Commit Graph

90 Commits

Author SHA1 Message Date
Denys Vlasenko
636c3b627c tls: merge sha1 and sha256 hmac functions
function                                             old     new   delta
hmac_begin                                             -     196    +196
hmac_sha256                                           61      68      +7
hmac                                                 250      87    -163
hmac_sha256_begin                                    190       -    -190
------------------------------------------------------------------------------
(add/remove: 1/1 grow/shrink: 1/1 up/down: 203/-353)         Total: -150 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-04-03 17:43:44 +02:00
Denys Vlasenko
0ec4d08ea3 tls: covert i/o loop from using select() to poll()
function                                             old     new   delta
tls_run_copy_loop                                    377     282     -95

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-02-16 16:51:18 +01:00
Denys Vlasenko
c31b54fd81 tls: fold AES CBC en/decryption into single functions
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-02-04 16:23:49 +01:00
Denys Vlasenko
5b05d9db29 wget/tls: session_id of zero length is ok (arxiv.org responds with such)
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-02-03 18:23:52 +01:00
Denys Vlasenko
89193f985b tls: can download kernels now :)
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-24 18:08:07 +01:00
Denys Vlasenko
1500b3a50d tls: if got CERTIFICATE_REQUEST, send an empty CERTIFICATE
wolfssl test server is not satisfied by an empty one,
but some real servers might be.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-24 17:06:10 +01:00
Denys Vlasenko
49ecee098d tls: add 2nd cipher_id, TLS_RSA_WITH_AES_128_CBC_SHA, so far it doesn't work
Good news that TLS_RSA_WITH_AES_256_CBC_SHA256 still works with new code ;)

This change adds inevitable extension to have different sized hashes and AES key sizes.
In libbb, md5_end() and shaX_end() are extended to return result size instead of void -
this helps *a lot* in tls (the cost is ~5 bytes per _end() function).

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-24 16:00:54 +01:00
Denys Vlasenko
7a18b9502a tls: reorder tls_handshake_data fields for smaller size, tweak comments
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-23 16:37:04 +01:00
Denys Vlasenko
b5bf1913d3 tls: send EMPTY_RENEGOTIATION_INFO_SCSV in our client hello
Hoped this can make cdn.kernel.org to like us more. Nope.
While at it, made error reporting more useful.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-23 16:12:17 +01:00
Denys Vlasenko
9492da7e63 tls: set TLS_DEBUG to 0; placate a gcc indentation warning
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-23 01:15:13 +01:00
Denys Vlasenko
9a647c326a separate TLS code into a library, use in in wget
A new applet, ssl_client, is the TLS debug thing now.
It doubles as wget's NOMMU helper.
In MMU mode, wget still forks, but then directly calls TLS code,
without execing.

This can also be applied to sendmail/popmail (SMTPS / SMTP+starttls support)
and nc --ssl (ncat, nmap's nc clone, has such option).

function                                             old     new   delta
tls_handshake                                          -    1691   +1691
tls_run_copy_loop                                      -     443    +443
ssl_client_main                                        -     128    +128
packed_usage                                       30978   31007     +29
wget_main                                           2508    2535     +27
applet_names                                        2553    2560      +7
...
xwrite_encrypted                                     360     342     -18
tls_main                                            2127       -   -2127
------------------------------------------------------------------------------
(add/remove: 4/1 grow/shrink: 13/8 up/down: 2351/-2195)       Total: 156 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-23 01:08:16 +01:00
Denys Vlasenko
f6e20724d4 tls: reorder tls_state fields for smaller offsets
function                                             old     new   delta
xwrite_encrypted                                     363     360      -3
xwrite_and_update_handshake_hash                     117     114      -3
tls_xread_handshake_block                             72      69      -3
tls_error_die                                        211     202      -9
tls_get_outbuf                                        64      49     -15
tls_main                                            2163    2127     -36
tls_xread_record                                     702     639     -63
------------------------------------------------------------------------------
(add/remove: 0/0 grow/shrink: 0/7 up/down: 0/-132)           Total: -132 bytes

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-21 02:08:34 +01:00
Denys Vlasenko
dd2577f21a tls: send SNI in the client hello
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 22:48:41 +01:00
Denys Vlasenko
0af5265180 tls: check size on "MAC-only, no crypt" code path too
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 21:23:10 +01:00
Denys Vlasenko
54b927d78b tls: AES decrypt does one unnecessary memmove
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 21:19:38 +01:00
Denys Vlasenko
3916139ac4 tls: make input buffer grow as needed
As it turns out, it goes only up to "inbuf_size:4608"
for kernel.org - fixed 18kb buffer was x4 larger than necessary.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 20:27:06 +01:00
Denys Vlasenko
38972a8df1 tls: improve i/o loop
With tls_has_buffered_record(), entire kernel.org response
is printed at once, without 6 second pause to see its delayed EOF.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 19:11:14 +01:00
Denys Vlasenko
e7863f394e tls: was psAesDecrypt'ing one block too many, trashing buffered data
For the first time

printf "GET / HTTP/1.1\r\nHost: kernel.org\r\n\r\n" | ./busybox tls kernel.org

successfully reads entire server response and TLS shutdown.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 18:04:04 +01:00
Denys Vlasenko
19e695ebad tls: do not use common_bufsiz
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 14:27:58 +01:00
Denys Vlasenko
a0aae9f714 tls: decode alerts and in particular, EOF alert.
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 14:12:10 +01:00
Denys Vlasenko
abbf17abcc tls: add the i/o loop - largish rework of i/o buffering
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-20 03:15:09 +01:00
Denys Vlasenko
432f1ae2ff tls: tested PSTM_X86_64, not enabling it - too large
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-19 16:32:38 +01:00
Denys Vlasenko
cccf8e735d tls: teach it to decrypt AES256-encrypted data
This adds decryption only.
There is no MAC verification, code simply throws away MAC.

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-19 00:20:45 +01:00
Denys Vlasenko
a9e1866806 tls: trim comments
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-18 21:00:23 +01:00
Denys Vlasenko
b5dfc3dfd6 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>
2017-01-18 20:37:24 +01:00
Denys Vlasenko
b7e9ae6e9f tls: added AES code and made it compile. not used yet
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-18 17:20:27 +01:00
Denys Vlasenko
c8ba23bcec tls: massage writing for encryption support; finer-grained debug
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-18 06:45:50 +01:00
Denys Vlasenko
5d1662ea1c tls: address one easy FIXME, tidy up comments
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-17 18:17:27 +01:00
Denys Vlasenko
e69d78c038 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>
2017-01-17 17:24:11 +01:00
Denys Vlasenko
fe0588df3b tls: rearrange function order, improve comments
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-17 17:04:24 +01:00
Denys Vlasenko
e2cb3b990f 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>
2017-01-17 16:53:36 +01:00
Denys Vlasenko
9a6897a48a 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>
2017-01-16 23:26:33 +01:00
Denys Vlasenko
936e83e694 tls: add sha256 hmac and prf code
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-16 04:25:01 +01:00
Denys Vlasenko
3f8ecd933a tls: rearrange code, add/improve comments, fix whitespace, no real changes here
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-15 14:16:51 +01:00
Denys Vlasenko
c5540d61f6 tls: send CHANGE_CIPHER_SPEC
To "actually implement it" will take more work...

Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-15 02:17:03 +01:00
Denys Vlasenko
f78ad0938b whitespace fix
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-15 00:18:22 +01:00
Denys Vlasenko
11d0096516 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>
2017-01-15 00:12:42 +01:00
Denys Vlasenko
2a17d1fc9b tls: DER length byte 0x81 is actually valid
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
2017-01-14 22:38:25 +01:00
Denys Vlasenko
b1003f7019 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>
2017-01-14 13:57:16 +01:00
Denys Vlasenko
ceff6b0ea9 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>
2017-01-14 12:49:32 +01:00