We should NULL out mac after freeing it, so that the cleanup code
doesn't try to free it again; also if the final key creation fails, we
should avoid freeing ret->comment when we're going to go to that same
cleanup code which will free 'comment' which contains the same pointer.
Thanks to Christopher Staite for pointing these out.
It became bash-dependent in r9229 because I used a bashism to remove
the 'r' from the front of $SVN_REV, but that's not needed any more.
[originally from svn r10281]
[r9229 == bd60f2fc5b]
The symbol alphabet used for encoding ranges of backward distances in
a Deflate compressed block contains 32 symbol values, but two of them
(symbols 30 and 31) have no meaning, and hence it is an encoding error
for them to appear in a compressed block. If a compressed file did so
anyway, this decompressor would index past the end of the distcodes[]
array. Oops.
This is clearly a bug, but I don't believe it's a vulnerability. The
nonsense record we load from distcodes[] in this situation contains an
indeterminate bogus value for 'extrabits' (how many more bits to read
from the input stream to complete the backward distance) and also for
the offset to add to the backward distance after that. But neither of
these can lead to a buffer overflow: if extrabits is so big that
dctx->nbits (which is capped at 32) never exceeds it, then the
decompressor will simply swallow all further data without producing
any output, and otherwise the decompressor will consume _some_ number
of spare bits from the input, work out a backward distance and an
offset in the sliding window which will be utter nonsense and probably
out of bounds, but fortunately will then AND the offset with 0x7FFF at
the last minute, which makes it safe again. So I think the worst that
a malicious compressor can do is to cause the decompressor to generate
strange data, which of course it could do anyway if it wanted to by
sending that same data legally compressed.
[originally from svn r10278]
I've shifted away from using the SVN revision number as a monotonic
version identifier (replacing it in the Windows version resource with
a count of days since an arbitrary epoch), and I've removed all uses
of SVN keyword expansion (replacing them with version information
written out by Buildscr).
While I'm at it, I've done a major rewrite of the affected code which
centralises all the computation of the assorted version numbers and
strings into Buildscr, so that they're all more or less alongside each
other rather than scattered across multiple source files.
I've also retired the MD5-based manifest file system. A long time ago,
it seemed like a good idea to arrange that binaries of PuTTY would
automatically cease to identify themselves as a particular upstream
version number if any changes were made to the source code, so that if
someone made a local tweak and distributed the result then I wouldn't
get blamed for the results. Since then I've decided the whole idea is
more trouble than it's worth, so now distribution tarballs will have
version information baked in and people can just cope with that.
[originally from svn r10262]
The winegcc hack I use for my Coverity builds is currently using a
version of wincrypt.h that's missing a couple of constants I use.
Ensure they're defined by hand, but (just in case I defined them
_wrong_) also provide a command-line define so I can do that only in
the case of Coverity builds.
[originally from svn r10234]
I had initially assumed that, since all of a user's per-connection
subdirectories live inside a top-level putty-connshare.$USER directory
that's not accessible to anyone else, there would be no need to
obfuscate the names of the internal directories for privacy, because
nobody would be able to look at them anyway.
Unfortunately, that's not true: 'netstat -ax' run by any user will
show up the full pathnames of Unix-domain sockets, including pathname
components that you wouldn't have had the access to go and look at
directly. So the Unix connection sharing socket names do need to be
obfuscated after all.
Since Unix doesn't have Windows's CryptProtectMemory, we have to do
this manually, by creating a file of random salt data inside the
top-level putty-connshare directory (if there isn't one there already)
and then hashing that salt with the "user@host" connection identifier
to get the socket directory name. What a pain.
[originally from svn r10222]
This option is available from the command line as '-hostkey', and is
also configurable through the GUI. When enabled, it completely
replaces all of the automated host key management: the server's host
key will be checked against the manually configured list, and the
connection will be allowed or disconnected on that basis, and the host
key store in the registry will not be either consulted or updated.
The main aim is to provide a means of automatically running Plink,
PSCP or PSFTP deep inside Windows services where HKEY_CURRENT_USER
isn't available to have stored the right host key in. But it also
permits you to specify a list of multiple host keys, which means a
second use case for the same mechanism will probably be round-robin
DNS names that select one of several servers with different host keys.
Host keys can be specified as the standard MD5 fingerprint or as an
SSH-2 base64 blob, and are canonicalised on input. (The base64 blob is
more unwieldy, especially with Windows command-line length limits, but
provides a means of specifying the _whole_ public key in case you
don't trust MD5. I haven't bothered to provide an analogous mechanism
for SSH-1, on the basis that anyone worrying about MD5 should have
stopped using SSH-1 already!)
[originally from svn r10220]
I'm about to add a list box which expects to contain some very long
but uninformative strings, and which is also quite vertically squashed
so there's not much room for a horizontal scroll bar to appear in it.
So here's an option in the list box specification structure which
causes the constructed GTKTreeView to use the 'ellipsize' option for
all its cell renderers, i.e. too-long strings are truncated with an
ellipsis.
Windows needs no change, because its list boxes already work this way.
[originally from svn r10219]
I'm about to need to refer to it from a source file that won't
necessarily always be linked against sshpubk.c, so it needs to live
somewhere less specialist. Now it sits alongside base64_encode_atom
(already in misc.c for another reason), which is neater anyway.
[originally from svn r10218]
I'm about to introduce a configuration option which is really a _set_
of string values (rather than an ordered list), and I'm going to
represent it in Conf as a string->string map (since that's a data type
we already support) in which every key which exists at all maps to the
empty string.
This change modifies settings.c so that it can write out such options
into the saved session as a comma-separated list of the key strings
only, rather than the form 'string1=,string2=,string3=' which you'd
get if you just used the existing wmap().
(Reading the result back in turns out not to need a code change - the
existing code already does what we want if it's reading a list of
key=value pairs and one of them doesn't have an = sign at all.)
[originally from svn r10217]
Without this, doing 'Restart Session' on Windows in a session with
sharing enabled but no actual sharing being done would crash, because
the first incarnation of the session would become an upstream and
establish a listening named pipe, which then wouldn't get cleaned up
when the session closed, so the restarted session would try to connect
to it, triggering a call to plug_accepting on a freed sharestate.
[originally from svn r10216]
A user points out that the person who writes a REG_SZ into the
registry can choose whether or not to NUL-terminate it properly, and
if they don't, RegQueryValueEx will retrieve it without the NUL. So if
someone does that to PuTTY's saved session data, then PuTTY may
retrieve nonsense strings.
Arguably this is the fault of whoever tampered with the saved session
data without doing it the same way we would have, but even so, there
ought to be some handling at our end other than silently returning the
wrong data, and putting the NUL back on seems more sensible than
complaining loudly.
[originally from svn r10215]
The IDM_RECONF handler unconditionally calls ldisc_configure to
reconfigure the line discipline for the new echo/edit settings, but in
fact ldisc can be NULL if no session is currently active. (Indeed, the
very next line acknowledges this, by testing it for NULL before
calling ldisc_send!) Thanks to Alexander Wong for the report.
[originally from svn r10214]
This is the same code I previously fixed for failing to check NULL
pointers coming back from ssh_pkt_getstring if the server's KEXINIT
ended early, leading to an embarrassing segfault in place of a fatal
error message. But I've now also had it pointed out to me that the
fatal error message passes the string as %s, which is inappropriate
because (being read straight out of the middle of an SSH packet) it
isn't necessarily zero-terminated!
This is still just an embarrassing segfault in place of a fatal error
message, and not exploitable as far as I can see, because the string
is passed to a dupprintf, which will either read off the end of
allocated address space and segfault non-exploitably, or else it will
find a NUL after all and carefully allocate enough space to format an
error message containing all of the previous junk. But still, how
embarrassing to have messed up the same code _twice_.
[originally from svn r10211]
Now Jacob has reminded me that 'resize-no-truncate' was already on the
wishlist, I notice that it suggested Clear Scrollback should remove
the preserved information off to the right. On the basis that that's
(at least partly) a privacy feature, that seems sensible, so let's do it.
[originally from svn r10210]
We now only truncate a termline to the current terminal width if we're
actually going to modify it. As a result, resizing to a narrower
terminal width and then immediately back again, with no terminal
output in between, should restore the previous screen contents. Only
lines that are actually modified while the terminal is narrow (and
scrolling them around doesn't count as modification) should now be
truncated.
This will be a bit nicer for Unix window resizing (since X lacks the
Windows distinction between mid-drag resize events and the ultimate
drag-release, so can't defer the call to term_size until the latter as
we can on Windows), but mostly it's inspired by having played with a
tiling window manager recently and hence realised that in some
environments windows will be resized back and forth without much
control as a side effect of just moving them around - so it's
generally desirable for resizes to be non-destructive.
[originally from svn r10208]
Robert de Bath points out that failure to remove the timer whose
callback returned FALSE may not have been the cause of runaway timer
explosion; another possibility is that a function called from
timer_trigger()'s call to run_timers() has already set a timer up by
the time run_timers() returns, and then we set another one up on top
of it. Fix that too.
[originally from svn r10206]
Mihkel Ader reports that on that system, timers apparently aren't
getting auto-destroyed when timer_trigger returns FALSE, so the change
in r10181 has caused GTK PuTTY to gradually allocate more and more
timers and consume more and more CPU as they all keep firing.
As far as I can see, this must surely be a bug in GTK 2 (the docs say
that timers _are_ auto-destroyed when their callback returns false),
and it doesn't seem to happen for me with GTK 2.4.23 on Ubuntu 14.04.
However, I'll try to work around it by _explicitly_ destroying each
old timer before we zero out the variable containing its id.
[originally from svn r10202]
[r10181 == e4c4bd2092]
Manfred Schwarb points out that when I moved the autoconf machinery up
from the unix subdirectory to the top level, in r10141, I missed a
couple of lingering $(srcdir)/.. in the make rule for version.o, as a
result of which the automatic checking of the manifest wasn't doing
its thing and tools built from a standard .tar.gz were reporting as
'Unidentified build'.
[originally from svn r10201]
[r10141 == a947c49bec]
We now expect that after the server has sent us CHANNEL_CLOSE, we
should not expect to see any replies to our outstanding channel
requests, and conversely after we have sent CHANNEL_CLOSE we avoid
sending any reply to channel requests from the server. This was the
consensus among implementors discussing the problem on ietf-ssh in
April 2014.
To cope with current OpenSSH's (and perhaps other servers we don't
know about yet) willingness to send request replies after
CHANNEL_CLOSE, I introduce a bug-compatibility flag which is detected
for every OpenSSH version up to and including the current 6.6 - but
not beyond, since https://bugzilla.mindrot.org/show_bug.cgi?id=1818
promises that 6.7 will also implement the new consensus behaviour.
[originally from svn r10200]
Philippe Maupertuis reports that on one particular machine, Windows
causes the named pipe created by upstream PuTTY to be owned by the
Administrators group SID rather than the user's SID, which defeats the
security check in the downstream PuTTY. No other machine has been
reported to do this, but nonetheless it's clearly a thing that can
sometimes happen, so we now work around it by specifying explicitly in
the security descriptor for the pipe that its owner should be the user
SID rather than any other SID we might have the right to use.
[originally from svn r10188]
winshare.c includes ssh.h, but if you defined NO_SECURITY it then
decides to fall back to including the stub noshare.c, which includes
ssh.h again. Fix by moving a block of includes inside the ifdef.
[originally from svn r10184]
It's an old prototype from part way through the development of
connection sharing, which I must have forgotten to fix because by the
time I changed the prototype no platform was using noshare.c any more.
[originally from svn r10183]
Timer objects evaporate when our timer_trigger callback is called, and
therefore we should not remember their ids beyond that time and
attempt to cancel them later. Previous versions of GTK silently
ignored us doing that, but upgrading to Ubuntu Trusty has given me a
version of GTK that complains about it, so let's stop doing it.
[originally from svn r10181]
Martin Prikryl reports that it had the exact same bug as old OpenSSH
(insisting that RSA signature integers be padded with leading zero
bytes to the same length as the RSA modulus, where in fact RFC 4253
section 6.6 says it ought to have _no_ padding), but is recently
fixed. The first version string to not have the bug is reported to be
"mod_sftp/0.9.9", so here we recognise everything less than that as
requiring our existing workaround.
[originally from svn r10161]
questionnaires in unfriendly formats like Excel, apparently in the
mistaken belief that we have some kind of incentive to answer them. I
hope I've managed to identify the key reason why they make this
mistake.
[originally from svn r10156]
It was intended to ensure that people still working with DOS filename
restrictions (or things approximating that, e.g. VFAT) wouldn't have
trouble. Those days are surely long gone, and now zip -k is causing
its own trouble with the new VS2010/VS2012 project files, which
include pairs of filenames that become the same under the zip -k
transformation and hence break the source archive build.
[originally from svn r10155]
Its previous policy of silence made sense before we did builds using
bob (gratuitous output on success caused cronmail) but now it just
makes it hard to spot problems.
[originally from svn r10153]
We were inventing a random number by starting with a long zero bignum
and then setting bits at random, which left an opportunity for the
result to be a non-normalised representation (with a leading zero
word) and hence fail an assertion in bignum_cmp.
[originally from svn r10147]
gcc 4.8 compiling with -O3 gives a new warning about the access to
st->pending at the top of lz77_compress, because for some reason it
thinks there's an out-of-bounds array access there (or perhaps just a
potential one, I'm not really sure which side -Warray-bounds is erring
on). Add an assertion reassuring it that st->npending can't get bigger
than the size of st->pending at the site it's complaining about, and a
second one at the site where st->npending is increased (just in case
my analysis of why it can't happen was wrong!). Also add a comment
explaining the assertions.
[originally from svn r10144]
This rearranges the object files so that they each live alongside
their original source file, instead of all being in the same
directory. To my way of thinking this is a more or less neutral change
(perhaps marginally less tidy), but autotools is apparently beginning
to think it's the One True Way and 1.14 will give a warning if you
don't have it enabled.
[originally from svn r10142]
Previously, 'configure' and its assorted machinery lived in the 'unix'
subdir, because that seemed like a clean place to keep it given that
all the other per-platform Makefiles live in their platform
directories. However, this never sat all that happily with autotools,
and even less so now that it likes to have object file pathnames
parallel source file pathnames: if you have Makefile.am refer to
source files outside its subdir as "../terminal.c" and enable
subdir-objects then any out-of-tree build calls the corresponding
object file "../terminal.o" and so your build products mostly end up
at the directory above your build dir! And as of autotools 1.14 my
previous compensatory bodge of prefixing every source file path in
Makefile.am with "$(srcdir)" has stopped working too.
So I'm giving in to necessity, and changing policy by moving the
configure machinery up to the top level of the source tree where
autotools will be less confused by it. This should not be taken as any
indication of the primacy of the Unix port, only of the recalcitrance
of autotools.
Whereas before we had a trivial script called 'configure' at the top
level that invoked unix/configure to effectively do an 'out-of-tree
build' (for make purposes) at the top level of the source tree, we now
have a similar script in unix/configure. So this _should_ make very
little difference: people who were previously running configure from
the top level should still be able to, and likewise people who were
running it from the unix subdir.
[originally from svn r10141]
I had somehow missed this completely out of the GTK mouse-button
handling and never noticed until now!
Of course, like any other mouse action, if you want it to be handled
locally rather than passed through then you can hold down Shift.
[originally from svn r10139]
On Windows (X mouse reporting of the mouse wheel isn't currently done
by the Unix front end, though I'm shortly about to fix that too) a
mouse wheel event is translated into a virtual button, and we send
both a press and a release of that button to terminal.c, which encodes
both in X mouse reporting escape sequences and passes them on to the
server. This isn't consistent with what xterm does - scroll-wheel
events are encoded _like_ button presses, but differ semantically in
that they don't have matching releases. So we're updating to match
xterm.
[originally from svn r10138]
One of my changes in uxnet.c was outside the NO_IPV6 ifdef, and broke
compilation in the normal mode. Revert all changes in that file and
replace with a reference to the 'step' parameter in the no-IPv6
version of the SOCKADDR_FAMILY macro, so that those warnings are
squelched anyway.
[originally from svn r10136]
[r10135 == e00a004e64]
Simon Tatham
Jacob Nevins