[backport to 3.15.2: drop the bit for "no_optgo", which did not exist
yet in that version]

When a user sends a name length value of 0xffffffff, nbd-server will try
to allocate one more byte for the \0 at the end, but that will result in
an integer overflow and a malloc(0), with the resulting write being to a
dangling pointer.

Fix by constraining the string size to 4096 bytes, as recommended by the
protocol standard.

[backport to 3.15.2: drop the NBD_OPT_INFO/NBD_OPT_GO part, which did
not exist yet in that version]

CVE-2022-26495

Reported-By: 王多 <duo.wang@chaitin.com>
Signed-Off-By: Wouter Verhelst <w@uter.be>

Although the configuration file specified a validity of 3650 days (i.e.,
almost 10 years), the actual certificate in this directory was generated
with a certificate validity of 1 year. It expired on april 19th, 2017
(i.e., 5 days ago), causing the test suite to fail from now on.

Fix by regenerating the certificate (not the keys) of the certificate
with the actual configuration in this directory.

Signed-off-by: Wouter Verhelst <w@uter.be>

The "if GNUTLS" conditional was nested inside an "if CLIENT"
conditional. That made sense for the client bits, but not for the server
bits.

Move those out into their rightful place.

Vagrant Cascadian authored Mar 21, 2017 and Wouter Verhelst committed Mar 21, 2017

The fix for #840612 only fixed the ipv4 case, the following should fix
it for ipv6 as well

Vagrant Cascadian authored Mar 20, 2017 and Wouter Verhelst committed Mar 21, 2017

LTSP uses NFS-like export names, and expects a "/" in the name to be part of an
export name, rather than it being the delimiter between a port and an export
name.

Since NFS-like export names don't have a port number, we can
special-case the ":/" format, and use the : as a delimiter then.

Tagging nbd 3.15.2

AC_SUBST'ing the nbd-client program worked when we didn't need to
compile the client twice. Now that we do, this makes us try to compile
the minimal (i.e., non-TLS) client even if we're on a system where the
client can't be compiled. This is obviously wrong.

Fix it by using AM_CONDITIONAL a second time, and nesting the whole
GnuTLS conditional below that.

Signed-off-by: Wouter Verhelst <w@uter.be>

The introduction of multithread handling of requests failed to consider
what happens with copyonwrite enabled. We would seek, then read (or
write), a file.

If we have copyonwrite enabled, this fails badly in that two threads can
race against eachother for the current file pointer. Instead, we should
use pread() or pwrite() to read from offset without relying on the
current file offset, as explained in #43.

Remove the myseek() function to ensure we don't make use of it anymore,
and fix up all callers.

The only exception where we can still use lseek() safely is in the
open_treefile() function, since there we hold a mutex and hence we can't
race against other threads.

TODO: test whether this fixes the problem.

Signed-off-by: Wouter Verhelst <w@uter.be>

AC_SUBST'ing the nbd-client program worked when we didn't need to
compile the client twice. Now that we do, this makes us try to compile
the minimal (i.e., non-TLS) client even if we're on a system where the
client can't be compiled. This is obviously wrong.

Fix it by using AM_CONDITIONAL a second time, and nesting the whole
GnuTLS conditional below that.

Signed-off-by: Wouter Verhelst <w@uter.be>

The introduction of multithread handling of requests failed to consider
what happens with copyonwrite enabled. We would seek, then read (or
write), a file.

If we have copyonwrite enabled, this fails badly in that two threads can
race against eachother for the current file pointer. Instead, we should
use pread() or pwrite() to read from offset without relying on the
current file offset, as explained in #43.

Remove the myseek() function to ensure we don't make use of it anymore,
and fix up all callers.

The only exception where we can still use lseek() safely is in the
open_treefile() function, since there we hold a mutex and hence we can't
race against other threads.

TODO: test whether this fixes the problem.

Signed-off-by: Wouter Verhelst <w@uter.be>

Closes #45

The options are supposed to be alphabetically ranked, but they weren't
anymore. Fix that.

Signed-off-by: Wouter Verhelst <w@uter.be>

Signed-off-by: Wouter Verhelst <w@uter.be>

GnuTLS <= 3.3 does not initialize at load time, and requires an explicit
call to gnutls_global_init(). Newer versions don't require it, but don't
break if the call is absent, either.

Add it for the benefit of older versions of GnuTLS

Signed-off-by: Wouter Verhelst <w@uter.be>

Tagging 3.15.1

We merged the WRITE_ZEROES support from a branch, where it was written
before STARTTLS was implemented, without properly adapting it to the way
things are now done in the light of STARTTLS, resulting in the
WRITE_ZEROES function sending its reply header in the clear,
unconditionally.

Fix by using the correct function instead.

Tagging nbd 3.15

Signed-off-by: Wouter Verhelst <w@uter.be>