We accept both "\u0000socket-name" and "@socket-name" as abstract
unix sockets. The first one is passed to the kernel pristine,
while the second is transformed '@'->'\0'.
The commit that added support for unix sockets accepts both
variants, but we internally stored it in the same way, using
"\u0000..." for both.
We want to support abstract sockets transparently to the user, so
that if the user configures unitd with '@', if we receive a query
about the current configuration, the user should see the same
exact thing that was configured. So, this commit avoids the
transformation in the internal state file, storing user input
pristine, and we only transform the '@' for a string that will
be used internally (not user-visible).
This commit (indirectly) fixes a small bug, where we created
abstract sockets with a trailing '\0' in their name due to calling
twice nxt_sockaddr_parse() on the same string. By calling that
function only once with each copy of the string, we have fixed that
bug.
Unix domain sockets are normally backed by files in the
filesystem. This has historically been problematic when closing
and opening again such sockets, since SO_REUSEADDR is ignored for
Unix sockets (POSIX left the behavior of SO_REUSEADDR as
implementation-defined, and most --if not all-- implementations
decided to just ignore this flag).
Many solutions are available for this problem, but all of them
have important caveats:
- unlink(2) the file when it's not needed anymore.
This is not easy, because the process that controls the fd may
not be the same process that created the file, and may not have
file permissions to remove it.
Further solutions can be applied to that caveat:
- unlink(2) the file right after creation.
This will remove the pathname from the filesystem without
closing the socket (it will continue to live until the last fd
is closed). This is not useful for us, since we need the
pathname of the socket as its interface.
- chown(2) or chmod(2) the directory that contains the socket.
For removing a file from the filesystem, a process needs
write permissions in the containing directory. We could
put sockets in dummy directories that can be chown(2)ed to
nobody. This could be dangerous, though, as we don't control
the socket names. It is our users who configure the socket
name in their configuration, and so it's easy that they don't
understand the many implications of not chosing an appropriate
socket pathname. A user could unknowingly put the socket in a
directory that is not supposed to be owned by user nobody, and
if we blindly chown(2) or chmod(2) the directory, we could be
creating a big security hole.
- Ask the main process to remove the socket.
This would require a very complex communication mechanism with
the main process, which is not impossible, but let's avoid it
if there are simpler solutions.
- Give the child process the CAP_DAC_OVERRIDE capability.
That is one of the most powerful capabilities. A process with
that capability can be considered root for most practical
aspects. Even if the capability is disabled for most of the
lifetime of the process, there's a slight chance that a
malicious actor could activate it and then easily do serious
damage to the system.
- unlink(2) the file right before calling bind(2).
This is dangerous because another process (for example, another
running instance of unitd(8)), could be using the socket, and
removing the pathname from the filesystem would be problematic.
To do this correctly, a lot of checks should be added before the
actual unlink(2), which is error-prone, and difficult to do
correctly, and atomically.
- Use abstract-namespace Unix domain sockets.
This is the simplest solution, as it only requires accepting a
slightly different syntax (basically a @ prefix) for the socket
name, to transform it into a string starting with a null byte
('\0') that the kernel can understand. The patch is minimal.
Since abstract sockets live in an abstract namespace, they don't
create files in the filesystem, so there's no need to remove
them later. The kernel removes the name when the last fd to it
has been closed.
One caveat is that only Linux currently supports this kind of
Unix sockets. Of course, a solution to that could be to ask
other kernels to implement such a feature.
Another caveat is that filesystem permissions can't be used to
control access to the socket file (since, of course, there's no
file). Anyone knowing the socket name can access to it. The
only method to control access to it is by using
network_namespaces(7). Since in unitd(8) we're using 0666 file
sockets, abstract sockets should be no more insecure than that
(anyone can already read/write to the listener sockets).
- Ask the kernel to implement a simpler way to unlink(2) socket
files when they are not needed anymore. I've suggested that to
the <linux-fsdevel@vger.kernel.org> mailing list, in:
<lore.kernel.org/linux-fsdevel/0bc5f919-bcfd-8fd0-a16b-9f060088158a@gmail.com/T>
In this commit, I decided to go for the easiest/simplest solution,
which is abstract sockets. In fact, we already had partial
support. This commit only fixes some small bug in the existing
code so that abstract Unix sockets work:
- Don't chmod(2) the socket if it's an abstract one.
This fixes the creation of abstract sockets, but doesn't make them
usable, since we produce them with a trailing '\0' in their name.
That will be fixed in the following commit.
This closes#669 issue on GitHub.
nxt_sockaddr_ntop() stopped being used in commit (git) 029942f4eb.
It has been replaced mostly by nxt_sockaddr_text().
commit 029942f4eb
Author: Igor Sysoev <igor@sysoev.ru>
Date: Wed Feb 22 15:09:59 2017 +0300
I/O operations refactoring.
nxt_job_sockaddr_parse() stopped being used in commit (git) 794248090a.
commit 794248090a
Author: Igor Sysoev <igor@sysoev.ru>
Date: Wed Mar 4 14:04:08 2020 +0300
Legacy upstream code removed.
Also, remove functions and types used only by those two functions:
nxt_job_sockaddr_unix_parse()
nxt_job_sockaddr_inet6_parse()
nxt_job_sockaddr_inet_parse()
nxt_job_sockaddr_parse_t
nxt_job_resolve()
nxt_job_resolve_t
Registering an isolated PID in the global PID hash is wrong
because it can be duplicated. Isolated processes are stored only
in the children list until the response for the WHOAMI message is
processed and the global PID is discovered.
To remove isolated siblings, a pointer to the children list is
introduced in the nxt_process_init_t struct.
This closes#633 issue on GitHub.
Some non-Linux systems implement pivot_root(2), even if they
don't document that. An example is MacOS:
$ grepc pivot_root / 2>/dev/null
.../sys/sysproto.h:3012:
int pivot_root(struct proc *, struct pivot_root_args *, int *);
Since the prototype of the syscall differs from that of Linux, we
can't use that syscall. Let's make sure the test only detects
pivot_root(2) under Linux. Also, rename the feature macro to make
clear that it's only about Linux's pivot_root(2).
This closes#737 issue on GitHub.
With NXT_HAVE_PIVOT_ROOT, we had issues in MacOS. Headers should
normally be included unconditionally, except of course if they
don't exist.
This fixes part of the #737 issue on GitHub.
Having the basename of the script pathname was incorrect. While
we don't have something more accurate, the best thing to do is to
have it empty (which should be the right thing most of the time).
This closes#715 issue on GitHub.
The bug was introduced in git commit
0032543fa6
'Ruby: added the Rack environment parameter "SCRIPT_NAME".'.
In src/nxt_http_route_addr.c::nxt_http_route_addr_pattern_parse() there
was potentially undefined behaviour when shifting a 32 bit value by 32
bits, this could happen if cidr_prefix was 0.
Promote the shiftee to unsigned long long to avoid this issue.
Allow $dollar (or ${dollar}) to translate to a literal $ to allow
support for sub-delimiters in URIs.
It is possible to have URLs like
https://example.com/path/15$1588/9925$2976.html
and thus it would be useful to be able to specify them in various bits
of the unit config such as the location setting.
However this hadn't been possible due to $ being used to denote
variables for substitution. E.g $host.
As was noted in the below GitHub issue it was suggested by @VBart to
use $sign to represent a literal $, however I feel $dollar is more
appropriate so we have a variable named after the thing it represents,
also @tippexs found[0] that &dollar is used in HTML to represent a $, so
there is some somewhat related precedent.
(The other idea to use $$ was rejected in my original pull-request[1]
for this issue.)
This means the above URL could be specified as
https://example.com/path/15${dollar}1588/9925${dollar}2976.html
in the unit config.
This is done by adding a variable called 'dollar' which is loaded into
the variables hash table which translates into a literal $.
This is then handled in nxt_var_next_part() where variables are parsed
for lookup and $dollar is set for substitution by a literal '$'. Actual
variable substitution happens in nxt_var_query_finish().
[0]: https://github.com/nginx/unit/pull/693#issuecomment-1130412323
[1]: https://github.com/nginx/unit/pull/693
Closes: https://github.com/nginx/unit/issues/675
The #endif was misplaced by accident during a refactor:
<029942f4eb>.
clang(1)'s -Wunreachable-code-break (implied by -Weverything) catches
that, but it is only produced for code compiled without support
for Unix sockets, which is probably the reason it was undetected:
no-one seems to be compiling Unit without Unix sockets support (at
least with clang(1)).
posix_spawn(3POSIX) was introduced by POSIX.1d
(IEEE Std 1003.1d-1999), and was later consolidated in
POSIX.1-2001, requiring it in all POSIX-compliant systems.
It's safe to assume it's always available, more than 20 years
after its standardization.
Link: <https://pubs.opengroup.org/onlinepubs/9699919799/basedefs/spawn.h.html>
As was pointed out by the cppcheck[0] static code analysis utility there
was a useless assignment in nxt_unit_request_read(). The size parameter
is passed in by value and was being modified without being used again.
[0]: https://cppcheck.sourceforge.io/
As was pointed out by the cppcheck[0] static code analysis utility, lib
was being set in nxt_unit_shm_open() regardless of platform when in fact
it's only used when (NXT_HAVE_MEMFD_CREATE || NXT_HAVE_SHM_OPEN).
Move the variable declaration & definition to be within the
#if (NXT_HAVE_MEMFD_CREATE || NXT_HAVE_SHM_OPEN)
block.
[0]: https://cppcheck.sourceforge.io/
In src/nxt_sockaddr.c::nxt_job_sockaddr_inet_parse() there is a check
that port > 0 then there is a check that port < 1 || port > 65535, well
we _know_ it can't be less than 1.
As was pointed out by the cppcheck[0] static code analysis utility we
can mark a couple of variables as 'const'. This acts as a hint to the
compiler about our intentions and the compiler will tell us when we
deviate from them.
[0]: https://cppcheck.sourceforge.io/
As was pointed out by the cppcheck[0] static code analysis utility we
can mark numerous function parameters as 'const'. This acts as a hint to
the compiler about our intentions and the compiler will tell us when we
deviate from them.
[0]: https://cppcheck.sourceforge.io/
The code for finding the extension made a few assumptions that are
no longer true. It didn't account for pathnames that didn't
contain '/', including the empty string, or the NULL string. That
code was used with "share", which always had a '/', but now it's
also used with "index", which should not have a '/' in it.
This fix works by limiting the search to the beginning of the
string, so that if no '/' is found in it, it doesn't continue
searching before the beginning of the string.
This also happens to work for NULL. It is technically Undefined
Behavior, as we rely on `NULL + 0 == NULL` and `NULL - NULL == 0`.
But that is the only sane behavior for an implementation, and all
existing POSIX implementations will Just Work for this code.
Relying on this UB is useful, because we don't need to add an
explicit check for NULL, and therefore we have faster code.
Although the current code can't have a NULL, I expect that when we
add support for variables in the index, it will be NULL in some
cases.
Link: <https://stackoverflow.com/q/67291052/6872717>
The same code seems to be defined behavior in C++, which normally
will share implementation in the compiler for these cases, and
therefore it is really unlikely to be in trouble.
Link: <https://stackoverflow.com/q/59409034/6872717>
This makes the replacement of forwarded request header
like client_ip and protocol more generic.
It's a prerequirement for protocol replacement.
No functional changes.
This supports a new option "index" that configures a custom index
file name to be served when a directory is requested. This
initial support only allows a single fixed string. An example:
{
"share": "/www/data/static/$uri",
"index": "lookatthis.htm"
}
When <example.com/foo/bar/> is requested,
</www/data/static/foo/bar/lookatthis.html> is served.
Default is "index.html".
===
nxt_conf_validator.c:
Accept "index" as a member of "share", and make sure it's a string.
===
I tried this feature in my own computer, where I tried the
following:
- Setting "index" to "lookatthis.htm", and check that the correct
file is being served (check both a different name and a
different extension).
- Not setting "index", and check that <index.html> is being
served.
- Settind "index" to an array of strings, and check that the
configuration fails:
{
"error": "Invalid configuration.",
"detail": "The \"index\" value must be a string, but not an array."
}
Before this patch, if "index" was a file, but not a regular file
nor a directory, so it may have been for example a FIFO, Unit
returned 404. But if "index" was a directory, Unit returned 301.
For consistency, this patch makes Unit return 404 for every
non-regular file, including directories.
Casts are usually very dangerous, disabling most compiler warnings
and basically removing type safety. This change adds 'const' to a
pointer where we don't need to write, improving type safety, and
that also allows removing some casts.
nxt_str_null() setted the loc.start pointer to NULL, which was
being passed to memcpy(3) through nxt_debug(). That caused
Undefined Behavior, so we now pass an empty string.
An empty string in Location was being handled specially by not sending a
Location header. This may occur after variable resolution, so we need to
consider this scenario.
The obsolete RFC 2616 defined the Location header as consisting of an absolute
URI <https://www.rfc-editor.org/rfc/rfc2616#section-14.30>, which cannot be an
empty string. However, the current RFC 7231 allows the Location to be a
relative URI <https://www.rfc-editor.org/rfc/rfc7231#section-7.1.2>, and a
relative URI may be an empty string <https://stackoverflow.com/a/43338457>.
Due to these considerations, this patch allows sending an empty Location header
without handling this case specially. This behavior will probably be more
straightforward to users, too. It also simplifies the code, which is now more
readable, fast, and conformant to the current RFC. We're skipping an
allocation at request time in a common case such as "action": {"return": 404}
Having a configurable index filename will require adding an index
field to this structure. The most natural name for that field is
'index', so the current index field should be renamed to allow for
that. A sensible name is 'share_idx', since it's the index of the
shares array in 'nxt_http_static_conf_t'.
Instead of 'share_index' I opted for the shorter 'share_idx'.
Also, when 'index' allows an array of filenames in a following
commit, another similar variable 'index_idx' should be created,
and having a different prefix and suffix seems more readable than
for example 'index_index'.
