This change addresses various open source compatibility issues, so that
we pass 313/411 of the tests in https://github.com/jart/libc-test where
earlier today we were passing about 30/411 of them, due to header toil.
Please note that Glibc only passes 341/411 so 313 today is pretty good!
- Make the conformance of libc/isystem/ headers nearly perfect
- Import more of the remaining math library routines from Musl
- Fix inconsistencies with type signatures of calls like umask
- Write tests for getpriority/setpriority which work great now
- conform to `struct sockaddr *` on remaining socket functions
- Import a bunch of uninteresting stdlib functions e.g. rand48
- Introduce readdir_r, scandir, pthread_kill, sigsetjmp, etc..
Follow the instructions in our `tool/scripts/cosmocc` toolchain to run
these tests yourself. You use `make CC=cosmocc` on the test repository
- Change IDT code so kprintf() isn't mandatory dependency
- Document current intentions around pthread_cancel()
- Make _npassert() an _unassert() in MODE=tiny
- fix rare thread exit race condition on openbsd
- pthread_getattr_np() now supplies detached status
- child threads may now pthread_join() the main thread
- introduce sigandset(), sigorset(), and sigisemptyset()
- introduce pthread_cleanup_push() and pthread_cleanup_pop()
- correctly check that virtual region we want to use is
unmapped, rather than accidentally clobbering existing pages
- avoid placing mmap'd memory at null virtual address
* [metal] Refactoring: separate out sys_writev_vga() and _vga_init() modules
* [metal] Read VGA info from BDA before long mode entry, not after
If using a pre-existing VGA text console, the VGA initialization
code now retrieves the cursor position & character height from
the BIOS data area while still in real mode — rather than
reading from the BIOS data area only after entering long mode.
(This should help make the code more correct, if Cosmopolitan
were to support UEFI graphics output in the future. If the
program were booted via UEFI, then the long mode IsMetal()
code would still be activated, but the BIOS data area might
not have been initialized in that case.)
This change also means that there are now a few more fields
in the `struct mman`.
* [metal] VGA console can now show "screen of death" upon a crash
There is now a new function _klog_vga(), which can be called
by kprintf() to output system messages — e.g. information about
CPU exceptions — on the VGA screen.
* [metal] CPU exception handler now dumps cr2 value
* [metal] Add demo of program crash reporting w/ bare metal VGA TTY
* [metal] Reduce size of "screen of death" code
You can now do things like implement mutexes using futexes in your
redbean lua code. This provides the fastest possible inter-process
communication for your production systems when SQLite alone as ipc
or things like pipes aren't sufficient.
We need to make sure no existing mappings exist between the
MAP_GROWSDOWN page and the guard page, since otherwise it's
not going to be able to grow down thus causing difficult to
troubleshoot failures.
This change also found a few POSIX compliance bugs with errnos. Another
bug was discovered where, on Windows, pread() and pwrite() could modify
the file position in cases where ReadFile() returned an error e.g. when
seeking past the end of file. We also have more tests!
This change lets you use system() in an easier and portable way. The
problem with the call in the past has always been that bourne and
cmd.com on Windows have less than nothing in common, so pretty much the
only command system() could be used for across platforms was maybe echo.
cmd.exe is also a security liability due to its escaping rules.
Since cocmd.com implements 85% of what we need from bourne, in a really
tiny way, it makes perfect sense to be embedded in these functionss. We
get a huge performance boost too.
Fixes#644
This makes it possible for us to use system() and popen() with paths
that redirect to filenames that contain spaces, e.g.
system("echo.com hello >\"hello there.txt\"")
It's difficult to solve this problem, because WIN32 only allows passing
one single argument when launching programs and each program is allowed
to tokenize that however it wants. Most software follows the convention
of cmd.exe which is poorly documented and positively byzantine.
In the future we're going to solve this by not using cmd.exe at all and
instead embedding the cocmd.com interpreter into the system() function.
In the meantime, our documentation has been updated to help recalibrate
any expectation the user might hold regarding the security of using the
Windows command interpreter.
Fixes#644
This change reduces the .bss memory requirement for all executables by
O(64kb). The brk system calls are now fully tested and figured out and
might be useful for tiny programs that only target System Five.
This change improves copy_file_range(), sendfile(), splice(), openpty(),
closefrom(), close_range(), fadvise() and posix_fadvise() in addition to
writing tests that confirm things like errno and seeking behavior across
platforms. We now less aggressively polyfill behavior with some of these
functions when the platform support isn't available. Please see:
https://justine.lol/cosmopolitan/functions.html
Since we're now on Windows 8, we can have clone() work as advertised on
Windows, where it sends a futex wake to the child tid. It's also likely
we no longer need to work around thread flakes on OpenBSD, in _wait0().
Doing this makes binaries tinier, since we don't need to have all the
extra code for supporting a 32-bit address space. It also benefits us
because we're able to use WIN32 futexes, which makes locking simpler.
b69f3d2488 is what officially ended our
Windows 7 support. This change is merely a formalization. You can use
old versions of Cosmo now and forevermore if you need Windows 7 since
our repository is hermetic and vendors all its dependencies.
Won't fix#617
- Fix preadv() and pwritev() for old distros
- Introduce _npassert() and _unassert() macros
- Prove that file locks work properly on Windows
- Support fcntl(F_DUPFD_CLOEXEC) on more systems
640 bytes for old kDos2Errno table
182 bytes for new kDos2Errno under hello2.com (MODE=fastbuild)
122 bytes for new kDos2Errno under hello2.com (MODE=tiny)
You can now run bare metal on bare metal!
* Fix handling of int 0x15 eax = 0xe820 memory map
* Fix some issues in initial page table creation
* hello4.com now works outside emulators
* Ensure area for identity page tables are zeroed first
* Simplify logic for creating page table entries, this partly
reverts 577c0f6226
* Add degenerate MBR partition entry, to ease testing
Co-authored-by: tkchia <tkchia-cosmo@gmx.com>
This makes breaking changes to add underscores to many non-standard
function names provided by the c library. MODE=tiny is now tinier and
we now use smaller locks that are better for tiny apps in this mode.
Some headers have been renamed to be in the same folder as the build
package, so it'll be easier to know which build dependency is needed.
Certain old misguided interfaces have been removed. Intel intrinsics
headers are now listed in libc/isystem (but not in the amalgamation)
to help further improve open source compatibility. Header complexity
has also been reduced. Lastly, more shell scripts are now available.
The organization of the source files is now much more rational.
Old experiments that didn't work out are now deleted. Naming of
things like files is now more intuitive.
This change fixes#496 where ASAN spotted a race condition that could
happen in multithreaded programs, with more than OPEN_MAX descriptors
when using ZipOS or Windows NT, which require tracking open file info
and this change fixes that table so it never relocates, thus allowing
us to continue to enjoy the benefits of avoiding locks while reading.
- You can now use _gc(malloc()) in multithreaded programs
- This change fixes a bug where fork() on NT disabled TLS
- Fixed TLS code morphing on XNU/NT, for R8-R15 registers
If your main module has this declaration:
STATIC_YOINK("vga_console");
Then a VGA driver will be linked into your executable which
displays your stdio characters on the PC display, whereas
before we could only use the serial port. Your display is an
ANSI terminal and it's still a work in progress.
It now works most excellently across all supported operating
sytsems (earlier it didn't work on NT and XNU). Demo code is
available in examples/clock.c and this change also adds some
of the newer ANSI C time functions like timespec_get(), plus
timespec_getres() which hasn't even come out yet as it's C23
pthread_mutex_lock() now uses a better algorithm which goes much faster
in multithreaded environments that have lock contention. This comes at
the cost of adding some fixed-cost overhead to mutex invocations. That
doesn't matter for Cosmopolitan because our core libraries all encode
locking operations as NOP instructions when in single-threaded mode.
Overhead only applies starting the moment you first call clone().
This change restores the .symtab symbol table files in our flagship
programs (e.g. redbean.com, python.com) needed to show backtraces. This
also rolls back earlier changes to zip.com w.r.t. temp directories since
the right way to do it turned out to be the -b DIR flag.
This change also improves the performance of zip.com. It turned out
mmap() wasn't being used, because zip.com was assuming a 4096-byte
granularity, but cosmo requires 65536. There was also a chance to speed
up stdio scanning using the unlocked functions.
- Polyfill pselect() on Windows
- Add -O NOFILE flag to pledge.com
- Polyfill ppoll() on NetBSD, XNU, and Windows
- Support negative numbers and errno in sizetol()
- Add .RSS, .NOFILE, and .MAXCORE to Landlock Make
- Fix issue with .PLEDGE preventing touching of output files
- Add __watch() function (like ftrace) for logging memory changes
It turned out that specifying all SRCS and INCS as dependencies on the
pattern rules for all headers, caused `make` memory usage to skyrocket
from 40mb ot 160mb. This change also reduces the build graph another 4%.
This change introduces the nointernet() function which may be called to
prevent a process and its descendants from communicating with publicly
routable Internet addresses. GNU Make has been modified to always call
this function. In the future Landlock Make will have a way to whitelist
subnets to override this behavior, or disable it entirely. Support is
available for Linux only. Our firewall does not require root access.
Calling nointernet() will return control to the caller inside a new
process that has a SECCOMP BPF filter installed, which traps network
related system calls. Your original process then becomes a permanent
ptrace() supervisor that monitors all processes and threads descending
from the returned child. Whenever a networking system call happens the
kernel will stop the process and wakes up the monitor, which then peeks
into the child memory to read the sockaddr_in to determine if it's ok.
The downside to doing this is that there can be only one supervisor at a
time using ptrace() on a process. So this firewall won't be enabled if
you run make under strace or inside gdb. It also makes testing tricky.
The earlier iterations did too much guesswork when it came to things
like stderr logging and syscall origin verification. This change will
make things more conformant to existing practices. The __pledge_mode
extension now can be configured in a better way.
There's also a new `-q` flag added to pledge.com, e.g.
o//tool/build/pledge.com -qv. ls
Is a good way to disable warnings about `tty` access attempts.
- 10.5% reduction of o//depend dependency graph
- 8.8% reduction in latency of make command
- Fix issue with temporary file cleanup
There's a new -w option in compile.com that turns off the recent
Landlock output path workaround for "good commands" which do not
unlink() the output file like GNU tooling does.
Our new GNU Make unveil sandboxing appears to have zero overhead
in the grand scheme of things. Full builds are pretty fast since
the only thing that's actually slowed us down is probably libcxx
make -j16 MODE=rel
RL: took 85,732,063µs wall time
RL: ballooned to 323,612kb in size
RL: needed 828,560,521µs cpu (11% kernel)
RL: caused 39,080,670 page faults (99% memcpy)
RL: 350,073 context switches (72% consensual)
RL: performed 0 reads and 11,494,960 write i/o operations
pledge() and unveil() no longer consider ENOSYS to be an error.
These functions have also been added to Python's cosmo module.
This change also removes some WIN32 APIs and System Five magnums
which we're not using and it's doubtful anyone else would be too
- We now kill the program on violations like OpenBSD
- We now print a message explaining which promise is needed
- This change also fixes a linkage bug with thread local storage
- Your sigaction() handlers should now be more thread safe
A new `__pledge_mode` global has been introduced to make pledge() more
customizable on Linux. For example:
__attribute__((__constructor__)) static void init(void) {
__pledge_mode = SECCOMP_RET_ERRNO | EPERM;
}
Can be used to restore our old permissive pledge() behavior.
The whole repository is now buildable with GNU Make Landlock sandboxing.
This proves that no Makefile targets exist which touch files other than
their declared prerequisites. In order to do this, we had to:
1. Stop code morphing GCC output in package.com and instead run a
newly introduced FIXUPOBJ.COM command after GCC invocations.
2. Disable all the crumby Python unit tests that do things like create
files in the current directory, or rename() files between folders.
This ended up being a lot of tests, but most of them are still ok.
3. Introduce an .UNSANDBOXED variable to GNU Make to disable Landlock.
We currently only do this for things like `make tags`.
4. This change deletes some GNU Make code that was preventing the
execve() optimization from working. This means it should no longer
be necessary in most cases for command invocations to be indirected
through the cocmd interpreter.
5. Missing dependencies had to be declared in certain places, in cases
where they couldn't be automatically determined by MKDEPS.COM
6. The libcxx header situation has finally been tamed. One of the
things that makes this difficult is MKDEPS.COM only wants to
consider the first 64kb of a file, in order to go fast. But libcxx
likes to have #include lines buried after huge documentation.
7. An .UNVEIL variable has been introduced to GNU Make just in case
we ever wish to explicitly specify additional things that need to
be whitelisted which aren't strictly prerequisites. This works in
a manner similar to the recently introduced .EXTRA_PREREQS feature.
There's now a new build/bootstrap/make.com prebuilt binary available. It
should no longer be possible to write invalid Makefile code.
This change also fixes a bug with gettid() being incorrect after fork().
We now implement the ENOENT behavior for getauxval(). The getuid() etc.
system calls are now faster too. Plus issetugid() will work on BSDs.
- Fix getpriority()
- Add AT_MINSIGSTKSZ
- Fix bugs in BPF code
- Show more stuff in printargs.com
- Write manual test for pledge.com
- pledge() now generates tinier BPF code
- Have pledge("exec") only enable execve()
- Fix pledge.com chroot setuid functionality
- Improve pledge.com unveiling of ape loader
This change fixes bugs, adds more system calls, and improves
compatibility with OpenBSD. Going forward, versions on the web will be
pinned to a permanent version. There were many other changes over the
last week which also improved this new release.
This change reconciles our pledge() implementation with the OpenBSD
kernel source code. We now a polyfill that's much closer to OpenBSD's
behavior. For example, it was discovered that "stdio" permits threads.
There were a bunch of Linux system calls that needed to be added, like
sched_yield(). The exec / execnative category division is now dropped.
We're instead using OpenBSD's "prot_exec" promise for launching APE
binaries and dynamic shared objects. We also now filter clone() flags.
The pledge.com command has been greatly improved. It now does unveiling
by default when Landlock is available. It's now smart enough to unveil a
superset of paths that OpenBSD automatically unveils with pledge(), such
as /etc/localtime. pledge.com also now checks if the executable being
launched is a dynamic shared object, in which case it unveils libraries.
These changes now make it possible to pledge curl on ubuntu 20.04 glibc:
pledge.com -p 'stdio rpath prot_exec inet dns tty sendfd recvfd' \
curl -s https://justine.lol/hello.txt
Here's what pledging curl on Alpine 3.16 with Musl Libc looks like:
pledge.com -p 'stdio rpath prot_exec dns inet' \
curl -s https://justine.lol/hello.txt
Here's what pledging curl.com w/ ape loader looks like:
pledge.com -p 'stdio rpath prot_exec dns inet' \
o//examples/curl.com https://justine.lol/hello.txt
The most secure sandbox, is curl.com converted to static ELF:
o//tool/build/assimilate.com o//examples/curl.com
pledge.com -p 'stdio rpath dns inet' \
o//examples/curl.com https://justine.lol/hello.txt
A weird corner case needed to be handled when resolving symbolic links
during the unveiling process, that's arguably a Landlock bug. It's not
surprising since Musl and Glibc are also inconsistent here too.
We had previously not enabled TLS in MODE=tiny in order to keep the
smallest example programs (e.g. life.com) just 16kb in size. But it
was error prone doing that, so now we just always enable it because
this change uses hacks to ensure it won't increase life.com's size.
This change also fixes a bug on NetBSD, where signal handlers would
break thread local storage if SA_SIGINFO was being used. This looks
like it might be a bug in NetBSD, but it's got a simple workaround.
The pledge.com command now supports the new [WIP] unveil() support. For
example, to strongly sandbox our command for listing directories.
o//tool/build/assimilate.com o//examples/ls.com
pledge.com -v /etc -p 'stdio rpath' o//examples/ls.com /etc
This file system sandboxing is going to be perfect for us, because APE
binaries are self-contained static executables that really don't use the
filesystem that much. On the other hand, with non-static executables,
sandboxing is going to be more difficult. For example, here's how to
sandbox the `ls` command on the latest Alpine:
pledge.com -v rx:/lib -v /usr/lib -v /etc -p 'stdio rpath exec' ls /etc
This change fixes the `execpromises` API with pledge().
This change also adds unix.unveil() to redbean.
Fixes#494
This change simplifies the thread-local storage support code. On Windows
and Mac OS X the startup latency of __enable_tls() has been reduced from
30ms to 1ms. On Windows, TLS memory accesses will now go much faster due
to better self-modifying code that prevents a function call and acquires
our thread information block pointer in a single instruction.
We now rewrite the binary image at runtime on Windows and XNU to change
mov %fs:0,%reg instructions to use %gs instead. There's also simpler
threading API introduced by this change and it's called _spawn() and
_join(), which has replaced most clone() usage.
- Fix DescribeSigset()
- Introduce new unix.rmrf() API
- Fix redbean sigaction() doc example code
- Fix unix.sigaction() w/ more than two args
- Improve redbean re module API (non-breaking)
- Enhance Lua with Python string multiplication
- Make third parameter of unix.socket() default to 0
