Improve handling of low entropy at start up from /dev/urandom by waiting for
a read(2) call on /dev/random to succeed. Once one such call has succeeded,
a shared memory segment is created and persisted as an indicator to other
processes that /dev/urandom is properly seeded.
This does not fully prevent against attacks weakening the entropy source.
An attacker who has control of the machine early in its boot sequence
could create the shared memory segment preventing detection of low entropy
conditions. However, this is no worse than the current situation.
An attacker would also be capable of removing the shared memory segment
and causing seeding to reoccur resulting in a denial of service attack.
This is partially mitigated by keeping the shared memory alive for the
duration of the process's existence. Thus, an attacker would not only need
to have called call shmctl(2) with the IPC_RMID command but the system
must subsequently enter a state where no instances of libcrypto exist in
any process. Even one long running process will prevent this attack.
The System V shared memory calls used here go back at least as far as
Linux kernel 2.0. Linux kernels 4.8 and later, don't have a reliable way
to detect that /dev/urandom has been properly seeded, so a failure is raised
for this case (i.e. the getentropy(2) call has already failed).
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/9595)
Because of that we can remove OPENSSL_UNISTD and some other
macros from e_os2.h and opensslconf.h
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/9204)
The DEVRANDOM_WAIT feature added a select() call to wait for the
`/dev/random` device to become readable before reading from the
`/dev/urandom` device. It was introduced in commit 38023b87f0
in order to mitigate the fact that the `/dev/urandom` device
does not block until the initial seeding of the kernel CSPRNG
has completed, contrary to the behaviour of the `getrandom()`
system call.
It turned out that this change had negative side effects on
performance which were not acceptable. After some discussion it
was decided to revert this feature and leave it up to the OS
resp. the platform maintainer to ensure a proper initialization
during early boot time.
Fixes#9078
This partially reverts commit 38023b87f0.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/9084)
Use select to wait for /dev/random in readable state,
but do not actually read anything from /dev/random,
use /dev/urandom first.
Use linux define __NR_getrandom instead of the
glibc define SYS_getrandom, in case the kernel headers
are more current than the glibc headers.
Fixes#8215
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/8251)
This function can end up being called during ERR_get_error() if we are
initialising. ERR_get_error() must preserve errno since it gets called via
SSL_get_error(). If that function returns SSL_ERROR_SYSCALL then you are
supposed to inspect errno.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7680)
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/7277)
These headers are internal and never exposed to a cpp compiler, hence no
need for the preamble.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/6554)
Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/4534)
Move struct timeval includes into e_os.h (where the Windows ones were).
Enaure that the include is guarded canonically.
Refer #4271
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4312)
cryptilib.h is the second.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4188)
The socket stuff for DJGPP and VMS was only partially moved to
include/internal/sockets.h...
Remains vxWorks.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4254)
Remove GETPID_IS_MEANINGLESS and osslargused.
Move socket-related things to new file internal/sockets.h; this is now
only needed by four(!!!) files. Compiles should be a bit faster.
Remove USE_SOCKETS ifdef's
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4209)
Apart from ssltest_old.c, the test suite relied on e_os.h for the
OSSL_NELEM macro and nothing else.
The ssltest_old.c also requires EXIT and some socket macros.
Create a new header to define the OSSL_NELEM macro and use that instead.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4186)
warning C4996: 'fileno': The POSIX name for this item is deprecated.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4172)
Move the definition of ossl_assert() out of e_os.h which is intended for OS
specific things. Instead it is moved into internal/cryptlib.h.
This also changes the definition to remove the (int) cast.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4073)
If we have an assert then in a debug build we want an abort() to occur.
In a production build we want the function to return an error.
This introduces a new macro to assist with that. The idea is to replace
existing use of OPENSSL_assert() with this new macro. The problem with
OPENSSL_assert() is that it aborts() on an assertion failure in both debug
and production builds. It should never be a library's decision to abort a
process (we don't get to decide when to kill the life support machine or
the nuclear reactor control system). Additionally if an attacker can
cause a reachable assert to be hit then this can be a source of DoS attacks
e.g. see CVE-2017-3733, CVE-2015-0293, CVE-2011-4577 and CVE-2002-1568.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/3496)
Add OPENSSL_SYS_UEFI to remove unused syslog and uid stuffs for
more clean UEFI build.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2961)
The previous change for Windows wasn't quite right. Corrected to use
%HOME%, %USERPROFILE% and %SYSTEMPROFILE%, in that order.
Also adding the default home for VMS, SYS$LOGIN:
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
This fixes the following error when the CRT debug heap (crtdbg.h) is used:
e_os.h(476): warning C4005: 'strdup': macro redefinition
C:\Program Files (x86)\Windows Kits\10\Include\10.0.10586.0\ucrt\crtdbg.h(319): note: see previous definition of 'strdup'
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1137)
Their only reason to exist was that they didn't exist in VMS before
version 7.0. We do not support such old versions any more.
However, for the benefit of systems that don't get strings.h included
by string.h, we include the former in e_os.h.
RT#4458
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Visual C version from version 2003 complain about certain function
names, for example:
apps\apps.c(2572) : warning C4996: 'open': The POSIX name for this item is deprecated. Instead, use the ISO C++ conformant name: _open. See online help for details.
This adds preprocessor aliases for them in e_os.h.
Additionally, crypto/conf/conf_lib.c needs to include e_os.h to catch
those aliases.
RT#4488
RT#4489
Reviewed-by: Matt Caswell <matt@openssl.org>
* Configure: Replaced -DTERMIO by -DTERMIOS in CFLAGS.
* crypto/bio/bss_dgram.c [WATT32]: Remove obsolete redefinition of
function names: sock_write, sock_read and sock_puts.
* crypto/bio/bss_sock.c [WATT32]: For Watt-32 2.2.11 sock_write,
sock_read and sock_puts are redefined to their private names so
their names must be undefined first before they can be redefined
again.
* crypto/bio/bss_file.c (file_fopen) [__DJGPP__]: Make a copy of the
passed file name and replace the leading dots in the dirname part
and the basname part of the file name, unless LFN is supported.
* e_os.h [__DJGPP__]: Undefine macro DEVRANDOM_EGD. Neither MS-DOS nor
FreeDOS provide 'egd' sockets.
New macro HAS_LFN_SUPPORT checks if underlying file system supports
long file names or not.
Include sys/un.h.
Define WATT32_NO_OLDIES.
* INSTALL.DJGPP: Update URL of WATT-32 library.
Submitted by Juan Manuel Guerrero <juan.guerrero@gmx.de>
RT#4217
Reviewed-by: Andy Polyakov <appro@openssl.org>
e_os.h was defining OPENSSL_NO_DGRAM if OPENSSL_NO_SOCK was defined.
This causes link problems on Windows because the generated .def files
still contain the DGRAM symbols even though they have not been compiled.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Since NDEBUG is defined unconditionally on command line for release
builds, we can omit *_DEBUG options in favour of effective "all-on"
in debug builds exercised though CI.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
