1 NOEXEC and NOFORK applets.
3 Unix shells traditionally execute some commands internally in the attempt
4 to 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
6 have to _reimplement_ these commands internally.
8 Busybox is unique in this regard because it already is a collection
9 of reimplemented Unix commands, and we can do the same trick
10 for speeding up busybox shells, and more. NOEXEC and NOFORK applets
11 are exactly those applets which are eligible for these tricks.
13 Applet will be subject to NOFORK/NOEXEC tricks if it is marked as such
14 in applets.h. FEATURE_PREFER_APPLETS is a config option which
15 globally enables usage of NOFORK/NOEXEC tricks.
16 If it is enabled, FEATURE_SH_STANDALONE can be enabled too,
17 and then shells will use NOFORK/NOEXEC tricks for ordinary commands.
18 NB: shell builtins use these tricks regardless of FEATURE_SH_STANDALONE
19 or FEATURE_PREFER_APPLETS.
21 In C, if you want to call a program and wait for it, use
22 spawn_and_wait(argv), BB_EXECVP(prog,argv) or BB_EXECLP(prog,argv0,...).
23 They check whether program name is an applet name and optionally
24 do NOFORK/NOEXEC thing depending on configuration.
29 NOEXEC applet should work correctly if another applet forks and then
30 executes exit(<applet>_main(argc,argv)) in the child. The rules
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 (although usually xfunc_error_retval's state is not a problem).
37 * do not expect that stdio wasn't used before. Calling set[v]buf()
41 NOEXEC applets save only one half of fork+exec overhead.
42 NOEXEC trick is disabled for NOMMU build.
47 NOFORK applet should work correctly if another applet simply runs
48 <applet>_main(argc,argv) and then continues with its business.
49 xargs, find, shells do it (grep for "spawn_and_wait" and
50 "run_nofork_applet" to find more users).
52 This poses much more serious limitations on what applet can do:
54 * all NOEXEC limitations apply.
55 * do not run for a long time or wait for user input:
56 hush shell only handles signals (like ^C) after you return
58 * do not ever exit() or exec().
59 - xfuncs are okay. They are using special trick to return
60 to the caller applet instead of dying when they detect "x" condition.
61 - you may "exit" to caller applet by calling xfunc_die(). Return value
62 is taken from xfunc_error_retval.
63 - fflush_stdout_and_exit(n) is ok to use.
64 * do not use shared global data, or save/restore shared global data
65 (e.g. bb_common_bufsiz1) prior to returning.
66 - getopt32() is ok to use. You do not need to save/restore option_mask32,
67 xfunc_error_retval, and logmode - it is already done by core code.
68 * if you allocate memory, you can use xmalloc() only on the very first
69 allocation. All other allocations should use malloc[_or_warn]().
70 After first allocation, you cannot use any xfuncs.
71 Otherwise, failing xfunc will return to caller applet
72 without freeing malloced data!
73 * the same applies to other resources, such as open fds: no xfuncs after
75 * All allocated data, opened files, signal handlers, termios settings
76 etc should be freed/closed/restored prior to return.
78 Currently, ash shell signal handling is implemented in a way that signals
79 have non-SA_RESTARTed handlers. This means that system calls can
80 return EINTR. An example of such problem is "yes" applet:
81 it is implemented so that it has a writing loop, this loop is exited on
82 any write error, and in the case of user pressing ^C the error was EINTR.
83 The problem is, the error causes stdout FILE* object to get into error
84 state, needing clearerr() - or else subsequent shell output will also
85 not work. ("yes" has been downgraded to NOEXEC, since hush signal handling
86 does not have this problem - which makes "yes" to not exit on ^C (bug).
87 But stray EINTRs can be seen in any NOFORK under ash, until ash is fixed).
89 NOFORK applets give the most of speed advantage, but are trickiest
90 to implement. In order to minimize amount of bugs and maintenance,
91 prime candidates for NOFORK-ification are those applets which
92 are small and easy to audit, and those which are more likely to be
93 frequently executed from shell/find/xargs, particularly in shell
94 script loops. Applets which mess with signal handlers, termios etc
95 are probably not worth the effort.
97 Applets which must be interruptible by ^C in shells can not be NOFORKs.
99 Any NOFORK applet is also a NOEXEC applet.
102 Calling NOFORK applets
104 API to call NOFORK applets is two functions:
106 run_nofork_applet(appno, argv)
107 spawn_and_wait(argv) // only if FEATURE_PREFER_APPLETS=y
109 First one is directly used by shells if FEATURE_SH_NOFORK=y.
110 Second one is used by many applets, but main users are xargs and find.
111 It itself calls run_nofork_applet(), if argv[0] is a name
114 run_nofork_applet() saves/inits/restores option parsing, xfunc_error_retval,
115 logmode, applet_name. Thus, for example, caller does not need to worry about
116 option_mask32 getting trashed.
119 Calling NOEXEC applets
121 It's the same trusty spawn_and_wait(argv). If FEATURE_PREFER_APPLETS=y,
122 it does NOEXEC trick. It resets xfunc_error_retval = 1 and
123 logmode = LOGMODE_STDIO in the child.
126 Relevant CONFIG options
128 FEATURE_PREFER_APPLETS
129 BB_EXECVP(cmd, argv) will try to exec /proc/self/exe
130 if command's name matches some applet name;
131 spawn_and_wait(argv) will do NOFORK/NOEXEC tricks
133 //TODO: the above two things probably should have separate options?
135 FEATURE_SH_STANDALONE
136 shells will try to exec /proc/self/exe if command's name matches
137 some applet name; shells will do NOEXEC trick on NOEXEC applets
139 //TODO: split (same as for PREFER_APPLETS)
142 shells will do NOFORK trick on NOFORK applets