README_distro_proposal.txt: writeup about runit adoption

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

top: fix help text: with !TERMIOS, no keys are affected top output

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

install: fix musl problem with dirname

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Do not print useless empty line after list of applets

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Code style fixes, no code changes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

lineedit: simplify code a bit

function                                             old     new   delta
lineedit_read_key                                    155     162      +7
put_prompt                                            51      46      -5
read_line_input                                     3722    3715      -7
cmdedit_setwidth                                      81      50     -31
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 1/3 up/down: 7/-43)             Total: -36 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

makedevs: make special node creation idempotent

When makedevs is called for a second time with the same device file,
it fails because the files already exist and mknod() gives -EEXIST.

Ignore EEXIST errors.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

lineedit: fix two bugs in SIGWINCH signal handling

(1) restore entire sigaction, not only signal handler function
(2) do not use stdio when not sure WINCH did not interrupt a printf() or such.

function                                             old     new   delta
cmdedit_setwidth                                       -      81     +81
read_line_input                                     3682    3722     +40
lineedit_read_key                                    138     155     +17
put_prompt                                            55      51      -4
win_changed                                           93      47     -46
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 2/2 up/down: 138/-50)            Total: 88 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

fdisk_gpt: simplify GPT partition name printing

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

fdisk: sanitize partition name printing; drop "Code" column; get rid of one static var

function                                             old     new   delta
list_table                                          2335    2373     +38
fill_bounds                                          131     128      -3
part_array_len                                         4       -      -4
get_boot                                            1584    1574     -10
------------------------------------------------------------------------------
(add/remove: 0/1 grow/shrink: 1/2 up/down: 38/-17)             Total: 21 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

dpkg-deb: shorten code, improve help text

function                                             old     new   delta
packed_usage                                       30261   30236     -25
dpkg_deb_main                                        437     401     -36
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-61)             Total: -61 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

dpkg-deb: remove unused FEATURE_DPKG_DEB_EXTRACT_ONLY config option

Its usage in C code was removed in 2004.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

find: fix handling of trailing slashes in -name PATTERN comparisons

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash,hush: make ^C in interactive mode visually much closer to bash behavior

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Move FEATURE_BUFFERS_USE_foo options to library tuning

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Move locale, unicode, and "use sendfile?" options to library tuning

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Collapse three levers of menuconfig to two levels.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash,hush: ^C from command line should set $? to 128+SIGINT

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

lineedit: fix handling of repeating Alt-b, Alt-f, Alt-d, Alt-Backspace

These key combinations should repeat correctly when the keys are
pressed and held.

Before this change, they do this erratically - many repeats are "eaten"
because they are treated as unrecognized ESC seqs:
ESC 0x7f is treated by Alt+baskspace, but ESC 0x7f ESC 0x7f ESC 0x7f
is unrecognized.

Escape sequences corresponding to these key combinations are moved from
read_line_input to lineedit_read_key.

Also, these key sequences are now enabled regardless of whether
FEATURE_EDITING_VI is set, since Vim does not actually support these key
combinations, but they are present in readline library.

function                                             old     new   delta
static.esccmds                                        93     103     +10
read_line_input                                     3737    3687     -50

Signed-off-by: Rostislav Skudnov <rostislav@tuxera.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

patch: fix matching failure

Fix matching failure when plist is advanced while checking for buffered
lines - the lines in the hunk that are about to be added should be
skipped when checking for matching context.

Also add a valid test case that fails with current busybox and is fixed
by the change.

Signed-off-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

patch: fix debug log failure

If we reach the end of plist it means the input has still data while we
are expecting EOF. Fix the log to avoid a crash.

Signed-off-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Remove outdated comments

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all udhcp applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all selinux/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

test: make [ and [[ forms individually selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Remove remnants of disabled "length" applet

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make dos2unix/unix2dos independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all coreutils/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make "hd" independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all util-linux/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Remove trailing empty lines

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all console-tools/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all mailutils/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

tc: consolidate its disabled bits

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make ping6 and traceroute6 selectable independently from ping/traceroute

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ipneigh: add this shortcur similar to other "ip <CMD>" shortcuts

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Rename FEATURE_FTP_WRITE and FEATURE_FTP_AUTHENTICATION to *_FTPD_*

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all networking/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all modutils/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make killall and killall5 selecatable independent from kill

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all procps/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Convert all miscutils/* applets to "new style" applet definitions

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Split miscutils/Config.src items into miscutils/*.c files

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

httpd: explain why we use sprintf and why it should be fine

While at it, fix a pathological case where it is not fine:
-r REALM with some 8-kbyte long REALM would overflow the buffer.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Fix allnoconfig warnings

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

applets.h.sh: a script to check applet names against config options

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make mkfs.vfat and mkdosfs individually selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make mke2fs and mkfs.ext2 individually selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make bzcat, lzcat, xzcat, zcat, lzopcat, unlzop individually selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make ifup and ifdown individually selectable.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make swapon and swapoff individually selectable.

For example, without swapoff, code shrinks by 277 bytes.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make grep/egrep/fgrep independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make halt/poweroff/reboot independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Make setarch/linux32/linux64 independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

dnsdomainname: split it from "hostname", make it independently selectable

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

tar: handle pax-encoded utf8 filenames and link names. Closes 9406

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: make getch/peek functions directly called

Indirect calls are more difficult to predict.
Unfortunately, on x64 direct call is 5 bytes while indirect "call (reg+ofs)"
is 3 bytes:

function                                             old     new   delta
i_getch                                                -      82     +82
i_peek                                                 -      63     +63
parse_stream                                        2531    2579     +48
parse_dollar                                         771     797     +26
parse_redirect                                       296     321     +25
add_till_closing_bracket                             408     420     +12
encode_string                                        256     265      +9
i_peek_and_eat_bkslash_nl                             93      99      +6
add_till_backquote                                   110     114      +4
parse_and_run_stream                                 139     141      +2
expand_vars_to_list                                 1143    1144      +1
static_peek                                            6       -      -6
setup_string_in_str                                   39      18     -21
setup_file_in_str                                     40      19     -21
static_get                                            27       -     -27
file_peek                                             52       -     -52
file_get                                              65       -     -65
------------------------------------------------------------------------------
(add/remove: 2/4 grow/shrink: 9/2 up/down: 278/-192)           Total: 86 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: renumber PIPE_foo, make PIPE_SEQ = 0

PIPE_SEQ is used most often, having it zero makes code smaller:

function                                             old     new   delta
done_word                                            719     707     -12
parse_stream                                        2546    2531     -15

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: small optimization in run_list

I thought gcc can detect this itself. It doesn't.

function                                             old     new   delta
run_list                                            1030    1021      -9

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: deindent large block of code, no code changes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: simplify insert_bg_jobs

function                                             old     new   delta
insert_bg_job                                        366     281     -85

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: make "wait %1" less likely to play with signal mask

Was playing with "sleep 3 | exit 3 & wait %1" and noticed that often
SIGCHLD arrives even before I get to signal masking. Can avoid it in this
case.

function                                             old     new   delta
wait_for_child_or_signal                             228     265     +37

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: rework "wait %jobspec" to work in non-interactive shells too

Also add tests. wait5.tests so far fails (but works for ash and dash).

function                                             old     new   delta
builtin_wait                                         305     283     -22

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: "wait $!; echo $?" should return 127 if $! already exited

It would be nice to provide bash-like "remember las exitcode"
thingy, but it's a bit complex. For now, match ash and dash.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: implement "wait %jobspec"

function                                             old     new   delta
parse_jobspec                                          -      83     +83
job_exited_or_stopped                                  -      79     +79
builtin_wait                                         236     302     +66
wait_for_child_or_signal                             199     228     +29
checkjobs                                            142     158     +16
builtin_jobs                                          59      68      +9
process_wait_result                                  453     408     -45
builtin_fg_bg                                        272     203     -69
------------------------------------------------------------------------------
(add/remove: 2/0 grow/shrink: 4/2 up/down: 282/-114)          Total: 168 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: factor out %jobspec parsing

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: comment fix

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: do not allow sh -c '{ echo boo }'

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

cpio: add ownership (-R) test cases

Add ownership (-R) test cases.

Signed-off-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: case logic for setting $? was still wrong

Resetting to 0 should happen in "esac". Matched branch must
still see previous $?.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: non-matching "case" statement sets $? to 0

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: allow { cmd } to not be terminated by semicolon in some cases

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: fix "duplicate local" code (forgot to re-enable interrupts)

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

comment and readme updates

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

su: expand help; simplify passing of -c CMD to run_shell()

Also, added a comment about bug 9401 (TIOCSTI input injection).

function                                             old     new   delta
packed_usage                                       30909   30932     +23
su_main                                              470     487     +17
sulogin_main                                         260     258      -2
run_applet_and_exit                                  681     678      -3
run_shell                                            166     126     -40

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

loadfont: 32k size limit is not enough

Since our "read to malloced buf" routines only gradually grow
allocations, let's be generous here and allow 128k.

Reported by Alex Henrie <alexhenrie24@gmail.com>

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: while (!got_sig) pause() is not reliable, use sigsuspend()

dash was doing it for a reason. Unfortunately, it had no comment why...
now I know.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

man: remove -Tascii from nroff invocation

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

man: make width selection more thorough; explain how to override it

Fedora's "man CMD >file" still uses terminal width, not 80 (but disables formatting),
this change mimics that.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

man: allow nroff and tbl commands be overridden; unmangle writing to files

Parse this in config files:
 DEFINE col ...
 DEFINE tbl ...
 DEFINE nroff ...
Add width options to nroff command line.
Use "tbl", not "gtbl", as default tbl command.
Export GROFF_NO_SGR=1 and use "col -b -p -x" instead of pager when writing to file.

function                                             old     new   delta
man_main                                             735     863    +128
if_redefined                                           -      64     +64
show_manpage                                         199     169     -30
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 1/1 up/down: 192/-30)           Total: 162 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: if using libc glob(), skip it if no metachars are in word

This saves making tons of pointless stat() calls

function                                             old     new   delta
expandarg                                            888     921     +33

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: fix bit-rotten debug infrastructure

DEBUG = 2 output was a bit messed up

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: make popfile() anfter popallfiles() safe

In this example:

ash -c 'readonly x; echo $(command eval x=2)'

evalstring() is called after forkchild(), which calls popallfiles().
On exception, evalstring() will popfile().

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: fix "(sleep 1; exit 3) & sleep 2; echo $?; wait $!; echo $?"

"wait $!" may be just a bit too late: backgrounded $! is gone.
Do not bomb out in this case.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: fix "wait PID"

It was not properly interruptible, and did not update job status
(the exited processes were still thought of as running).

function                                             old     new   delta
process_wait_result                                    -     453    +453
wait_for_child_or_signal                               -     199    +199
run_list                                             996    1002      +6
checkjobs_and_fg_shell                                41      43      +2
builtin_wait                                         328     215    -113
checkjobs                                            516     142    -374
------------------------------------------------------------------------------
(add/remove: 2/0 grow/shrink: 2/2 up/down: 660/-487)          Total: 173 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: use pause(), not sigsuspend(), in wait builtin

Same effect, smaller code

function                                             old     new   delta
dowait                                               463     374     -89

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: fix interactive "command eval STRING" exiting on errors.

This bug is also present in current dash

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [JOBS] Fix dowait signal race

Upstream commit:

    Date: Sun, 22 Feb 2009 18:10:01 +0800
    [JOBS] Fix dowait signal race

    This test program by Alexey Gladkov can cause dash to enter an
    infinite loop in waitcmd.

    #!/bin/dash
    trap "echo TRAP" USR1
    stub() {
        echo ">>> STUB $1" >&2
        sleep $1
        echo "<<< STUB $1" >&2
        kill -USR1 $$
    }
    stub 3 &
    stub 2 &
    until { echo "###"; wait; } do
    echo "*** $?"
    done

    The problem is that if we get a signal after the wait3 system
    call has returned but before we get to INTON in dowait, then
    we can jump back up to the top and lose the exit status.  So
    if we then wait for the job that has just exited, then it'll
    stay there forever.

    I made the original change that caused this bug to fix pretty
    much the same bug but in the opposite direction.  That is, if
    we get a signal after we enter wait3 but before we hit the kernel
    then it too can cause the wait to go on forever (assuming the
    child doesn't exit).

    In fact this is pretty much exactly the scenario that you'll
    find in glibc's documentation on pause().  The solution is given
    there too, in the form of sigsuspend, which is the only way to
    do the check and wait atomically.

    So this patch fixes Alexey's race without reintroducing the old
    bug by converting the blocking wait3 to a sigsuspend.

    In order to do this we need to set a signal handler for SIGCHLD,
    so the code has been modified to always do that.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
I failed to reproduce the bug (it requires precise timing), but it seems real.

function                                             old     new   delta
dowait                                               284     463    +179
setsignal                                            301     326     +25
signal_handler                                        59      76     +17
ash_main                                            1481    1487      +6
localcmd                                             350     348      -2
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 4/1 up/down: 227/-2)            Total: 225 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [SIGNAL] Remove EXSIG

Upstream commit 1:

    Date: Sun, 22 Feb 2009 18:16:13 +0800
    [SIGNAL] Remove EXSIG

    Now that waitcmd no longer uses EXSIG we can remove it.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Upstream commit 2:

    Date: Thu, 2 Oct 2014 21:07:55 +0800
    [ERROR] Set exitstatus in onint

    Currently the exit status when we receive SIGINT is set in evalcommand
    which means that it doesn't always get set.  For example, if you press
    CTRL-C at the prompt of an interactive dash, the exit status is not
    set to 130 as it is in many other Bourne shells.

    This patch fixes this by moving the setting of the exit status into
    onint which also simplifies evalcommand.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Upstream commit 3:

    Date: Fri, 3 Oct 2014 14:07:07 +0800
    [EVAL] Do not clobber exitstatus in evalcommand

    All originators of EXERROR have been setting the exitstatus for
    a while now.  So it is no longer appropriate to set it explicitly
    in evalcommand.

    In fact doing so may cause the original exitstatus to be lost.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Last three coomits:
function                                             old     new   delta
waitcmd                                              186     224     +38
dowait                                               276     284      +8
waitforjob                                           104     107      +3
localcmd                                             348     350      +2
showjobs                                              64      61      -3
forkshell                                            263     260      -3
raise_interrupt                                       93      67     -26
blocking_wait_with_raise_on_sig                       40       -     -40
evalcommand                                         1264    1208     -56
evaltree                                             809     498    -311

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: open-code blocking_dowait_with_raise_on_sig()

There is in fact only one callsite.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: return to DOWAIT_* constants similar to dash, no logic changes

This loses an insignificant optimization, but may allow backporting
of some recent-ish dash fixes.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: delete leftovers from "simplify EOF/newline handling in list parser" commit

This commit should have deleted these two statements:

    commit c0e007663d30f83b0e5e074db34dcffaa8915e99
    Author: Ron Yorston <rmy@pobox.com>
    Date:   Thu Oct 29 11:30:55 2015 +0000
    ash: simplify EOF/newline handling in list parser

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [EXPAND] Fix ifsfirst/ifslastp leak

Upstream commit:

    Date: Wed, 8 Sep 2010 20:07:26 +0800
    [EXPAND] Fix ifsfirst/ifslastp leak

    As it stands expandarg may return with a non-NULL ifslastp which
    then confuses any subsequent ifsbreakup user that doesn't clear
    it directly.

    What's worse, if we get interrupted before we hit ifsfree in
    expandarg we will leak memory.

    This patch fixes this by always calling ifsfree in expandarg
    thus ensuring that ifslastp is always NULL on the normal path.
    It also adds an ifsfree call to the RESET path to ensure that
    memory isn't leaked.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fallout 1:

    Date: Mon, 18 Oct 2010 10:55:42 +0800
    [EXPAND] Fix ifsfirst/ifslastp leak in casematch

    The commit f42e443bb511ed3224f09b4fcf0772438ebdbbfa

        [EXPAND] Fix ifsfirst/ifslastp leak

    revealed yet another ifsfirst/ifslastp leak in casematch.
    Previously it was hidden because ifsfirst/ifslastp was cleared
    unconditionally on entry (which caused the leakage of those
    entries).

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fallout 2:

    Date: Sun, 28 Nov 2010 21:09:51 +0800
    [EXPAND] Free IFS state in evalbackcmd

    On Sun, Nov 07, 2010 at 04:04:20PM -0600, Jonathan Nieder wrote:
    > Herbert Xu wrote:
    > > commit f42e443bb511ed3224f09b4fcf0772438ebdbbfa
    > > Author: Herbert Xu <herbert@gondor.apana.org.au>
    > > Date:   Wed Sep 8 20:07:26 2010 +0800
    > >
    > >     [EXPAND] Fix ifsfirst/ifslastp leak
    >
    > Another puzzle bisecting to f42e443bb.  This one comes from the
    > grub-mkconfig script:
    >
    >  $ sh -c 'datadir=/usr/share; pkgdatadir=${datadir}/`cat`' 2>&1 | cat -A
    >  cat: M-^\^M^F^HM-4^M^F^HM-(^M^F^H: No such file or directory$
    >  cat: M-(^M^F^H: No such file or directory$
    >
    > Still reproducible with 016b529.  I'll try to find time to look into
    > it, but thought you might like to know nevertheless.

    This is the symptom of another leak.  In this case evalbackcmd
    occurs in the middle of an expansion (as it should) but the forked
    child never clears the previous IFS state.

    This patch adds the missing ifsfree call.

    This wasn't as much of a problem as the previously discovered leaks
    since all it means is that the child gets to carry around the parent's
    expansion state and the child is usually short-lived.

Reported-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fallout 3:

    Date: Tue, 15 Mar 2011 16:01:34 +0800
    [EXPAND] Free IFS state after here document expansion

    Here's another bug bisecting to f42e443bb ([EXPAND] Fix
    ifsfirst/ifslastp leak, 2010-09-08).  It was found with the following
    test case, based on the configure script for Tracker:

        dash -x -c '
                <<-_ACEOF
                $@
                _ACEOF
                exec
        ' - abcdefgh
        +
        + exec   ?a
        exec: 1: : Permission denied

    The missing ifsfree call is in expandarg when it returns to openhere
    during here document expansion.

Reported-by: Aurelien Jarno <aurel32@debian.org>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
function                                             old     new   delta
ifsfree                                                -      66     +66
ash_main                                            1490    1495      +5
argstr                                              1154    1159      +5
evalcase                                             275     270      -5
expandarg                                            972     888     -84
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 2/2 up/down: 76/-89)            Total: -13 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: move ifsbreakup() and ifsfree() up

Preparatory patch.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: partially sync with dash on "fork if traps are set" logic

Upstream commit "[EVAL] Force fork if any trap is set, not just on EXIT"
had a similar code as our fix to that bug.
Eliminate some superficial differences.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [SHELL] Expand ENV before using it

Upstream commit:

    Date: Sun, 13 Jul 2008 21:51:52 +0800
    [SHELL] Expand ENV before using it

    Per POSIX ENV needs to undergo parameter expansion.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: comment tweaks, no code changes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: optimize tryexec(): avoid one allocation

There was a bug in tryexec which bbox had fixed in 2003.
dash had a smaller fix in 2007. Copy it. It is smaller,
although it is also more quirky (requires argv[-1] to exist).

Upstream commit 1:

    Date: Mon, 15 Oct 2007 20:24:28 +0800
    [EXEC] Fixed execing of scripts with no hash-bang

    The function tryexec used the original name instead of the path found through
    PATH search.  This patch fixes that.

    Test case:

        trap 'rm -f $TMP' EXIT
        TMP=$(tempfile -s nosuchthing)

        cat <<- EOF > $TMP
                echo OK
        EOF
        chmod u+x $TMP

        cd /
        PATH=${TMP%/*} ${TMP##*/}

    Old result:

        /bin/sh: Can't open filelgY4Fanosuchthing

    New result:

        OK

Upstream commit 2:

    Date: Sun, 23 Dec 2007 11:02:26 +0800
    [EVAL] Fix bad pointer arithmetic in evalcommand

    dash dies on sparc with a SIGBUS due to an arithmetic error introduced
    with commit 03b4958, this patch fixes it.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
function                                             old     new   delta
evalcommand                                         1261    1264      +3
dotcmd                                               321     319      -2
tryexec                                              115      64     -51
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 1/2 up/down: 3/-53)             Total: -50 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [CD] Lookup PWD after going through CDPATH

Upstream commit:

    Date: Mon, 31 Aug 2009 22:06:41 +1000
    [CD] Lookup PWD after going through CDPATH

    On Tue, Jul 14, 2009 at 09:39:03PM +0000, Eric Blake wrote:
    > For the cd command, POSIX 2008 requires that after all pathnames in CDPATH
    > have been tested and failed in step 5, then step 6 interprets the directory
    > argument relative to PWD.  In other words, this demonstrates a bug:
    >
    > $ dash -c 'cd /tmp; mkdir -p foo; CDPATH=oops; cd foo; echo $?; pwd'
    > cd: 1: can't cd to foo
    > 2
    > /tmp
    >
    > while bash gets it correct:
    >
    > $ bash -c 'cd /tmp; mkdir -p foo; CDPATH=oops; cd foo; echo $?; pwd'
    > 0
    > /tmp/foo

    This patch fixes the problem.

Reported-by: Eric Blake <ebb9@byu.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
function                                             old     new   delta
cdcmd                                                667     680     +13

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: [MEMALLOC] Made grabstackblock an inline wrapper for stalloc

Upstream patch:

    Date: Fri, 5 Oct 2007 23:26:45 +0800
    [MEMALLOC] Made grabstackblock an inline wrapper for stalloc

    The function grabstackblock is identical in semantics to stalloc within its
    input constraints.

function                                             old     new   delta
dotcmd                                               319     321      +2
grabstackblock                                        19       5     -14

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>