source: branches/3.2/mindi-busybox/docs/nofork_noexec.txt @ 3186

Last change on this file since 3186 was 2725, checked in by bruno, 9 years ago
  • Update mindi-busybox to 1.18.3 to avoid problems with the tar command which is now failing on recent versions with busybox 1.7.3
  • Property svn:eol-style set to native
File size: 3.4 KB
Line 
1    NOEXEC and NOFORK applets.
2
3Unix shells traditionally execute some commands internally in the attempt
4to dramatically speed up execution. It will be slow as hell if for every
5"echo blah" shell will fork and exec /bin/echo. To this end, shells
6have to _reimplement_ these commands internally.
7
8Busybox is unique in this regard because it already is a collection
9of reimplemented Unix commands, and we can do the same trick
10for speeding up busybox shells, and more. NOEXEC and NOFORK applets
11are exactly those applets which are eligible for these tricks.
12
13Applet will be subject to NOFORK/NOEXEC tricks if it is marked as such
14in applets.h. FEATURE_PREFER_APPLETS is a config option which
15globally enables usage of NOFORK/NOEXEC tricks.
16If it is enabled, FEATURE_SH_STANDALONE can be enabled too,
17and then shells will use NOFORK/NOEXEC tricks for ordinary commands.
18NB: shell builtins use these tricks regardless of FEATURE_SH_STANDALONE
19or FEATURE_PREFER_APPLETS.
20
21In C, if you want to call a program and wait for it, use
22spawn_and_wait(argv), BB_EXECVP(prog,argv) or BB_EXECLP(prog,argv0,...).
23They check whether program name is an applet name and optionally
24do NOFORK/NOEXEC thing depending on configuration.
25
26
27    NOEXEC
28
29NOEXEC applet should work correctly if another applet forks and then
30executes exit(<applet>_main(argc,argv)) in the child. The rules
31roughly are:
32
33* do not expect shared global variables/buffers to be in their
34  "initialized" state. Examples: xfunc_error_retval can be != 1,
35  bb_common_bufsiz1 can be scribbled over, ...
36* do not expect that stdio wasn't used before. Calling set[v]buf()
37  can be disastrous.
38* ...
39
40NOEXEC applets save only one half of fork+exec overhead.
41NOEXEC trick is disabled for NOMMU build.
42
43
44    NOFORK
45
46NOFORK applet should work correctly if another applet simply runs
47<applet>_main(argc,argv) and then continues with its business (xargs,
48find, shells can do it). This poses much more serious limitations
49on what applet can/cannot do:
50
51* all NOEXEC limitations apply.
52* do not ever exit() or exec().
53  - xfuncs are okay. They are using special trick to return
54    to the caller applet instead of dying when they detect "x" condition.
55  - you may "exit" to caller applet by calling xfunc_die(). Return value
56    is taken from xfunc_error_retval.
57  - fflush_stdout_and_exit(n) is ok to use.
58* do not use shared global data, or save/restore shared global data
59  prior to returning. (e.g. bb_common_bufsiz1 is off-limits).
60  - getopt32() is ok to use. You do not need to save/restore option_mask32,
61    it is already done by core code.
62* if you allocate memory, you can use xmalloc() only on the very first
63  allocation. All other allocations should use malloc[_or_warn]().
64  After first allocation, you cannot use any xfuncs.
65  Otherwise, failing xfunc will return to caller applet
66  without freeing malloced data!
67* All allocated data, opened files, signal handlers, termios settings,
68  O_NONBLOCK flags etc should be freed/closed/restored prior to return.
69* ...
70
71NOFORK applets give the most of speed advantage, but are trickiest
72to implement. In order to minimize amount of bugs and maintenance,
73prime candidates for NOFORK-ification are those applets which
74are small and easy to audit, and those which are more likely to be
75frequently executed from shell/find/xargs, particularly in shell
76script loops. Applets which mess with signal handlers, termios etc
77are probably not worth the effort.
78
79Any NOFORK applet is also a NOEXEC applet.
Note: See TracBrowser for help on using the repository browser.