| 1 |
EXCEPTIONS |
| 2 |
|
| 3 |
When using exceptions in commands like do_request_service, when should |
| 4 |
we use the passed in exception handler, and when should we use the |
| 5 |
handler in the connection struct? Perhaps we should not have any |
| 6 |
connection->e handler at all? |
| 7 |
|
| 8 |
|
| 9 |
S-EXPRESSIONS |
| 10 |
|
| 11 |
Change the format used for encrypted private key to include the public |
| 12 |
key as well. This way, the private key can be decrypted on-demand. |
| 13 |
This seems cleaner than having the client read both identity and |
| 14 |
identity.pub. |
| 15 |
|
| 16 |
After retiring the complex "streamed" parser, sexp-conv can no longer |
| 17 |
handle arbitrarily large files. That should be fixed some time. |
| 18 |
Perhaps sexp-conv could use a parser of it's own, as that is the only |
| 19 |
program for which it would make sense to understand all features of |
| 20 |
the "advanced" syntax. |
| 21 |
|
| 22 |
|
| 23 |
USERS |
| 24 |
|
| 25 |
Use real ACL:s instead of the authorized keys hack. Implement a |
| 26 |
general user-database abstraction. Also have a look at 'Pluggable Non |
| 27 |
Interactive Authentication Modules' at |
| 28 |
http://www.msu.ru/pniam/pniam.html. |
| 29 |
|
| 30 |
Log a message if user authentication fails because root is not allowed |
| 31 |
to log in. |
| 32 |
|
| 33 |
Create an --telnet-mode/--auth-by-login option to lshd, which will |
| 34 |
accept a none-authentication, but instead of forking a login shell for |
| 35 |
the user, it should fork a /bin/login running as root with options to |
| 36 |
make it display a password prompt. |
| 37 |
|
| 38 |
|
| 39 |
CHARSETS |
| 40 |
|
| 41 |
The conversion from UTF8->unicode->latin1 doesn't handle accented |
| 42 |
characters properly. Probably needs a complete rewrite. |
| 43 |
|
| 44 |
Check out GNU libidn. |
| 45 |
|
| 46 |
|
| 47 |
CONFIGURATION |
| 48 |
|
| 49 |
Figure out what configuration files we need. |
| 50 |
|
| 51 |
|
| 52 |
LOGGING |
| 53 |
|
| 54 |
Use syslog to log server startup, shutdown, and user authentication. |
| 55 |
|
| 56 |
|
| 57 |
SERVER ISSUES |
| 58 |
|
| 59 |
For compatibility with other sshd:s and bash, lshd should set some |
| 60 |
more environment variables when starting user processes. |
| 61 |
|
| 62 |
SSH_CLIENT=<client-ip> <client-port> <server-port> |
| 63 |
SSH_TTY=<tty-name> (if a pty was allocated) |
| 64 |
SSH_ORIGINAL_COMMAND |
| 65 |
|
| 66 |
bash looks at SSH_CLIENT, and reads .bashrc if it is set. |
| 67 |
|
| 68 |
The lshd server should register a signal handler, probably for |
| 69 |
SIGTERM, that causes it to close it's listening socket, and exit after |
| 70 |
current connections are gone (or after a timeout). |
| 71 |
|
| 72 |
|
| 73 |
ALLOCATION |
| 74 |
|
| 75 |
Have the allocator initialize all objects automatically. |
| 76 |
|
| 77 |
Consider adding reference counts to strings. |
| 78 |
|
| 79 |
Unify the debug malloc system to add a magic number to every block |
| 80 |
header, with distinct values for space, strings, objects and lists. |
| 81 |
|
| 82 |
|
| 83 |
OBJECT SYSTEM |
| 84 |
|
| 85 |
Better typechecking of non-heap objects. To do this one would also |
| 86 |
need valid isa-pointers in classes (as classes are statically |
| 87 |
allocated). If this is done properly, the meta feature could also be |
| 88 |
cleaned up a little. |
| 89 |
|
| 90 |
Consider adding "const" methods. One such method to use that for is |
| 91 |
ALIST_GET. |
| 92 |
|
| 93 |
Make sure that the first (self) argument to the method-like macros |
| 94 |
never have side effects. |
| 95 |
|
| 96 |
|
| 97 |
CHANNELS |
| 98 |
|
| 99 |
Implement window-change requests on the server side. |
| 100 |
|
| 101 |
Look at client_session.c:do_client_session_eof() and |
| 102 |
server_session.c:do_eof() and see if they can be unified or perhaps |
| 103 |
even deleted. |
| 104 |
|
| 105 |
The CHANNEL_CLOSE_AT_EOF should be enabled by default. It is the |
| 106 |
correct behaviour for all channels except the server side of sessions. |
| 107 |
|
| 108 |
The following bug should be eliminated: I use lshg to get a large |
| 109 |
file, |
| 110 |
|
| 111 |
src/lshg -v --trace --debug >lsh-test.log.gz sture.lysator.liu.se |
| 112 |
cat 'incoming/lsh-test.log.gz' 2>lshg.log |
| 113 |
|
| 114 |
The file is 0x614ef9 bytes, but only 0x610000 bytes gets through. |
| 115 |
Looking at the lshg log, the final SSH_MSG_DATA packet(s) is missing, |
| 116 |
so the data is lost either in lsh, or in the remote opensshd. |
| 117 |
|
| 118 |
|
| 119 |
CONTROL LANGUAGE |
| 120 |
|
| 121 |
Improve the tail recursiveness of builtin functions. |
| 122 |
|
| 123 |
|
| 124 |
FORWARDING |
| 125 |
|
| 126 |
Generalize the -L and -R options to |
| 127 |
[<source-ip>:]<source-port>:<target-ip>:<target-port> |
| 128 |
where the first component is optional. |
| 129 |
|
| 130 |
Implement other types of forwarding; ssh-agent, UDP, etc. |
| 131 |
|
| 132 |
Review the naming of functions in tcpforward.c; it appears a little |
| 133 |
inconsistent. |
| 134 |
|
| 135 |
There's one known bug in lshd's X11 forwarding: It will start sending |
| 136 |
data on the session before the reply to the client's "shell" or "exec" |
| 137 |
request. |
| 138 |
|
| 139 |
|
| 140 |
RANDOMNESS |
| 141 |
|
| 142 |
Look at prngd, |
| 143 |
http://www.aet.tu-cottbus.de/personen/jaenicke/postfix_tls/prngd.html |
| 144 |
(recommended by Oystein Viggen <[email protected]>), and perhaps |
| 145 |
also at egd. |
| 146 |
|
| 147 |
|
| 148 |
SPKI ISSUES |
| 149 |
|
| 150 |
Use the name "rsa-pkcs1" consistently for all keys, and perhaps |
| 151 |
include the name of the hash function in the signature objects. |
| 152 |
|
| 153 |
Encryption of private keys seems broken, decryption gives the error |
| 154 |
|
| 155 |
lsh: Unknown encryption algorithm for pkcs5v2. |
| 156 |
lsh: Decrypting private key failed. |
| 157 |
|
| 158 |
|
| 159 |
DENIAL OF SERVICE |
| 160 |
|
| 161 |
Implement some limit on the amount of data that may be buffered for |
| 162 |
write on a connection. When the limit is exceeded, the connection |
| 163 |
should be dropped. The problem: if a client connects and sends a lot |
| 164 |
of packets, without ever reading anything from its socket, the server |
| 165 |
will eventually run out of memory. |
| 166 |
|
| 167 |
|
| 168 |
BUILD ISSUES |
| 169 |
|
| 170 |
Have autoconf automatically add -I $prefix/include and -L $prefix/lib ? |
| 171 |
|
| 172 |
Remaking Makefile.am from subdirectories probably cd:s to the wrong |
| 173 |
place, when building outside of the source tree. |
| 174 |
|
| 175 |
Building outside of the source tree creates machine independent |
| 176 |
automatically generated files in the wrong directory (they should be |
| 177 |
in the source tree). |
| 178 |
|
| 179 |
The definition of SBINDIR is broken, in gets the value |
| 180 |
"${exec_prefix}/sbin". |
| 181 |
|
| 182 |
Use separate version.h.in and version.h files, to avoid rebuilding |
| 183 |
everything when the version number is increased. |
| 184 |
|
| 185 |
|
| 186 |
PROTOCOL ISSUES |
| 187 |
|
| 188 |
Update the signal-related messages to conform to |
| 189 |
draft-ietf-secsh-connect-08.txt. |
| 190 |
|
| 191 |
Some SSH Inc sshd server sends USERAUTH_FAILURE messages with a |
| 192 |
trailing ",", like "publickey,password,". Consider hacking |
| 193 |
parse.c:parse_next_atom to allow that. |
| 194 |
|
| 195 |
The server currently allows the client to send random banner text |
| 196 |
before its version string. Such text is not allowed by the protocol |
| 197 |
spec. |
| 198 |
|
| 199 |
Add a timeout, to drop connections that hasn't older than (say) five |
| 200 |
minutes without having authenticated properly. |
| 201 |
|
| 202 |
|
| 203 |
USER INTERFACE ISSUES |
| 204 |
|
| 205 |
Hack do_spki_decrypt() in spki_commands.c to ask again if the provided |
| 206 |
password is incorrect. |
| 207 |
|
| 208 |
Implement an escape char, like ~ with rsh. |
| 209 |
|
| 210 |
The --interface option to lshd can only handle numeric ip addresses. |
| 211 |
|
| 212 |
|
| 213 |
I/O ISSUES |
| 214 |
|
| 215 |
How should we handle POLLERR in io_iter()? |
| 216 |
|
| 217 |
Add a descriptive string to the lsh_fd struct, to aid debugging. |
| 218 |
|
| 219 |
|
| 220 |
TESTING |
| 221 |
|
| 222 |
The testsuite is far from complete. Some parts that it misses are |
| 223 |
|
| 224 |
* Sending and receiving of SSH_MSG_REQUEST_FAILURE and |
| 225 |
SSH_MSG_CHANNEL_FAILURE, and handling of the corresponding |
| 226 |
exceptions. |
| 227 |
|
| 228 |
* Same for SSH_MSG_CHANNEL_OPEN_FAILURE. |
| 229 |
|
| 230 |
* All SSH_MSG_CHANNEL_EXTENDED_DATA (i.e. stderr data) |
| 231 |
|
| 232 |
* i/o exceptions on channels. |
| 233 |
|
| 234 |
* Failed key exchanges. SRP key exchange. |
| 235 |
|
| 236 |
* PTY allocation, requests, etc. |
| 237 |
|
| 238 |
* Userauth banner, failures, passwords,kerberos password helper. |
| 239 |
|
| 240 |
* connection_handle_pending(), connection_ignore_handler, |
| 241 |
connection_fail_handler, connection_unimplemented_handler, |
| 242 |
connection_forward_handler. |
| 243 |
|
| 244 |
* Gatewayed channel requests and global requests. |
| 245 |
|
| 246 |
* lsh-export-key. |
| 247 |
|
| 248 |
* sexp-conv with options --select, --*-hash. |
| 249 |
|
| 250 |
* sexp display types. |
| 251 |
|
| 252 |
* spki, tag prefix and tag any. |
| 253 |
|
| 254 |
* Encrypted provate keys. |
| 255 |
|
| 256 |
* TCP forwarding: cancel-tcpip-forward, various exception. |
| 257 |
|
| 258 |
* utmp logging. |
| 259 |
|
| 260 |
|
| 261 |
MISC |
| 262 |
|
| 263 |
Try to find out why read() sometimes returns -1 and sets errno==EPIPE, |
| 264 |
on sparc-linux, 2.0.33, redhat-5.1, glibc. |
| 265 |
Note: Debian's sparc port uses glibc2.1, rather than 2.0; maybe Red Hat does |
| 266 |
too? Quite a lot of things have changed between 2.0 and 2.1; this might well |
| 267 |
be a documented feature of 2.1. |
| 268 |
|
| 269 |
"lsh" is already used as the name of a shell (included in Debian; |
| 270 |
Description: Baby Shell for Novices with DOS compatible commands). Perhaps |
| 271 |
we need to change our name? |
| 272 |
|
| 273 |
Make it cleaner wrt. more gcc warnings. |
| 274 |
|
| 275 |
Fix desTest.c to use proper declarations of its function pointers, and |
| 276 |
have autoconf check for rusage(). |
| 277 |
|
| 278 |
It seems that it is a little difficult to detect that an asyncronous |
| 279 |
connect() failed; poll() indicates that both reading and writing is |
| 280 |
possible, but write failes (with EPIPE/SIGPIPE) later. Perhaps a |
| 281 |
linux-problem? |
| 282 |
|
| 283 |
Perhaps some of the dh-code in publickey_crypto should be moved to |
| 284 |
keyexchange.c? |
| 285 |
|
| 286 |
Fix the src/symmetric/desTest program. It doesn't complain if the |
| 287 |
desCode is compiled with empty keymap.h, parity.h and rotors.h files. |
| 288 |
|
| 289 |
About POLLHUP: |
| 290 |
|
| 291 |
I've found something that can either be an io.c or a linux kernel bug. |
| 292 |
This time it is the poll() system call, which causes troubles. If an |
| 293 |
AF_UNIX connection is closed by the client, the kernel doesn't set |
| 294 |
POLLIN in revents, instead it sets only POLLHUP. (and therefore the io |
| 295 |
backend doesn't read the last zero length packet) The following patch |
| 296 |
solves the problem for me: |
| 297 |
|
| 298 |
- if (fds[i].revents & POLLIN) |
| 299 |
- READ_FD(fd); |
| 300 |
+ if (fds[i].revents & (POLLIN | POLLHUP)) |
| 301 |
+ READ_FD(fd); |
| 302 |
|
| 303 |
jpoll.h doesn't define POLLHUP, so I added a |
| 304 |
|
| 305 |
#define POLLHUP 0 |
| 306 |
|
| 307 |
line there. I seldom use poll() so I don't know what the correct |
| 308 |
behaviour should be, but I am afraid this is a kernel bug again. |
| 309 |
|
| 310 |
-- |
| 311 |
Bazsi |
| 312 |
|
| 313 |
Find out why opening /dev/ptmx failes occasionally (errno = 19, no |
| 314 |
such device). |
| 315 |
|
| 316 |
Add some workaround for the poll()-bug in linux/glibc-2.07 reported by |
| 317 |
Bazsi. Hmm, this is probably not a bug. I have to find out what the |
| 318 |
right way is to handle the poll conditions POLLERR, POLLHUP and |
| 319 |
POLLERR. |
| 320 |
|
| 321 |
Consider removing the write-attribute from ssh_channel. |
| 322 |
|
| 323 |
Let init_channel take enough arguments to initialize the window- and |
| 324 |
packet-size fields properly. It's too easy to forget them. |
| 325 |
|
| 326 |
Add const to arguments of type struct exception_handler *? |
| 327 |
|
| 328 |
Do something reasonable in do_buffered_read() if want_read is cleared. |
| 329 |
|
| 330 |
The supplied argp is compiled even if it isn't used. |
| 331 |
|
| 332 |
Move miscellaneous lsh_string-functions from format.c to a separate |
| 333 |
file. |
| 334 |
|
| 335 |
Consider moving SPKI tuple-related functions to a separate file. |
| 336 |
|
| 337 |
With SPKI, consider whether or not we should require any specific |
| 338 |
order of subexpressions. The most central function for this is |
| 339 |
sexp_get_un. |
| 340 |
|
| 341 |
Extend the GABA: (expr ...) constructions with a single parameter |
| 342 |
could be used directly as a command_simple. Extend gaba.scm to do that |
| 343 |
automatically. Extend the (exp ...) construction to take a return |
| 344 |
type, and automatically cast the result to that type. |
| 345 |
|
| 346 |
Add -f, to fork and exit when userauthentication is done and any |
| 347 |
tunneling is set up. |
| 348 |
|
| 349 |
Consider using dynamically allocated strings for exception messages. |
| 350 |
Without this, it is impossible to use non-constant messages, for |
| 351 |
instance messages supplied by the peer. |
| 352 |
|
| 353 |
Try to unify the handling of queued channel requests and global |
| 354 |
requests. |
| 355 |
|
| 356 |
Get rid of the double close loop in io_iter. Perhaps start using a |
| 357 |
doubly linked list for fds? Add a backend-pointer to fd:s? Use an |
| 358 |
extra stack of fds that should be closed? |
| 359 |
|
| 360 |
If we add the backend field to lsh_fd objects, the code for listen |
| 361 |
could also be simplified, and perhaps we can get rid of the backend |
| 362 |
pointer in some other places as well. |
| 363 |
|
| 364 |
Exceptions from listen are handled badly in several places. |
| 365 |
|
| 366 |
Should lsh fail if some port forwardings fails? In particular, |
| 367 |
|
| 368 |
$ lsh -R... -N |
| 369 |
|
| 370 |
is pretty useless if the server doesn't let us bind the remote port. |
| 371 |
|
| 372 |
Add reasonable limits to all calls to parse_bignum(), sexp2bignum_u() |
| 373 |
and sexp_get_un(). |
| 374 |
|
| 375 |
There are dsa-specific details in many places, lsh.c |
| 376 |
server_publickey.c, server_authorization.c, server_keyexchange.c. Try |
| 377 |
to write more generic functions that can deal with both dsa and rsa. |
| 378 |
|
| 379 |
Review the default algorithm preference list in |
| 380 |
algorithms.c:default_crypto_algorithms(). Perhaps make the list more |
| 381 |
conservative, and add a "pseudo-algorithm" all to include all |
| 382 |
supported algorithms in the list? |
| 383 |
|
| 384 |
Use static objects for crypto algorithms with fixed key sizes and |
| 385 |
other parameters. |
| 386 |
|
| 387 |
Replace most defines with enums, for improved type checking. |
| 388 |
|
| 389 |
Write more testcases for rsa. |
| 390 |
|
| 391 |
Perhaps change the LOOKUP_VERIFIER method to return an spki_subject? |
| 392 |
|
| 393 |
Make connection_unlock install a callout that calls |
| 394 |
connection_handle_pending(). |
| 395 |
|
| 396 |
Consider how much we really need to differentiate between |
| 397 |
rsa-pkcs1-sha1 / rsa-pkcs1-md5 and between spki-sign-rsa / |
| 398 |
spki-sign-dss. |
| 399 |
|
| 400 |
Building fails if srp support is disabled. The building of srp-gen |
| 401 |
must be conditionalized. |
| 402 |
|
| 403 |
lshg -L doesn't work. The lsh process dies on the first connection to |
| 404 |
the forwarded port. Hmm. Or rather, it seems that the remote server |
| 405 |
(opensshd) disconnects. |
| 406 |
|
| 407 |
A successful tcp-forwarding (set up using lsh -L... -N): |
| 408 |
|
| 409 |
DEBUG: Sent CHANNEL_OPEN (size 72 = 0x48) |
| 410 |
00000000: 5a0000000c6469726563742d74637069 Z....direct-tcpi |
| 411 |
00000010: 70000000000000271000007f9c000000 p......'........ |
| 412 |
00000020: 126b6f6d2e6c797361746f722e6c6975 .kom.lysator.liu |
| 413 |
00000030: 2e73650000131e000000093132372e30 .se........127.0 |
| 414 |
00000040: 2e302e3100000dc8 .0.1.... |
| 415 |
|
| 416 |
When using lsh -G -N; lshg -L... -n: |
| 417 |
|
| 418 |
DEBUG: gateway received CHANNEL_OPEN (size 72 = 0x48) |
| 419 |
00000000: 5a0000000c6469726563742d74637069 Z....direct-tcpi |
| 420 |
00000010: 70000000000000271000007f9c000000 p......'........ |
| 421 |
00000020: 126b6f6d2e6c797361746f722e6c6975 .kom.lysator.liu |
| 422 |
00000030: 2e73650000131e000000093132372e30 .se........127.0 |
| 423 |
00000040: 2e302e3100000dc3 .0.1.... |
| 424 |
|
| 425 |
handle_connection: Received packet of type 90 (CHANNEL_OPEN) |
| 426 |
Allocated local channel number 0 |
| 427 |
Allocated local channel number 0 |
| 428 |
Registering local channel 0. |
| 429 |
DEBUG: Sent CHANNEL_OPEN (size 29 = 0x1d) |
| 430 |
00000000: 5a0000000c6469726563742d74637069 Z....direct-tcpi |
| 431 |
00000010: 70000000000000271000007f9c p......'..... |
| 432 |
|
| 433 |
|
| 434 |
|
| 435 |
make_char_classes seems not to work with guile-1.3.4. Bug in guile? |
| 436 |
|
| 437 |
: ERROR: Two copies of (leaf #f (#\377)) (leaf #f (#\377)) |
| 438 |
: make[1]: *** [sexp_table.h] Error 2 |
| 439 |
|
| 440 |
or |
| 441 |
|
| 442 |
: ERROR: Stack overflow |
| 443 |
: make: *** [sexp_table.h] Error 2 |
| 444 |
|
| 445 |
Figure out whether or not to make --cvs-workaround the default. |
| 446 |
According to Neil Jarram, "lsh-1.1.2 with --cvs-workaround=oe always |
| 447 |
succeeds." Update: According to rms, --cvs-workaround=e works, and |
| 448 |
--cvs-workaround=o doesn't, so it seems that it's stderr that causes |
| 449 |
problems. So we need to avoid setting stderr into non-blocking mode. |
| 450 |
|
| 451 |
Add a keep-alive option, which sends SSH_MSG_IGNORE at regular |
| 452 |
intervals. |
| 453 |
|
| 454 |
The lshd --daemonic option doesn't work with the |
| 455 |
RANDOM_POLL_BACKGROUND. |
| 456 |
|
| 457 |
: [email protected] writes: |
| 458 |
: |
| 459 |
: > When I run |
| 460 |
: > |
| 461 |
: > lshd --daemonic --ssh1-fallback=/usr/sbin/sshd |
| 462 |
: > |
| 463 |
: > It dies on first connection and says : |
| 464 |
: > |
| 465 |
: > Apr 25 04:49:02 subversions lshd[25773]: Background randomness poll failed. |
| 466 |
: > Apr 25 04:49:02 subversions lshd[25773]: lshd: Could not get enough entropy from the environment. |
| 467 |
: |
| 468 |
: Strange. A guess on what's happening: |
| 469 |
: |
| 470 |
: 1. lshd starts a background process for collecting randomness, |
| 471 |
: |
| 472 |
: 2. next it forks and exits to get into the background, |
| 473 |
: |
| 474 |
: 3. it tries to waitpid the process created in 1, but it's no longer |
| 475 |
: the parent of that process, so waiting fails. |
| 476 |
|
| 477 |
When a forwarding created with lshg -L ... goes down (at the local |
| 478 |
end, I think), the channel is not taken down properly, and there are a |
| 479 |
lot of "lsh: write_buffer: Attempt to write data to closed buffer." |
| 480 |
warnings by lsh: |
| 481 |
|
| 482 |
: bash-2.03$ lshg: Exiting: Connection reset by peer |
| 483 |
: lsh: write_buffer: Attempt to write data to closed buffer. |
| 484 |
: lsh: write_buffer: Attempt to write data to closed buffer. |
| 485 |
|
| 486 |
To fix this, we must put some resources on the lsh's |
| 487 |
gateway-connection's resource list, that closes the chained |
| 488 |
connections. |
| 489 |
|
| 490 |
According to rms, lsh is significantly slower than ssh when logging in |
| 491 |
over a slow connection. Figure out why. |