tls: teach it to send AES256-encrypted data

>> CLIENT_HELLO
wrote 50 bytes
insize:0 tail:0
got block len:74
got HANDSHAKE
<< SERVER_HELLO
insize:79 tail:0
got block len:2397
got HANDSHAKE
<< CERTIFICATE
key bytes:271, first:0x00
server_rsa_pub_key.size:256
insize:2402 tail:0
got block len:4
got HANDSHAKE
<< SERVER_HELLO_DONE
>> CLIENT_KEY_EXCHANGE
wrote 267 bytes
master secret:c51df5b1e3b3f57373cdd8ea28e8ce562059636cf9f585d0b89c7f4bacec97e674d7b91f93e7b500cb64637f240c3b78
client_write_MAC_key:3b0b7e2bab241b629c37eb3a3824f09b39fe71a00876b0c8026dda16ef0d2f82
client_write_key:d36e801470ed2f0a8fc886ac25df57ffbe4265d06e3192122c4ef4df1e32fab2
>> CHANGE_CIPHER_SPEC
from secret: c51df5b1e3b3f57373cdd8ea28e8ce562059636cf9f585d0b89c7f4bacec97e674d7b91f93e7b500cb64637f240c3b78
from labelSeed: 636c69656e742066696e6973686564b22e0e6008b8ee218cc02e4a93e4a42b570535f9b57662e262d43b379d125b69
=> digest: a45bfee8ed6507a2a9920d0c
>> FINISHED
before crypt: 5 hdr + 16 data + 32 hash bytes
writing 5 + 16 IV + 64 encrypted bytes, padding_length:0x0f
wrote 85 bytes
insize:9 tail:0
got block len:1
<< CHANGE_CIPHER_SPEC
insize:6 tail:0
got block len:80
< hdr_type:22 ver:3.3 len:80 type:21 len24:9541723 |1591985b...a3da|

The last line is the server's FINISHED response, encrypted.

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

tls: added AES code and made it compile. not used yet

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

tls: massage writing for encryption support; finer-grained debug

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

tls: address one easy FIXME, tidy up comments

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

tls: process CHANGE_CIPHER_SPEC and FINISHED from server

Successfully finishes handshake with test servers using NULL-SHA256
cipher.

The "only" thing remaining before there is a chance
this can actually work with real servers is AES encrypt/decrypt.

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

tls: rearrange function order, improve comments

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

tls: make our send_client_finished() pass server check

sha256 hash should be calculated over incoming handshake packets too!

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

tls: format FINISHED message properly for unencrypted, but sha256 signed mode

Now it at least looks correct, but unfortunately "openssl s_server"
says my hash is wrong.

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

Assorted warning fixes and added a comment, no code changes

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

tls: add sha256 hmac and prf code

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

cryptpw: support "rounds=NNNNNNN$" thing in salts

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

libbb: eliminate redundant variable in sha_crypt

function                                             old     new   delta
sha_crypt                                           1136    1130      -6

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

sha512: use larger constant table only if sha512 is in fact selected

function                                             old     new   delta
sha_K                                                640     256    -384

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

tls: rearrange code, add/improve comments, fix whitespace, no real changes here

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

tls: send CHANGE_CIPHER_SPEC

To "actually implement it" will take more work...

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

whitespace fix

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

tls: format and send CLIENT_KEY_EXCHANGE

$ ./busybox tls kernel.org
insize:0 tail:0
got block len:74
got HANDSHAKE
got SERVER_HELLO
insize:79 tail:4265
got block len:4392
got HANDSHAKE
got CERTIFICATE
entered der @0x8b217a7:0x30 len:1452 inner_byte @0x8b217ab:0x30
entered der @0x8b217ab:0x30 len:1172 inner_byte @0x8b217af:0xa0
skipped der 0xa0, next byte 0x02
skipped der 0x02, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
entered der @0x8b218b4:0x30 len:418 inner_byte @0x8b218b8:0x30
skipped der 0x30, next byte 0x03
entered der @0x8b218c7:0x03 len:399 inner_byte @0x8b218cb:0x00
key bytes:399, first:0x00
entered der @0x8b218cc:0x30 len:394 inner_byte @0x8b218d0:0x02
binary bytes:385, first:0x00
skipped der 0x02, next byte 0x02
binary bytes:3, first:0x01
server_rsa_pub_key.size:384
insize:4397 tail:9
got block len:4
got SERVER_HELLO_DONE
insize:9 tail:0
^C

Next step: send CHANGE_CIPHER_SPEC... and actually implement it.

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

tls: DER length byte 0x81 is actually valid

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

tls: a bit more work

$ ./busybox tls kernel.org
insize:0 tail:0
got block len:74
got HANDSHAKE
got SERVER_HELLO
insize:79 tail:4406
got block len:4392
got HANDSHAKE
got CERTIFICATE
entered der @0x8f7e723:0x30 len:1452 inner_byte @0x8f7e727:0x30
entered der @0x8f7e727:0x30 len:1172 inner_byte @0x8f7e72b:0xa0
skipped der 0xa0, next byte 0x02
skipped der 0x02, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
skipped der 0x30, next byte 0x30
entered der @0x8f7e830:0x30 len:418 inner_byte @0x8f7e834:0x30
skipped der 0x30, next byte 0x03
entered der @0x8f7e843:0x03 len:399 inner_byte @0x8f7e847:0x00
copying key bytes:399, first:0x00
insize:4397 tail:9
got block len:4
got SERVER_HELLO_DONE

Now need to teach it to send ClientKeyExchange...

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

tls: work-in-progress TLS1.2 test applet

function                                             old     new   delta
tls_main                                               -     733    +733
dump                                                   -     230    +230
xread_tls_block                                        -     180    +180
get_der_len                                            -      76     +76
enter_der_item                                         -      70     +70
skip_der_item                                          -      56     +56
get24be                                                -      24     +24
tls_error_die                                          -      19     +19
packed_usage                                       31010   31027     +17
applet_names                                        2549    2553      +4
applet_main                                         1472    1476      +4
applet_suid                                           92      93      +1
applet_install_loc                                   184     185      +1
------------------------------------------------------------------------------
(add/remove: 9/0 grow/shrink: 5/0 up/down: 1415/0)           Total: 1415 bytes

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

wget: fix for brain-damaged HTTP servers. Closes 9471

write(3, "GET / HTTP/1.1\r\nUser-Agent: Wget\r\nConnection: close\r\n\r\n", 74) = 74
shutdown(3, SHUT_WR)    = 0
alarm(900)              = 900
read(3, "", 1024)       = 0
write(2, "wget: error getting response\n", 29) = 29
exit(1)

The peer simply does not return anything. It closes its connection.

Probably it detects wget closing its writing end: shutdown(3, SHUT_WR).

The point it, closing write side of the socket is _valid_ for HTTP.
wget sent the full request, it won't be sending anything more:
it will only receive the response, and that's it.

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

ash: commented-out possible fix for 7694

bash has a feature: it restores termios after a successful wait for
a foreground job which had at least one stopped or sigkilled member.
The probable rationale is that SIGSTOP and SIGKILL can preclude task from
properly restoring tty state. Should we do this too?
A reproducer: ^Z an interactive python:

$ python
Python 2.7.12 (...)
>>> ^Z
    { python leaves tty in -icanon -echo state. We do survive that... }
 [1]+  Stopped                    python
    { ...however, next program (python no.2) does not survive it well: }
$ python
Python 2.7.12 (...)
>>> Traceback (most recent call last):
    { above, I typed "qwerty<CR>", but -echo state is still in effect }
  File "<stdin>", line 1, in <module>
NameError: name 'qwerty' is not defined

The implementation is modeled on bash code and seems to work.
However, I'm not sure we should do this. For one: what if I'd fg
the stopped python instead? It'll be confused by "restored" tty state.

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

ash: 16-bit ->nprocs field is a pain for many CPUs

function                                             old     new   delta
getoptscmd                                           527     540     +13
getjob                                               280     286      +6
makejob                                              278     282      +4
forkchild                                            602     600      -2
waitcmd                                              208     205      -3
showjob                                              382     379      -3
getstatus                                             83      80      -3
dowait                                               408     405      -3
freejob                                               93      89      -4
fg_bgcmd                                             290     286      -4
forkshell                                            260     255      -5
killcmd                                              224     218      -6
jobno                                                 17      11      -6
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 3/10 up/down: 23/-39)           Total: -16 bytes

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

Move FEATURE_AUTOWIDTH config option to two applets which use it

No code changes

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

Move FEATURE_USE_TERMIOS config option to two applets which use it

No code changes

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

libbb: consolidate the code to set termios unbuffered mode

function                                             old     new   delta
set_termios_to_raw                                     -     116    +116
count_lines                                           72      74      +2
powertop_main                                       1458    1430     -28
top_main                                             943     914     -29
more_main                                            759     714     -45
fsck_minix_main                                     2969    2921     -48
conspy_main                                         1197    1135     -62
rawmode                                               99      36     -63
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 1/6 up/down: 118/-275)         Total: -157 bytes

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

hush: reinstate [[ builtin

Mike deleted it:

    commit 39456a18a104b228de240b265bd943251219849d
    Author: Mike Frysinger <vapier@gentoo.org>
    Date:   Sat Mar 28 12:21:57 2009 +0000

    stop lying about [[ test support

probably because it was not properly ifdefed around, and was enabled
even when bash compat is off.

I just tested it - it works:

$ [ *.diff = z.diff ]; echo $?
0
$ [[ *.diff = z.diff ]]; echo $?
1

Of course, not all numerous bash tricks of [[ ]] are implemented...

function                                             old     new   delta
bltins2                                               60      72     +12

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

hush: split bash compatible extensions into separate defines. No code changes

Splitting these options makes it self-documenting about what
bash-compatible features we have.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: split bash compatible extensions into separate defines. No code changes

Splitting these options makes it self-documenting about what
bash-compatible features we have.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

busybox: stop depending on FEATURE_AUTOWIDTH for applet list

Many other appletw don't - they unconditionally use get_terminal_wodth(),
and here the amount of code saved by FEATURE_AUTOWIDTH=n is tiny.

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

top: FEATURE_USE_TERMIOS shouldn't control reading of screen size

It controls whether we take input.

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

more: hardcode FEATURE_USE_TERMIOS=y in this applet; code shrink

function                                             old     new   delta
get_wh                                                 -      27     +27
tcsetattr_tty_TCSANOW                                  -      18     +18
gotsig                                                35      27      -8
more_main                                            835     759     -76
------------------------------------------------------------------------------
(add/remove: 2/0 grow/shrink: 0/2 up/down: 45/-84)            Total: -39 bytes

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

hush: shorten output of "help" builtin

   text    data     bss     dec     hex filename
 891272     485    6856  898613   db635 busybox_old
 891232     485    6856  898573   db60d busybox_unstripped

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

hush: reorder builtins (cd and pwd ought to be close, etc), no code changes

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

init: rename FEATURE_EXTRA_QUIET -> FEATURE_INIT_QUIET

The former name had no INIT anywhere in its name, sounded generic

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

ash: revert "make dot command search current directory first"

Reverts this:
    commit 8ad78e1ec7b2e873953f9f476fb63b5893526c39
    Author: Denis Vlasenko <vda.linux@googlemail.com>
    Date:   Sun Feb 15 12:40:30 2009 +0000
    ash: make dot command search current directory first, as bash does.

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

shells: make hush test optional, rename ASH_BUILTIN_foo -> ASH_foo

This makes hash and ash more symmetrical wrt config menu and config
options.

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

Big cleanup in config help and description

Redundant help texts (one which only repeats the description)
are deleted.

Descriptions and help texts are trimmed.

Some config options are moved, even across menus.

No config option _names_ are changed.

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

Allow FAST_FUNC to be overridden at build time

Busybox uses FAST_FUNC macro to tweak with IA-32 calling conventions in
order to make the function call slightly smaller or slightly faster.
However, when I experiment with GCC's LTO (Link Time Optimization), I
discovered that FAST_FUNC could hinder LTO's optimization so that the
resulting executable become a few bytes larger (than what is compiled
without FAST_FUNC).

This change allows to specify e.g.
CONFIG_EXTRA_CFLAGS="-DFAST_FUNC= -flto"
and compile with LTO without a source code hack.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

kill: optimizations for single-applet build

* Fix a bug with a configuration in which the shell's kill builtin
  would be mistreated as a killall command (i.e. '-q' works, and
  'kill process_name' succeeds when it shouldn't):

    CONFIG_ASH_JOB_CONTROL=y
    CONFIG_HUSH_KILL=y
    # CONFIG_KILL is not set
    CONFIG_KILLALL=y
    # CONFIG_KILLALL5 is not set

* Optimize out unneeded code when the relevant applets are not
  selected.

* Move kbuild lines about shells' kill builtins from Kbuild.src to
  kill.c, to accompany the new HAVE_SH_KILL macro. I hope this would
  make maintanence a little bit easier.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: make echo builtin optional

It's a bit overkill (who would want it off?) but ash already has it
configurable. Let's be symmetric.

Also tweak kbuild logic to use ASH_BUILTIN_ECHO to select echo.o,
not ASH.

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

kill: need not build kill.c when ash's job control is off

ash kill builtin depends on the job control config option.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

make_single_applets.sh: show errors and warnings

While at it, fix one warning in modprobe-small.c

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

unlzma: expand comments, no code changes

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

unlzma: fix erroneous "while" instead of "if". Closes 4682

These parts of the code essentially check whether
stepping back by rep0 goes negative or not.

LZMA SDK from lzma1604.7z has the following in the corresponding places:

... = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)]

Clearly, not loop here.

Technically, "while" here works: if condition is false (because pos
underflowed), it iterates once, adds header.dict_size (a.k.a. dicBufSize),
this makes pos positive but smaller than header.dict_size, and loop exits.

Now we'll just check for negative result of subtraction, which is less code:

function                                             old     new   delta
unpack_lzma_stream                                  2659    2641     -18

(I hope 2 Gbyte+ dictionaries won't be in use soon).

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

unzip: match "Defl:?" display with info-zip; cosmetic code shuffling

Large nested indented code blocks made more sane with a few gotos.

function                                             old     new   delta
unzip_main                                          2491    2519     +28

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

unzip: optional support for xz

function                                             old     new   delta
unzip_main                                          2476    2491     +15

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

unzip: optional support for bzip2 and lzma

function                                             old     new   delta
unzip_main                                          2376    2476    +100
bbunpack                                             750     745      -5

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

modprobe-small: build fixes

Was throwing some build errors:

CONFIG_MODPROBE_SMALL=y
CONFIG_DEPMOD=y
CONFIG_LSMOD=y
CONFIG_MODINFO=y
CONFIG_FEATURE_MODPROBE_SMALL_CHECK_ALREADY_LOADED=y

error: unused variable 'exitcode'

modutils/modprobe-small.c: In function 'modprobe_main':
modutils/modprobe-small.c:1060: error: control reaches end of non-void function

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

modprobe-small: optimizations for single applet build

(I'm requesting for a review first because I fear such an aggressive
change could lead to bugs. While I observe the sizes have reduced, I
haven't test the functionality of each applet after that. So please
test before merging.)

Aggressively cut off unneeded code when the relevant applets are not
built.

Correct dependencies of FEATURE_MODPROBE_SMALL_OPTIONS_ON_CMDLINE and
FEATURE_MODPROBE_SMALL_CHECK_ALREADY_LOADED.

Don't bother with the '-r' option check if only rmmod is built (assume
true then), or when neither rmmod or mobprobe is built (assume false
then).

Size comparison before and after the change (single applet
configuration):

     text    data     bss     dec     hex filename
    34778     946     112   35836    8bfc old/busybox_DEPMOD
    34151     946     112   35209    8989 new/busybox_DEPMOD

    34903     946     112   35961    8c79 old/busybox_INSMOD
    28316     778     112   29206    7216 new/busybox_INSMOD

    35228     962     112   36302    8dce old/busybox_LSMOD
     5011     706      40    5757    167d new/busybox_LSMOD

    34830     946     112   35888    8c30 old/busybox_MODPROBE
    34795     946     112   35853    8c0d new/busybox_MODPROBE

    34718     946     112   35776    8bc0 old/busybox_RMMOD
     7502     714     104    8320    2080 new/busybox_RMMOD

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

Allow 'gzip -d' and 'bzip2 -d' without gunzip or bunzip2

Idea copied from the "ip" applet.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: fix a bug in argv restoration after sourcing a file

if sourced file "shift"ed argvs so that $1 is NULL, restore wasn't done.

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

hush: fix a bug in argv restoration after sourcing a file

if sourced file "shift"ed argvs so that $1 is NULL, restore wasn't done.

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

hush: global_args_malloced is used only if set builtin is enabled

function                                             old     new   delta
run_pipe                                            1623    1635     +12
builtin_source                                       210     222     +12
save_and_replace_G_args                               70      60     -10
builtin_shift                                        132      94     -38
restore_G_args                                        98       -     -98
------------------------------------------------------------------------------
(add/remove: 0/1 grow/shrink: 2/2 up/down: 24/-146)          Total: -122 bytes

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

hush: fix 'defined but not used' warning

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

hush: remove redundand test for ENABLE_HUSH_JOB

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

hush: fix kill builtin without jobs support

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

typo fix

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

hush: conditionalize print_escaped() on EXPORT || TRAP

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

hush: make memleak builtin optional

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

hush: make export builtin optional

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

hush: make umask builtin optional

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

hush: make set/unset builtins optional

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

hush: make read and trap builtins optional

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

hush: kill builtin and kill %jobspec support

Also made it and printf, type and wait builtins optional.

function                                             old     new   delta
builtin_kill                                           -     323    +323
bltins1                                              336     348     +12
builtin_type                                         114     116      +2
------------------------------------------------------------------------------
(add/remove: 1/0 grow/shrink: 2/0 up/down: 337/0)             Total: 337 bytes

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

hush: support %%, %+ and % jobspec (meaning "current job")

function                                             old     new   delta
parse_jobspec                                         83     133     +50
builtin_wait                                         278     283      +5

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

httpd: fix address family for reverse proxy client socket

When httpd proxies a request to another server, it first creates
an AF_INET socket, then resolves the server name to a sockaddr,
then connects to it. This fails if the server name resolves to
an IPv6 address.

This patch ensures that the socket is created with the correct
address family (AF_INET6 if the server resolves to an IPv6 address
and AF_INET otherwise).

Signed-off-by: Laurent Bercot <ska-dietlibc@skarnet.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

build system: no longer prompt for PLATFORM_LINUX option

With the new "select PLATFORM_LINUX" mechanism
(commit e3b1a1fd28558f7a1b3c0ec33313bedb675be8a1), the PLATFORM_LINUX
option alone no longer has any purpose of changing program behavior or
affecting compiled code. So there is no longer need to prompt user of
this config question.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

bunzip2: fix code bloat caused by zcat's seamless magic

This example single-applet configuration would trigger the bloat:

    CONFIG_FEATURE_SEAMLESS_XZ=y
    CONFIG_FEATURE_SEAMLESS_LZMA=y
    CONFIG_FEATURE_SEAMLESS_BZ2=y
    CONFIG_FEATURE_SEAMLESS_GZ=y
    CONFIG_BUNZIP2=y
    # CONFIG_ZCAT is not set
    # All other applets disabled

Here, the resulting "busybox-bunzip2" binary would contain
unpack_gz_stream, unpack_lzma_stream and unpack_xz_stream functions
code. In other words, the gzip, lzma and xz decompressors' code are
linked into the binary unnecessarily.

This happens because SEAMLESS_MAGIC != 0 and compiler is unable
to figure out that SEAMLESS_MAGIC bit is never set.

Fix this by disabling SEAMLESS_MAGIC option flag (setting its value
to 0) when zcat is disabled. This will help the compiler optimize out
bbunpack() and no longer generate open_zipped() function call.

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ash: fix "kill %1" not working if CONFIG_ASH is disabled

ix ash "kill %1" not working if CONFIG_ASH is disabled but
ash is launched by 'sh' or 'bash' name

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

unzip: do not use CDF.extra_len, read local file header. Closes 9536

While at it, shorten many field and variable names.

function                                             old     new   delta
unzip_main                                          2334    2376     +42

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

ash: fix open fds leaking in redirects. Closes 9561

commit e19923f6652a638ac39c84012e97f52cf5a7568e deleted clearredir()
call in shellexec():

ash: [REDIR] Remove redundant CLOEXEC calls
Upstream commit:

Now that we're marking file descriptors as CLOEXEC in savefd, we no longer
need to close them on exec or in setinputfd.

but it missed one place where we don't set CLOEXEC. Fixing this.

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

Automatically disable FEATURE_COMPRESS_USAGE for small builds.

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

ash: explicltly group ash options

This would makes all ash options indented inside "ash" in menuconfig.
It appears that menuconfig has a limit at tracking multiple dependency
lines like this (it looks like a "diamond problem" but I'm not sure if
it is):

               ---ASH <----------
              /                  \       ASH_OPTIMIZE_FOR_SIZE
    !NOMMU <-*----SH_IS_ASH <----[OR] <--ASH_INTERNAL_GLOB
              \                  /       ASH_RANDOM_SUPPORT
               ---BASH_IS_ASH <--        [...]

The kconfig-language document [1] states that:

> If a menu entry somehow depends on the previous entry, it can be
> made a submenu of it. First, the previous (parent) symbol must be
> part of the dependency list and then one of these two conditions
> must be true:
> - the child entry must become invisible, if the parent is set to 'n'

    [BusyBox ash used to satisfy this, but no longer does]

> - the child entry must only be visible, if the parent is visible

    [BusyBox ash configs actually satisfy this, but because of
     "diamond" above this might not be easily detected]

So I found out a direct workaround: by making ash options explicitly
depend on !NOMMU, we can tell menuconfig that rule 2 above is satisfied
without any more tracking.

               ---------------------
              /                     \
    !NOMMU <-*-----ASH <--------     \
              \                 \     \        ASH_OPTIMIZE_FOR_SIZE
               *---SH_IS_ASH <---[OR]-[AND] <--ASH_INTERNAL_GLOB
                \                /             ASH_RANDOM_SUPPORT
                 --BASH_IS_ASH <-              [...]

So all ash options would now be indented under "ash".

[1] "Documentation/kbuild/kconfig-language.txt" in Linux kernel source

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

ntpd: improve postponed hostname resolution

Run the namelookup from the main loop so a misspelled first ntp server
name does not block everything forever.

This fixes the following situation which would block forever:
  $ sudo ./busybox ntpd -dn -p foobar  -p pool.ntp.org
  ntpd: bad address 'foobar'
  ntpd: bad address 'foobar'
  ntpd: bad address 'foobar'
  ...

New behavior:
  ntpd: bad address 'foobar'
  ntpd: sending query to 137.190.2.4
  ntpd: reply from 137.190.2.4: offset:-1.009775 delay:0.175550 status:0x24 strat:1 refid:0x00535047 rootdelay:0.000000 reach:0x01
  ntpd: sending query to 137.190.2.4
  ntpd: reply from 137.190.2.4: offset:-1.009605 delay:0.175461 status:0x24 strat:1 refid:0x00535047 rootdelay:0.000000 reach:0x03
  ntpd: sending query to 137.190.2.4
  ntpd: reply from 137.190.2.4: offset:-1.005327 delay:0.167027 status:0x24 strat:1 refid:0x00535047 rootdelay:0.000000 reach:0x07
  ntpd: sending query to 137.190.2.4
  ntpd: bad address 'foobar'
  ntpd: reply from 137.190.2.4: offset:-1.046349 delay:0.248705 status:0x24 strat:1 refid:0x00535047 rootdelay:0.000000 reach:0x0f

This patch is based on Kaarle Ritvanens work.
http://lists.busybox.net/pipermail/busybox/2016-May/084197.html

function                                             old     new   delta
ntpd_main                                           1061    1079     +18
ntp_init                                             556     560      +4
resolve_peer_hostname                                 81      75      -6
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 2/1 up/down: 22/-6)              Total: 16 bytes

Signed-off-by: Kaarle Ritvanen <kaarle.ritvanen@datakunkku.fi>
Signed-off-by: Natanael Copa <ncopa@alpinelinux.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

unzip: remove now-pointless lseek which returns current position

archival/unzip.c: In function 'read_next_cdf':
archival/unzip.c:271:8: warning: variable 'org' set but
                                 not used [-Wunused-but-set-variable]
  off_t org;
        ^~~

Signed-off-by: Cristian Ionescu-Idbohrn <cristian.ionescu-idbohrn@axis.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

typo fix in config help text

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

unzip: properly use CDF to find compressed files. Closes 9536

function                                             old     new   delta
unzip_main                                          2437    2350     -87

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

modutils: remove special handling of uClibc

Commit 3a45b87ac36f (modutils: support finit_module syscall) introduced
macro finit_module. But it is not defined for uClibc.

The compilation for busybox fails for MIPS with:
With uClibc, we get following build errors:

  modutils/lib.a(modutils.o): In function `bb_init_module':
  modutils.c:(.text.bb_init_module+0x94): undefined reference to `finit_module'
  modutils.c:(.text.bb_init_module+0xa0): undefined reference to `finit_module'

We can just use syscall() without any need for the
uClibc wrappers.

Newer versions of uClibc-ng (>1.0.20) will remove the
module syscall wrappers.

Found via Buildroot autobuilders:
http://autobuild.buildroot.net/results/556/55655daef23788fb3967f801ec8b79e9bed7122b/build-end.log

function                                             old     new   delta
bb_delete_module                                      26      32      +6
bb_init_module                                        90      95      +5
delete_module                                         37       -     -37
init_module                                           53       -     -53
------------------------------------------------------------------------------
(add/remove: 0/4 grow/shrink: 2/0 up/down: 11/-90)            Total: -79 bytes

Reported-by: Rahul Bedarkar <rahul.bedarkar@imgtec.com>
Signed-off-by: Waldemar Brodkorb <wbx@openadk.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

shell: clarify help text of CONFIG_{SH,BASH}_IS_* options

Mention the behavior if user selects CONFIG_SH_IS_ASH but not
CONFIG_ASH. We will be explicit that invocations like "busybox ash"
will not work for such configuration.

Also clarify help text of CONFIG_BASH_IS_* that bash compatibility in
ash is not complete. (It shouldn't be anyway - ash can't support every
bash quirk out there.)

Signed-off-by: Kang-Che Sung <explorer09@gmail.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

udhcp: tweak config order and menu item names

All other applets are listed simply by their name, no reason why
dumpleases doesn't do that.

Group all udhcpd feature options directly after it.

Put "NOT READY" into udhcpc6 item (some users actually tried to use it,
and complained).

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

Fix dependency for IFUPDOWN_UDHCPC_CMD_OPTIONS

Commit a8c696bf09d8151323f6e99348c4bc8989f829c8 makes ifup and ifdown
individually selectable, but forgets to update the dependency to
IFUPDOWN_UDHCPC_CMD_OPTIONS, so it is not selectable anymore.

This patch fixes the dependency by checking for IFUP or IFDOWN, instead
of the obsolete IFUPDOWN.

Also, it drops dependency on UDHCPC: udhcpc on the target system
does not have to come from the _same_ binary.

Signed-off-by: Jörg Krause <joerg.krause@embedded.rocks>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

README_distro_proposal.txt: typo fixes

Signed-off-by: Tito Ragusa <farmatito@tiscali.it>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

appletlib: avoid warning on unused function ingroup

libbb/appletlib.c:558:12: warning: 'ingroup' defined but
                                   not used [-Wunused-function]
 static int ingroup(uid_t u, gid_t g)
            ^~~~~~~

That function is used only if FEATURE_SUID_CONFIG is also enabled.

Signed-off-by: Cristian Ionescu-Idbohrn <cristian.ionescu-idbohrn@axis.com>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

hush: correct exitcode for unterminated ')' - exitcode2.tests testcase

function                                             old     new   delta
parse_stream                                        2595    2609     +14

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

ash: fix error code regression

The commit 'ash,hush: set exit code 127 in "sh /does/not/exist" case'
only partly implemented the dash commit '[ERROR] Allow the originator
of EXERROR to set the exit status'.  This resulted in incorrect error
codes for a syntax error:

   $ )
   $ echo $?
   0

or a redirection error for a special builtin:

   $ rm -f xxx
   $ eval cat <xxx
   $ echo $?
   0

Signed-off-by: Ron Yorston <rmy@pobox.com>
Reported-by: Martijn Dekker <martijn@inlv.org>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>

qemu_multiarch_testing: small improvements

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

tcpudp: define SO_ORIGINAL_DST directly, not via include

musl does not like including linux/netfilter_ipv4.h
(enum / #define collision in two headers, resulting in "3 = 3"
type situation in enum definition).

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

depmod: don't build it if MODPROBE_SMALL=y

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

Update to examples/var_service/README_distro_proposal.txt

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

make_single_applets.sh: SH_IS_HUSH needs special handling too

Well, in fact it works (make oldconfig throws only a warning when it sees
both SH_IS_HUSH=y and SH_IS_NONE=y), but let's be nice.

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

Make it possible to select "sh" and "bash" aliases without selecting ash or hush

The same can be done for msh, but we are probably better off just deleting it
in a next versio or two.

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

Make RMMOD=y by default

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

Make INSMOD=y by default

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

Remove FEATURE_IFUPDOWN_IP_BUILTIN and FEATURE_IFUPDOWN_IFCONFIG_BUILTIN

They merely enable ip or ifconfig/route. There is already a way to do this
on the same menuconfig page.

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

modprobe-small: make applets individually selectable

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

Make DPKG=y and DPKG_DEB=y by default

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

Selinux complains about deprecated selinux/flask.h, stop including it

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

Make FEATURE_USERNAME_COMPLETION=y by default

This matches bash behavior on Fedora

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

make_single_applets.sh: a tool to check single-applet builds

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