ifupdown: rewrite state file atomically

By user's request.
Decided to not use fcntl(F_SETLKW) in lieu of problems with locking
on networked filesystems. The existence of /var/run/ifstate.new
is treated as a write lock. rename() provides atomicity.

function                                             old     new   delta
ifupdown_main                                       1019    1122    +103

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

tcpsvd: don't keep shared fd open if fd limit is reached. closes 9331

Also, much improved help text.

function                                             old     new   delta
packed_usage                                       30652   30851    +199
tcpudpsvd_main                                      1782    1784      +2

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

ash: all blocks in function node copying must be SHELL_ALIGNed

Previous commit probably introduced a bug:
non-matching size calculation in size counting and
actual copying caused by SHELL_ALIGN being applied differently!
This won't bite if string sizes are also SHELL_ALIGNed.
Thus fixing.

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

ash testsuite: fix false positives

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

ash: get rid of two global data variables

function                                             old     new   delta
calcsize                                             126     147     +21
funcstring_end                                         -       4      +4
sizenodelist                                          28      24      -4
funcstringsize                                         4       -      -4
funcstring                                             4       -      -4
funcblocksize                                          4       -      -4
nodeckstrdup                                          48      39      -9
evaltree                                             828     788     -40
------------------------------------------------------------------------------
(add/remove: 1/3 grow/shrink: 1/4 up/down: 25/-105)           Total: -40 bytes
   text    data     bss     dec     hex filename
 943376     916   14292  958584   ea078 busybox_old
 943344     916   14284  958544   ea050 busybox_unstripped

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

ash: comment out free(p) just before _exit, tweak some outdated comments

Planned to sync exitshell up to dash, turned out ours is better :)

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

wget: there is no -s option in GNU Wget 1.18, we should be compatible

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

dhcp: merge leases.c and static_leases.c into dhcpd.c

function                                             old     new   delta
send_offer                                           292     461    +169
udhcpd_main                                         1531    1588     +57
read_leases                                          330     332      +2
add_lease                                            314     312      -2
find_lease_by_mac                                     68       -     -68
find_free_or_expired_nip                             174       -    -174
------------------------------------------------------------------------------
(add/remove: 0/4 grow/shrink: 3/1 up/down: 228/-244)          Total: -16 bytes

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

dhcp: merge most of static_leases.c into dhcpd.c

function                                             old     new   delta
read_staticlease                                     121     222    +101
add_static_lease                                      48       -     -48
log_static_leases                                     68       -     -68
------------------------------------------------------------------------------
(add/remove: 0/2 grow/shrink: 1/0 up/down: 101/-116)          Total: -15 bytes

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

dhcp: merge files.c into dhcpd.c

Added NOINLINE to two function, since my version of gcc would actualy increase
code size otherwise.

I see no size changes.

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

examples: add example of a DHCP server

As usual, by multiplying directories - "dhcpd_eth0", "dhcpd_wlan1"
you can run many servers on different interfaces.

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

hush: add commented-out debug printouts in "memleak" built-in

Allocation addresses of malloc() are jittery,
thought I had a mem leak in hush, but it was malloc variability.

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

hush: fix a memory corruption when exported variable is modified

The construct such as this:

t=1
export t
t=new_value1

had a small probability of momentarily using free()d value.

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

hush: fix exitcode on exec failure with EACCES - should be 126

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

sh testsuite: add tests for exitcode on failure to exec

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

shell: delete all msh tests

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

hush testsuite: fix another false positive

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

hust testsuite: fix a false positive

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

ash testsuite: add most of hust tests which pass for ash

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

ash: [PARSER] Add nlprompt/nlnoprompt helpers

Upstream commit:

    Date: Mon, 29 Sep 2014 22:53:53 +0800
    [PARSER] Add nlprompt/nlnoprompt helpers

    This patch adds the nlprompt/nlnoprompt helpers to isolate code
    dealing with newlines and prompting.

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

whitespace fixes

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

hush: add var4.tests, var5.tests

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

hush: fix var3.tests

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

sh: do not print empty line at the end of "help" output

It's pointless.

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

sh testsuite: sync ash-redir/ and hush-redir/

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

sh testsuite: sync ash-misc/source* and hush-misc/source*

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

sh testsuite: sync ash-vars/ and hush-vars/

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

ash testsuite: update some of variable tests with newer versions from hush

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

sh testsuite: create hush-redir/* and move files around

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

sh testsuite: create hush-heredoc/* and move files around

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

hush testsuite: add many tests from ash testsuite

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

ash: undo "tokname hack"

dash has tokendlist[] array to decide which tokens end lists.
We store it as first byte of each tokname_array[i].

Switch to bit array, name it like dash (tokendlist), drop special
1st byte of tokname_array[i]. This brings us closer to dash, and
shrinks the binary, because many more string aliasing opportunities
are now open:

function                                             old     new   delta
pstrcmp1                                               -      16     +16
readtoken1                                          2852    2858      +6
list                                                 326     327      +1
pstrcmp                                               16      15      -1
tokname                                               45      42      -3
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 2/2 up/down: 23/-4)              Total: 19 bytes
   text    data     bss     dec     hex filename
 943556     916   14292  958764   ea12c busybox_old
 943463     916   14292  958671   ea0cf busybox_unstripped
^^^^^^^ note this!

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

ash: placate gcc: "warning: ! is only applied to the left hand side of =="

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

hush testsuite: add glob_dir.tests

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

ash: fix globbing bugs when using glibc glob()

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

ash: style fixes

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

ash: support "--" in "source" builtin

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

ash: expand: Fixed "$@" expansion when EXP_FULL is false

Upstream commit:

    Date: Thu, 1 Jan 2015 07:53:10 +1100
    expand: Fixed "$@" expansion when EXP_FULL is false

    The commit 3c06acdac0b1ba0e0acdda513a57ee6e31385dce ([EXPAND]
    Split unquoted $@/$* correctly when IFS is set but empty) broke
    the case where $@ is in quotes and EXP_FULL is false.

    In that case we should still emit IFS as field splitting is not
    performed.

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

test: fix compile error in last commit

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

test: memory leak: free group id list

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

ash,hush: add a test that "continue" does not cripple traps

Both shells pass this test.

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

ash: fix return_in_trap1.tests failure

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

ash testsuite: add return_in_trap1.tests

Currently it fails

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

hush: 'return' should have effect earlier

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

ash: [EXPAND] Optimise nulonly away and just use quoted as before

Upstream commit:

    Date: Wed, 8 Oct 2014 20:09:56 +0800
    [EXPAND] Optimise nulonly away and just use quoted as before

    This patch makes a small optimisation by using the same value for
    quoted between evalvar and varvalue by eliminating nulonly and
    passing along quoted instead.

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

ash: [EXPAND] Do not split quoted VSLENGTH and VSTRIM

Upstream patch:

    Date: Wed, 8 Oct 2014 15:42:08 +0800
    [EXPAND] Do not split quoted VSLENGTH and VSTRIM

    Currently VSLENGTH and VSTRIM* are field-split even within quotes.
    This is obviously wrong.  This patch fixes that.

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

ash: [EXPAND] Split unquoted $@/$* correctly when IFS is set but empty

Upstream commit:

    Date: Wed, 8 Oct 2014 15:24:23 +0800
    [EXPAND] Split unquoted $@/$* correctly when IFS is set but empty

    Currently we do not field-split $@/$* when it isn't quoted and IFS
    is set but empty.  This is obviously wrong.  This patch fixes this.

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

hush: add func6.tests

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

ash: fix "return N" not setting $? in loop conditionals

Upstream commit 1:

    Date: Mon, 6 Oct 2014 20:45:04 +0800
    [EVAL] Move common skipcount logic into skiploop

    The functions evalloop and evalfor share the logic on checking
    and updating skipcount.  This patch moves that into the helper
    function skiploop.

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

    Date: Mon, 6 Oct 2014 21:22:43 +0800
    [BUILTIN] Allow return in loop conditional to set exit status

    https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=332954

    When return is used in a loop conditional the exit status will
    be lost because we always set the exit status at the end of the
    loop to that of the last command executed in the body.

    This is counterintuitive and contrary to what most other shells do.

    This patch fixes this by always preserving the exit status of
    return when it is used in a loop conditional.

    The patch was originally written by Gerrit Pape <pape@smarden.org>.

Reported-by: Stephane Chazelas <stephane_chazelas@yahoo.fr>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash testsuite: remove two more inadvertent bashisms

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

ash testsuite: remove two inadvertent bashisms

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

ash: [ERROR] Set exitstatus in onint

Partially backported this commit:

    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>
The part after "if (evalbltin(cmdentry.u.cmd, argc, argv, flags))"
causes testsuite failures in signal handling, so left unchanged.

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

ash: [EVAL] Revert SKIPEVAL into EXEXIT

Upstream commit:

    Date:   Tue Aug 11 20:56:53 2009 +1000
    [EVAL] Revert SKIPEVAL into EXEXIT

    Now that eval handles EV_TESTED correctly, we can remove the
    SKIPEVAL hack and simply use EXEXIT for set -e.

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

ash: [EVAL] Pass EV_TESTED into evalcmd

Upstream commit:

    Date:   Tue Aug 11 20:48:15 2009 +1000
    [EVAL] Pass EV_TESTED into evalcmd

    This patch fixes the case where the eval command is used with
    set -e and as part of a construct that should not cause the
    shell to abort, e.g., as part of the condition of an if statement.

    This is achieved by propagating the EV_TESTED flag into the
    evalstring function through evalcmd.  As this alters the prototype
    of evalcmd it is now invoked explicitly by evalbltin.  The built-in
    infrastructure has been changed to accomodate this special case.

    In order to ensure that the EXIT trap is properly executed this
    patch clears evalskip in exitshell.  This wasn't needed before
    because of the broken way evalstring worked where it always clears
    evalskip when called by minusc.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Although, I failed to create a reproducer for this.

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

ash: fix a thinko in the last commit

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

ash: make internal globbing code selectable from config

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

ash: explain how "command" is handled, and shrink it a bit

function                                             old     new   delta
getoptscmd                                           641     632      -9
commandcmd                                            79      69     -10
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 0/2 up/down: 0/-19)             Total: -19 bytes

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

ash: [VAR] Initialise OPTIND after importing environment

Upstream commit 1:

    Date: Fri, 23 Aug 2013 21:27:42 +1000
    [VAR] Initialise OPTIND after importing environment

    On Sat, Mar 23, 2013 at 01:46:20AM +0000, Chris F.A. Johnson wrote:
    >   According to both the dash man page and the POSIX spec, "When the
    >   shell is invoked, OPTIND is initialized to 1."
    >
    >   However, it actually takes the value of the environment variable
    >   if it exists:
    >
    > $ OPTIND=4 dash -c 'echo "$OPTIND"'
    > 4
    > $ OPTIND=4 bash -c 'echo "$OPTIND"'
    > 1
    > $ OPTIND=4 ksh -c 'echo "$OPTIND"'
    > 1
    > $ OPTIND=4 ksh93 -c 'echo "$OPTIND"'
    > 1

    This patch fixes this by initialising OPTIND after importing the
    environment.

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

    Date: Tue, 7 Oct 2014 22:24:42 +0800
    [VAR] Use setvareq to set OPTIND initially

    There is no need to setvarint to set the initial value of OPTIND
    of one.  This patch switchs to setvareq which also lets us avoid
    an unnecessary memory allocation.

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

ash: [BUILTIN] Merge SKIPFUNC/SKIPFILE and only clear SKIPFUNC when leaving dotcmd

Upstream commit:

    Date: Sat, 9 Jul 2011 22:05:22 +0800
    [BUILTIN] Merge SKIPFUNC/SKIPFILE and only clear SKIPFUNC when leaving dotcmd

    Currently upon leaving a dotcmd the evalskip state is reset so
    if a continue/break statement is used within a dot script it would
    have no effect outside of the dot script.

    This is inconsistent with other shells.

    This patch is based on one by Jilles Tjoelker and only clears
    SKIPFUNC when leaving a dot script.  As a result continue/break
    will remain in effect.

    It also merges SKIPFUNC/SKIPFILE as they have no practical difference.

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

hush: enable "msh is deprecated" message in msh stub

After giving a few more years for everyone to notice and migrate,
can nuke all remains of msh.

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

ash: Avoid overflow for very long variable name

Upstream commit:

    Date: Tue, 3 Jul 2012 17:32:33 +0800
    Avoid overflow for very long variable name

    Otherwise, this:
      $ perl -le 'print "v"x(2**31+1) ."=1"' | dash
    provokes integer overflow

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

ash: [VAR] Sanitise environment variable names on entry

Upstream commit:

    Date: Sat, 25 Feb 2012 15:35:18 +0800
    [VAR] Sanitise environment variable names on entry

    On Tue, Feb 14, 2012 at 10:48:48AM +0000, harald@redhat.com wrote:
    > "export -p" prints all environment variables, without checking if the
    > environment variable is a valid dash variable name.
    >
    > IMHO, the only valid usecase for "export -p" is to eval the output.
    >
    > $ eval $(export -p); echo OK
    > OK
    >
    > Without this patch the following test does error out with:
    >
    > test.py:
    > import os
    > os.environ["test-test"]="test"
    > os.environ["test_test"]="test"
    > os.execv("./dash", [ './dash', '-c', 'eval $(export -p); echo OK' ])
    >
    > $ python test.py
    > ./dash: 1: export: test-test: bad variable name
    >
    > Of course the results can be more evil, if the environment variable
    > name is crafted, that it injects valid shell code.

    This patch fixes the issue by sanitising all environment variable names
    upon entry into the shell.

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

ash: [EVAL] Avoid using undefined handler

Upstream commit:

    Date: Fri, 8 Jul 2011 16:41:24 +0800
    [EVAL] Avoid using undefined handler

    * src/eval.c (evalbltin, evalfun): Set savehandler before calling
    setjmp with the possible "goto *done", where savehandler is used.
    Otherwise, clang warns that "Assigned value is garbage or undefined"
    at the point where "savehandler" is used on the RHS.

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

ash: remove dead stores and unused variables

Upstream commit 1:

    Date: Fri, 8 Jul 2011 16:12:20 +0800
    [MEMALLOC] Avoid gcc warning: variable 'oldstackp' set but not used

    * src/memalloc.c (growstackblock): Remove declaration and set of
    set-but-not-used variable.  Also remove a stray space-before-TAB.

Signed-off-by: Jim Meyering <meyering@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Upstream commit 2:

    Date: Fri, 8 Jul 2011 16:16:11 +0800
    [MEMALLOC] Avoid clang warning about dead store to "size"

    * src/memalloc.c (makestrspace): Remove dead store.

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

ash: [MEMALLOC] Add pushstackmark

Upstream commit:

    Author: Herbert Xu <herbert@gondor.apana.org.au>
    Date:   Sat Oct 6 00:45:52 2007 +0800

    [MEMALLOC] Add pushstackmark

    This patch gets rid of the stack mark tracking hack by allocating a little
    bit of stack memory if we're at risk of planting a stack mark which may be
    grown later.  To do this a new function pushstackmark is added which lets
    the user pick a bigger amount to allocate since some users do that anyway
    after setting a stack mark.

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

hush: fix typo in comment

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

hush: fix interactive input handling of backslash+newline

function                                             old     new   delta
fgetc_interactive                                      -     258    +258
i_peek_and_eat_bkslash_nl                             43      93     +50
static_peek2                                           7       -      -7
setup_string_in_str                                   46      39      -7
setup_file_in_str                                     47      40      -7
file_peek                                             72      52     -20
expand_vars_to_list                                 1167    1143     -24
file_peek2                                            74       -     -74
file_get                                             326      65    -261
------------------------------------------------------------------------------
(add/remove: 1/2 grow/shrink: 1/5 up/down: 308/-400)          Total: -92 bytes

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

hush: G.user_input_buf[] is needed only if line editing is enabled

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

shell testsuite: add trailing newline to var_unbackslash1.tests

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

hush: speed optimizations

Make o_addchr() faster: do not call o_grow_by() each time.
Create i_getch_and_eat_bkslash_nl(), use it instead of peek+getch pair.

function                                             old     new   delta
o_addchr                                              42      54     +12
parse_dollar                                         761     771     +10
o_grow_by                                             48      37     -11
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 3/1 up/down: 24/-11)             Total: 11 bytes

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

hush: fix handling of by backslash-newline in $((arith)) and $(cmd)

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

hush: rework input char buffering to allow more than one-deep peek

...this time with actual hush.c changes too :)

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

ash: fix arithmetic closing )) split by backslash-newline

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

ash: [PARSER] Handle backslash newlines properly after dollar sign

Fixes var_unbackslash1.tests failure.

Upstream commit:

    [PARSER] Handle backslash newlines properly after dollar sign

    On Tue, Aug 26, 2014 at 12:34:42PM +0000, Eric Blake wrote:
    > On 08/26/2014 06:15 AM, Oleg Bulatov wrote:
    > > While playing with sh generators I found that dash and bash have different
    > > interpretations for <slash><newline> sequence.
    > >
    > > $ dash -c 'EDIT=xxx; echo $EDIT\
    > >> OR'
    > > xxxOR
    >
    > Buggy.
    > >
    > > $ dash -c 'echo "$\
    > > (pwd)"'
    > > $(pwd)
    > >
    > > Is it undefined behaviour in POSIX?
    >
    > No, it's well-defined, and dash is buggy.
    ...

    I agree.  This patch should resolve this problem and similar ones
    affecting blackslash newlines after we encounter a dollar sign.

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

hush: rework input char buffering to allow more than one-deep peek

This fixes backslash+newline continuation in
$VAR\
NAME
construct. (ash has a bug there as well).

function                                             old     new   delta
file_peek2                                             -      74     +74
parse_dollar                                         746     773     +27
expand_vars_to_list                                 1143    1167     +24
setup_string_in_str                                   32      46     +14
setup_file_in_str                                     33      47     +14
file_get                                             264     278     +14
static_peek2                                           -       7      +7
file_peek                                             91      72     -19
------------------------------------------------------------------------------
(add/remove: 2/0 grow/shrink: 5/1 up/down: 174/-19)           Total: 155 bytes

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

ash: input: Allow two consecutive calls to pungetc

Upstream commit:

    input: Allow two consecutive calls to pungetc

    The commit ef91d3d6a4c39421fd3a391e02cd82f9f3aee4a8 ([PARSER]
    Handle backslash newlines properly after dollar sign) created
    cases where we make two consecutive calls to pungetc.  As we
    don't explicitly support that there are corner cases where you
    end up with garbage input leading to undefined behaviour.

    This patch adds explicit support for two consecutive calls to
    pungetc.

Reported-by: Jilles Tjoelker <jilles@stack.nl>
Reported-by: Juergen Daubert <jue@jue.li>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In bbox case, bashism >& may need two pungetc() too.

function                                             old     new   delta
pgetc                                                514     555     +41
pushstring                                           114     144     +30
basepf                                                52      76     +24
popstring                                            134     151     +17
parse_command                                       1584    1585      +1
pungetc                                               12       9      -3
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 5/1 up/down: 113/-3)            Total: 110 bytes

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

hush: fix 'eval ""' handling

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

hush: fix ". EMPTY_LINE" not setting $? to 0

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

ash: jobs: Don't attempt to access job table for job %0

Upstream commit:

    jobs: Don't attempt to access job table for job %0

    If job %0 is (mistakenly) specified, an out-of-bounds access to the
    jobtab occurs in function getjob() if num = 0:

            jp = jobtab + 0 - 1

    Fix this by checking that the job number is larger than 0 before
    accessing the jobtab.

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

ash: [EVAL] Make eval with empty arguments return 0

This is a backport of upstream commit:

    [EVAL] Make eval with empty arguments return 0

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

ash: [EVAL] Fix use-after-free in dotrap/evalstring

From upstream:

    [EVAL] Fix use-after-free in dotrap/evalstring

    The function dotrap calls evalstring using the stored trap string.
    If evalstring then unsets that exact trap string then we will end
    up using freed memory.

    This patch fixes it by making evalstring always duplicate the string
    before using it.

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

ash: Remove unused EV_BACKCMD flag

    The original ash defered forking commands in backquotes so builtins
    could be run in the same context as the shell.  This behavior was
    controlled using the EV_BACKCMD to evaltree.

    Unfortunately, as Matthias Scheler noticed in 1999 (NetBSD PR/7814),
    the result was counterintuitive; for example, echo "`cd /`" would
    change the cwd.  So ash 0.3.5 left out that optimization.

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

ash: eval: Return status in eval functions

Backported from dash:

    eval: Return status in eval functions

    The exit status is currently clobbered too early for case statements
    and loops.  This patch fixes it by making the eval functions return
    the current exit status and setting them in one place -- evaltree.

    Harald van Dijk pointed out a number of bugs in the original patch.

function                                             old     new   delta
evalcommand                                         1226    1242     +16
cmdloop                                              383     398     +15
evalfor                                              223     227      +4
evalcase                                             271     275      +4
localcmd                                             348     350      +2
evaltreenr                                           927     928      +1
evaltree                                             927     928      +1
evalsubshell                                         150     151      +1
evalpipe                                             356     357      +1
parse_command                                       1585    1584      -1
evalloop                                             177     164     -13
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 9/2 up/down: 45/-14)             Total: 31 bytes

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

traceroute: cleanup and fixes for packet size calculations

Remove FEATURE_TRACEROUTE_SOURCE_ROUTE: it's off by default, and
source routing is not used in real world.

Tested that "traceroute -n ::1 100" and "traceroute -n 127.0.0.1 100"
both send 100 byte IP packets (this matches what traceroute on Fedora
Rawhide is doing).

function                                             old     new   delta
common_traceroute_main                              3731    3738      +7

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

vi: fix '' command (goto to prev context)

The '' command in vi doesn't currently work because after the first
apostrophe is read, the next character is converted to an integer
between 0 and 25 inclusive (for indexing the array of marks). The
comparison of the converted character with an apostrophe therefore never
succeeds, meaning that '' doesn't do anything.

Based on the patch by Francis Rounds <francis.rounds@4bridgeworks.com>

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

mount: for cifs, dont insert "ip=ADDR" option if user gave it explicitly

This makes it possible to use scoped IPv6 addresses:

mount -t cifs -o ip=<ADDR>%<iface_id> //<ADDR>/test test

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

examples: wpa_supplicant.conf has a wrong field deleted in examples

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

mount: tweak "IPv6 scoped addr" commit

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

libbb: make xmalloc_sockaddr2dotted use NI_NUMERICSCOPE

Gives "mount -t cifs //fe80::6a05:caff:fe3e:dbf5%eth0/test test"
a chance to work: mount must pass "ip=numeric_IPv6%numeric_iface_id"
in the omunt option string. Currently, it does not.

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

libbb/speed_table.c: expand comments

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

typo fix in comment

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

ash: in heredoc code, fix access past the end of allocated memory. Closes 9276

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

ash: fix handling of NULs in $'abc\000def\x00asd'. Closes 9286

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

ifupdown: improve help text, add comment about run-parts error msg

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

libbb: handle \S in /etc/issue

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

AUTHORS: Add myself to AUTHORS

Instead of complaining that my authorship of the rewrite of
the mdev to use /sys/dev is totally gone from the git history
I bravely take credit by adding myself to the AUTHORS file
instead, he he.

Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: use glob() from libc

Adapted from dash.
The "homegrown" glob code is retained (ifdef'ed out).
This changes was inspired by bug 9261, which detected out-of bounds use of heap
for 2098 byte long name in the "homegrown" code. This is still not fixed...

function                                             old     new   delta
expandarg                                            960     982     +22
static.syntax_index_table                             26      25      -1
static.spec_symbls                                    27      26      -1
static.metachars                                       4       -      -4
addfname                                              42       -     -42
msort                                                126       -    -126
expmeta                                              528       -    -528
------------------------------------------------------------------------------
(add/remove: 0/4 grow/shrink: 1/2 up/down: 22/-702)          Total: -680 bytes

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

examples/var_service/supplicant_if: new service example

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

ifupdown: when flushing addresses with "ip", add label %label%

User report:

or our board we setup eth0:0 on a 10.10.10.x/29 netwrok.

The problem is ip addr flush dev eth0:0 removes all ip addresses from
eth0.  You can see this if you run
ip -stat -stat addr flush dev eth0:0

2: eth0    inet 172.27.105.10/22 brd 172.27.107.255 scope global eth0
       valid_lft forever preferred_lft forever
2: eth0    inet 10.10.10.9/29 scope global eth0:0
       valid_lft forever preferred_lft forever
2: eth0    inet6 fe80::a2f6:fdff:fe18:2b13/64 scope link
       valid_lft forever preferred_lft forever

*** Round 1, deleting 3 addresses ***
*** Flush is complete after 1 round ***

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

ash: fix handling of bashism $'xxx' with high-bit chars. Closes 9236

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