are only visible within |
---|
63 | that pseudo-process, and in any processes (or pseudo-processes) launched from |
---|
64 | it. All file and directory accesses from the pseudo-process will correctly |
---|
65 | map the virtual working directory to the real working directory appropriately. |
---|
66 | |
---|
67 | =item wait() and waitpid() |
---|
68 | |
---|
69 | wait() and waitpid() can be passed a pseudo-process ID returned by fork(). |
---|
70 | These calls will properly wait for the termination of the pseudo-process |
---|
71 | and return its status. |
---|
72 | |
---|
73 | =item kill() |
---|
74 | |
---|
75 | kill() can be used to terminate a pseudo-process by passing it the ID returned |
---|
76 | by fork(). This should not be used except under dire circumstances, because |
---|
77 | the operating system may not guarantee integrity of the process resources |
---|
78 | when a running thread is terminated. Note that using kill() on a |
---|
79 | pseudo-process() may typically cause memory leaks, because the thread that |
---|
80 | implements the pseudo-process does not get a chance to clean up its resources. |
---|
81 | |
---|
82 | =item exec() |
---|
83 | |
---|
84 | Calling exec() within a pseudo-process actually spawns the requested |
---|
85 | executable in a separate process and waits for it to complete before |
---|
86 | exiting with the same exit status as that process. This means that the |
---|
87 | process ID reported within the running executable will be different from |
---|
88 | what the earlier Perl fork() might have returned. Similarly, any process |
---|
89 | manipulation functions applied to the ID returned by fork() will affect the |
---|
90 | waiting pseudo-process that called exec(), not the real process it is |
---|
91 | waiting for after the exec(). |
---|
92 | |
---|
93 | =item exit() |
---|
94 | |
---|
95 | exit() always exits just the executing pseudo-process, after automatically |
---|
96 | wait()-ing for any outstanding child pseudo-processes. Note that this means |
---|
97 | that the process as a whole will not exit unless all running pseudo-processes |
---|
98 | have exited. |
---|
99 | |
---|
100 | =item Open handles to files, directories and network sockets |
---|
101 | |
---|
102 | All open handles are dup()-ed in pseudo-processes, so that closing |
---|
103 | any handles in one process does not affect the others. See below for |
---|
104 | some limitations. |
---|
105 | |
---|
106 | =back |
---|
107 | |
---|
108 | =head2 Resource limits |
---|
109 | |
---|
110 | In the eyes of the operating system, pseudo-processes created via the fork() |
---|
111 | emulation are simply threads in the same process. This means that any |
---|
112 | process-level limits imposed by the operating system apply to all |
---|
113 | pseudo-processes taken together. This includes any limits imposed by the |
---|
114 | operating system on the number of open file, directory and socket handles, |
---|
115 | limits on disk space usage, limits on memory size, limits on CPU utilization |
---|
116 | etc. |
---|
117 | |
---|
118 | =head2 Killing the parent process |
---|
119 | |
---|
120 | If the parent process is killed (either using Perl's kill() builtin, or |
---|
121 | using some external means) all the pseudo-processes are killed as well, |
---|
122 | and the whole process exits. |
---|
123 | |
---|
124 | =head2 Lifetime of the parent process and pseudo-processes |
---|
125 | |
---|
126 | During the normal course of events, the parent process and every |
---|
127 | pseudo-process started by it will wait for their respective pseudo-children |
---|
128 | to complete before they exit. This means that the parent and every |
---|
129 | pseudo-child created by it that is also a pseudo-parent will only exit |
---|
130 | after their pseudo-children have exited. |
---|
131 | |
---|
132 | A way to mark a pseudo-processes as running detached from their parent (so |
---|
133 | that the parent would not have to wait() for them if it doesn't want to) |
---|
134 | will be provided in future. |
---|
135 | |
---|
136 | =head2 CAVEATS AND LIMITATIONS |
---|
137 | |
---|
138 | =over 8 |
---|
139 | |
---|
140 | =item BEGIN blocks |
---|
141 | |
---|
142 | The fork() emulation will not work entirely correctly when called from |
---|
143 | within a BEGIN block. The forked copy will run the contents of the |
---|
144 | BEGIN block, but will not continue parsing the source stream after the |
---|
145 | BEGIN block. For example, consider the following code: |
---|
146 | |
---|
147 | BEGIN { |
---|
148 | fork and exit; # fork child and exit the parent |
---|
149 | print "inner\n"; |
---|
150 | } |
---|
151 | print "outer\n"; |
---|
152 | |
---|
153 | This will print: |
---|
154 | |
---|
155 | inner |
---|
156 | |
---|
157 | rather than the expected: |
---|
158 | |
---|
159 | inner |
---|
160 | outer |
---|
161 | |
---|
162 | This limitation arises from fundamental technical difficulties in |
---|
163 | cloning and restarting the stacks used by the Perl parser in the |
---|
164 | middle of a parse. |
---|
165 | |
---|
166 | =item Open filehandles |
---|
167 | |
---|
168 | Any filehandles open at the time of the fork() will be dup()-ed. Thus, |
---|
169 | the files can be closed independently in the parent and child, but beware |
---|
170 | that the dup()-ed handles will still share the same seek pointer. Changing |
---|
171 | the seek position in the parent will change it in the child and vice-versa. |
---|
172 | One can avoid this by opening files that need distinct seek pointers |
---|
173 | separately in the child. |
---|
174 | |
---|
175 | =item Forking pipe open() not yet implemented |
---|
176 | |
---|
177 | The C<open(FOO, "|-")> and C<open(BAR, "-|")> constructs are not yet |
---|
178 | implemented. This limitation can be easily worked around in new code |
---|
179 | by creating a pipe explicitly. The following example shows how to |
---|
180 | write to a forked child: |
---|
181 | |
---|
182 | # simulate open(FOO, "|-") |
---|
183 | sub pipe_to_fork ($) { |
---|
184 | my $parent = shift; |
---|
185 | pipe my $child, $parent or die; |
---|
186 | my $pid = fork(); |
---|
187 | die "fork() failed: $!" unless defined $pid; |
---|
188 | if ($pid) { |
---|
189 | close $child; |
---|
190 | } |
---|
191 | else { |
---|
192 | close $parent; |
---|
193 | open(STDIN, "<&=" . fileno($child)) or die; |
---|
194 | } |
---|
195 | $pid; |
---|
196 | } |
---|
197 | |
---|
198 | if (pipe_to_fork('FOO')) { |
---|
199 | # parent |
---|
200 | print FOO "pipe_to_fork\n"; |
---|
201 | close FOO; |
---|
202 | } |
---|
203 | else { |
---|
204 | # child |
---|
205 | while (<STDIN>) { print; } |
---|
206 | close STDIN; |
---|
207 | exit(0); |
---|
208 | } |
---|
209 | |
---|
210 | And this one reads from the child: |
---|
211 | |
---|
212 | # simulate open(FOO, "-|") |
---|
213 | sub pipe_from_fork ($) { |
---|
214 | my $parent = shift; |
---|
215 | pipe $parent, my $child or die; |
---|
216 | my $pid = fork(); |
---|
217 | die "fork() failed: $!" unless defined $pid; |
---|
218 | if ($pid) { |
---|
219 | close $child; |
---|
220 | } |
---|
221 | else { |
---|
222 | close $parent; |
---|
223 | open(STDOUT, ">&=" . fileno($child)) or die; |
---|
224 | } |
---|
225 | $pid; |
---|
226 | } |
---|
227 | |
---|
228 | if (pipe_from_fork('BAR')) { |
---|
229 | # parent |
---|
230 | while (<BAR>) { print; } |
---|
231 | close BAR; |
---|
232 | } |
---|
233 | else { |
---|
234 | # child |
---|
235 | print "pipe_from_fork\n"; |
---|
236 | close STDOUT; |
---|
237 | exit(0); |
---|
238 | } |
---|
239 | |
---|
240 | Forking pipe open() constructs will be supported in future. |
---|
241 | |
---|
242 | =item Global state maintained by XSUBs |
---|
243 | |
---|
244 | External subroutines (XSUBs) that maintain their own global state may |
---|
245 | not work correctly. Such XSUBs will either need to maintain locks to |
---|
246 | protect simultaneous access to global data from different pseudo-processes, |
---|
247 | or maintain all their state on the Perl symbol table, which is copied |
---|
248 | naturally when fork() is called. A callback mechanism that provides |
---|
249 | extensions an opportunity to clone their state will be provided in the |
---|
250 | near future. |
---|
251 | |
---|
252 | =item Interpreter embedded in larger application |
---|
253 | |
---|
254 | The fork() emulation may not behave as expected when it is executed in an |
---|
255 | application which embeds a Perl interpreter and calls Perl APIs that can |
---|
256 | evaluate bits of Perl code. This stems from the fact that the emulation |
---|
257 | only has knowledge about the Perl interpreter's own data structures and |
---|
258 | knows nothing about the containing application's state. For example, any |
---|
259 | state carried on the application's own call stack is out of reach. |
---|
260 | |
---|
261 | =item Thread-safety of extensions |
---|
262 | |
---|
263 | Since the fork() emulation runs code in multiple threads, extensions |
---|
264 | calling into non-thread-safe libraries may not work reliably when |
---|
265 | calling fork(). As Perl's threading support gradually becomes more |
---|
266 | widely adopted even on platforms with a native fork(), such extensions |
---|
267 | are expected to be fixed for thread-safety. |
---|
268 | |
---|
269 | =back |
---|
270 | |
---|
271 | =head1 BUGS |
---|
272 | |
---|
273 | =over 8 |
---|
274 | |
---|
275 | =item * |
---|
276 | |
---|
277 | Having pseudo-process IDs be negative integers breaks down for the integer |
---|
278 | C<-1> because the wait() and waitpid() functions treat this number as |
---|
279 | being special. The tacit assumption in the current implementation is that |
---|
280 | the system never allocates a thread ID of C<1> for user threads. A better |
---|
281 | representation for pseudo-process IDs will be implemented in future. |
---|
282 | |
---|
283 | =item * |
---|
284 | |
---|
285 | This document may be incomplete in some respects. |
---|
286 | |
---|
287 | =back |
---|
288 | |
---|
289 | =head1 AUTHOR |
---|
290 | |
---|
291 | Support for concurrent interpreters and the fork() emulation was implemented |
---|
292 | by ActiveState, with funding from Microsoft Corporation. |
---|
293 | |
---|
294 | This document is authored and maintained by Gurusamy Sarathy |
---|
295 | E<lt>gsar@activestate.comE<gt>. |
---|
296 | |
---|
297 | =head1 SEE ALSO |
---|
298 | |
---|
299 | L<perlfunc/"fork">, L<perlipc> |
---|
300 | |
---|
301 | =cut |
---|