Recursive mutexes now go as fast as normal mutexes. The tradeoff is they
are no longer safe to use in signal handlers. However you can still have
signal safe mutexes if you set your mutex to both recursive and pshared.
You can also make functions that use recursive mutexes signal safe using
sigprocmask to ensure recursion doesn't happen due to any signal handler
The impact of this change is that, on Windows, many functions which edit
the file descriptor table rely on recursive mutexes, e.g. open(). If you
develop your app so it uses pread() and pwrite() then your app should go
very fast when performing a heavily multithreaded and contended workload
For example, when scaling to 40+ cores, *NSYNC mutexes can go as much as
1000x faster (in CPU time) than the naive recursive lock implementation.
Now recursive will use *NSYNC under the hood when it's possible to do so
This change solves an issue where many threads attempting to spawn forks
at once would cause fork() performance to degrade with the thread count.
Things got real nasty on NetBSD, which slowed down the whole test fleet,
because there's no vfork() and we're forced to use fork() in our server.
threads count task
1 1062 fork+exit+wait
2 668 fork+exit+wait
4 66 fork+exit+wait
8 19 fork+exit+wait
16 22 fork+exit+wait
32 16 fork+exit+wait
Things are now much less bad on NetBSD, but not great, since it does not
have futexes; we rely on its semaphore file descriptors to do conditions
threads count task
1 1085 fork+exit+wait
2 842 fork+exit+wait
4 532 fork+exit+wait
8 400 fork+exit+wait
16 276 fork+exit+wait
32 66 fork+exit+wait
With OpenBSD which also lacks vfork(), things were just as bad as NetBSD
threads count task
1 584 fork+exit+wait
2 687 fork+exit+wait
4 206 fork+exit+wait
8 24 fork+exit+wait
16 33 fork+exit+wait
32 26 fork+exit+wait
But since OpenBSD has futexes fork() works terrifically thanks to *NSYNC
threads count task
1 525 fork+exit+wait
2 580 fork+exit+wait
4 451 fork+exit+wait
8 479 fork+exit+wait
16 408 fork+exit+wait
32 373 fork+exit+wait
This issue would most likely only manifest itself, when pthread_atfork()
callers manage to slip a spin lock into the outermost position of fork's
list of locks. Since fork() is very slow, a spin lock can be devastating
Needless to say vfork() rules and anyone who says differently is kidding
themselves. Look at what a FreeBSD 14.1 virtual machine with equal specs
can do over the course of three hundred milliseconds.
threads count task
1 2559 vfork+exit+wait
2 5389 vfork+exit+wait
4 34933 vfork+exit+wait
8 43273 vfork+exit+wait
16 49648 vfork+exit+wait
32 40247 vfork+exit+wait
So it's a shame that so few OSes support vfork(). It creates an unsavory
situation, where someone wanting to build a server that spawns processes
would be better served to not use threads and favor a multiprocess model
Cosmopolitan now supports mremap(), which is only supported on Linux and
NetBSD. First, it allows memory mappings to be relocated without copying
them; this can dramatically speed up data structures like std::vector if
the array size grows larger than 256kb. The mremap() system call is also
10x faster than munmap() when shrinking large memory mappings.
There's now two functions, getpagesize() and getgransize() which help to
write portable code that uses mmap(MAP_FIXED). Alternative sysconf() may
be called with our new _SC_GRANSIZE. The madvise() system call now has a
better wrapper with improved documentation.
We now have a C++ red-black tree implementation that implements standard
template library compatible APIs while compiling 10x faster than libcxx.
It's not as beautiful as the red-black tree implementation in Plinko but
this will get the job done and the test proves it upholds all invariants
This change also restores CheckForMemoryLeaks() support and fixes a real
actual bug I discovered with Doug Lea's dlmalloc_inspect_all() function.
Actually Portable Executable now supports Android. Cosmo's old mmap code
required a 47 bit address space. The new implementation is very agnostic
and supports both smaller address spaces (e.g. embedded) and even modern
56-bit PML5T paging for x86 which finally came true on Zen4 Threadripper
Cosmopolitan no longer requires UNIX systems to observe the Windows 64kb
granularity; i.e. sysconf(_SC_PAGE_SIZE) will now report the host native
page size. This fixes a longstanding POSIX conformance issue, concerning
file mappings that overlap the end of file. Other aspects of conformance
have been improved too, such as the subtleties of address assignment and
and the various subtleties surrounding MAP_FIXED and MAP_FIXED_NOREPLACE
On Windows, mappings larger than 100 megabytes won't be broken down into
thousands of independent 64kb mappings. Support for MAP_STACK is removed
by this change; please use NewCosmoStack() instead.
Stack overflow avoidance is now being implemented using the POSIX thread
APIs. Please use GetStackBottom() and GetStackAddr(), instead of the old
error-prone GetStackAddr() and HaveStackMemory() APIs which are removed.
If pthread_create() is linked into the binary, then the cosmo runtime
will create an independent dlmalloc arena for each core. Whenever the
malloc() function is used it will index `g_heaps[sched_getcpu() / 2]`
to find the arena with the greatest hyperthread / numa locality. This
may be configured via an environment variable. For example if you say
`export COSMOPOLITAN_HEAP_COUNT=1` then you can restore the old ways.
Your process may be configured to have anywhere between 1 - 128 heaps
We need this revision because it makes multithreaded C++ applications
faster. For example, an HTTP server I'm working on that makes extreme
use of the STL went from 16k to 2000k requests per second, after this
change was made. To understand why, try out the malloc_test benchmark
which calls malloc() + realloc() in a loop across many threads, which
sees a a 250x improvement in process clock time and 200x on wall time
The tradeoff is this adds ~25ns of latency to individual malloc calls
compared to MODE=tiny, once the cosmo runtime has transitioned into a
fully multi-threaded state. If you don't need malloc() to be scalable
then cosmo provides many options for you. For starters the heap count
variable above can be set to put the process back in single heap mode
plus you can go even faster still, if you include tinymalloc.inc like
many of the programs in tool/build/.. are already doing since that'll
shave tens of kb off your binary footprint too. Theres also MODE=tiny
which is configured to use just 1 plain old dlmalloc arena by default
Another tradeoff is we need more memory now (except in MODE=tiny), to
track the provenance of memory allocation. This is so allocations can
be freely shared across threads, and because OSes can reschedule code
to different CPUs at any time.
This change adds a TLS freelist for small dynamic memory allocations.
Cosmopolitan's TIB is now 512 bytes in size. Single-threaded malloc()
performance isn't impacted by this, until pthread_create() is called.
Single-threaded programs may also want to consider using:
#include "libc/mem/tinymalloc.inc"
Which will shave 30k off the executable size and sometimes go faster.
Now that these functions are behind _COSMO_SOURCE there's no reason for
having the ugly underscore anymore. To use these functions, you need to
pass -mcosmo to cosmocc.
We torture test dlmalloc() in test/libc/stdio/memory_test.c. That test
was crashing on occasion on Apple M1 microprocessors when dlmalloc was
using *NSYNC locks. It was relatively easy to spot the cause, which is
this one particular compare and swap operation, which needed to change
to use sequentially-consistent ordering rather than an acquire barrier
Our makefile generator now accepts badly formatted include lines. It's
now more hermetic with better error checking in the cosmo repo, and it
can be configured to not be hermetic at all.
- On Windows connect() can now be interrupted by a signal; connect() w/
O_NONBLOCK will now raise EINPROGRESS; and connect() with SO_SNDTIMEO
will raise ETIMEDOUT after the interval has elapsed.
- We now get the AcceptEx(), ConnectEx(), and TransmitFile() functions
from the WIN32 API the officially blessed way, using WSAIoctl().
- Do nothing on Windows when fsync() is called on a directory handle.
This was raising EACCES earlier becaues GENERIC_WRITE is required on
the handle. It's possible to FlushFileBuffers() a directory handle if
it's opened with write access but MSDN doesn't document what it does.
If you have any idea, please let us know!
- Prefer manual reset event objects for read() and write() on Windows.
- Do some code cleanup on our dlmalloc customizations.
- Fix errno type error in Windows blocking routines.
- Make the futex polyfill simpler and faster.
- Remove PAGESIZE constant
- Fix realloc() documentation
- Fix ttyname_r() error reporting
- Make forking more reliable on Windows
- Make execvp() a few microseconds faster
- Make system() a few microseconds faster
- Tighten up the socket-related magic numbers
- Loosen restrictions on mmap() offset alignment
- Improve GetProgramExecutableName() with getenv("_")
- Use mkstemp() as basis for mktemp(), tmpfile(), tmpfd()
- Fix flakes in pthread_cancel_test, unix_test, fork_test
- Fix recently introduced futex stack overflow regression
- Let sockets be passed as stdio to subprocesses on Windows
- Improve security of bind() on Windows w/ SO_EXCLUSIVEADDRUSE
This change ports APE Loader to Linux AARCH64, so that Raspberry Pi
users can run programs like redbean, without the executable needing
to modify itself. Progress has also slipped into this change on the
issue of making progress better conforming to user expectations and
industry standards regarding which symbols we're allowed to declare
This change takes an entirely new approach to the incremental linking of
pkzip executables. The assets created by zipobj.com are now treated like
debug data. After a .com.dbg is compiled, fixupobj.com should be run, so
it can apply fixups to the offsets and move the zip directory to the end
of the file. Since debug data doesn't get objcopy'd, a new tool has been
introduced called zipcopy.com which should be run after objcopy whenever
a .com file is created. This is all automated by the `cosmocc` toolchain
which is rapidly becoming the new recommended approach.
This change also introduces the new C23 checked arithmetic macros.
If threads are being used, then fork() will now acquire and release and
runtime locks so that fork() may be safely used from threads. This also
makes vfork() thread safe, because pthread mutexes will do nothing when
the process is a child of vfork(). More torture tests have been written
to confirm this all works like a charm. Additionally:
- Invent hexpcpy() api
- Rename nsync_malloc_() to kmalloc()
- Complete posix named semaphore implementation
- Make pthread_create() asynchronous signal safe
- Add rm, rmdir, and touch to command interpreter builtins
- Invent sigisprecious() and modify sigset functions to use it
- Add unit tests for posix_spawn() attributes and fix its bugs
One unresolved problem is the reclaiming of *NSYNC waiter memory in the
forked child processes, within apps which have threads waiting on locks
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.
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().
- 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 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.
- add vdso dump utility
- tests now log stack usage
- rename g_ftrace to __ftrace
- make internal spinlocks go faster
- add conformant c11 atomics library
- function tracing now logs stack usage
- make function call tracing thread safe
- add -X unsecure (no ssl) mode to redbean
- munmap() has more consistent behavior now
- pacify fsync() calls on python unit tests
- make --strace flag work better in redbean
- start minimizing and documenting compiler flags
- Fix bugs in kDos2Errno definition
- malloc() should now be thread safe
- Fix bug in rollup.com header generator
- Fix open(O_APPEND) on the New Technology
- Fix select() on the New Technology and test it
- Work towards refactoring i/o for thread safety
- Socket reads and writes on NT now poll for signals
- Work towards i/o completion ports on the New Technology
- Make read() and write() intermittently check for signals
- Blinkenlights keyboard i/o so much better on NT w/ poll()
- You can now poll() files and sockets at the same time on NT
- Fix bug in appendr() that manifests with dlmalloc footers off
- Document sigaction()
- Simplify New Technology fork() code
- Testing and many bug fixes for mprotect()
- Distribute Intel Xed ILD in the amalgamation
- Turn Xed enums into defines to avoid DWARF bloat
- Improve polyfilling of SA_SIGINFO on BSDs and fix bugs
- setpgid(getpid(), getpid()) on Windows will ignore CTRL-C
- Work around issues relating to NT mappings being executable
- Permit automatic executable stack override via `ape_stack_pf`
We defined `noinline` as an abbreviation for the longer version
`__attribute__((__noinline__))` which caused name clashes since
third party codebases often write it as `__attribute__((noinline))`.
This commit makes numerous refinements to cosmopolitan memory handling.
The default stack size has been reduced from 2mb to 128kb. A new macro
is now provided so you can easily reconfigure the stack size to be any
value you want. Work around the breaking change by adding to your main:
STATIC_STACK_SIZE(0x00200000); // 2mb stack
If you're not sure how much stack you need, then you can use:
STATIC_YOINK("stack_usage_logging");
After which you can `sort -nr o/$MODE/stack.log`. Based on the unit test
suite, nothing in the Cosmopolitan repository (except for Python) needs
a stack size greater than 30kb. There are also new macros for detecting
the size and address of the stack at runtime, e.g. GetStackAddr(). We
also now support sigaltstack() so if you want to see nice looking crash
reports whenever a stack overflow happens, you can put this in main():
ShowCrashReports();
Under `make MODE=dbg` and `make MODE=asan` the unit testing framework
will now automatically print backtraces of memory allocations when
things like memory leaks happen. Bugs are now fixed in ASAN global
variable overrun detection. The memtrack and asan runtimes also handle
edge cases now. The new tools helped to identify a few memory leaks,
which are fixed by this change.
This change should fix an issue reported in #288 with ARG_MAX limits.
Fixing this doubled the performance of MKDEPS.COM and AR.COM yet again.
- python now mixes audio 10x faster
- python octal notation is restored
- chibicc now builds code 3x faster
- chibicc now has help documentation
- chibicc can now generate basic python bindings
- linenoise now supports some paredit-like features
See #141
- Python static hello world now 1.8mb
- Python static fully loaded now 10mb
- Python HTTPS client now uses MbedTLS
- Python REPL now completes import stmts
- Increase stack size for Python for now
- Begin synthesizing posixpath and ntpath
- Restore Python \N{UNICODE NAME} support
- Restore Python NFKD symbol normalization
- Add optimized code path for Intel SHA-NI
- Get more Python unit tests passing faster
- Get Python help() pagination working on NT
- Python hashlib now supports MbedTLS PBKDF2
- Make memcpy/memmove/memcmp/bcmp/etc. faster
- Add Mersenne Twister and Vigna to LIBC_RAND
- Provide privileged __printf() for error code
- Fix zipos opendir() so that it reports ENOTDIR
- Add basic chmod() implementation for Windows NT
- Add Cosmo's best functions to Python cosmo module
- Pin function trace indent depth to that of caller
- Show memory diagram on invalid access in MODE=dbg
- Differentiate stack overflow on crash in MODE=dbg
- Add stb_truetype and tools for analyzing font files
- Upgrade to UNICODE 13 and reduce its binary footprint
- COMPILE.COM now logs resource usage of build commands
- Start implementing basic poll() support on bare metal
- Set getauxval(AT_EXECFN) to GetModuleFileName() on NT
- Add descriptions to strerror() in non-TINY build modes
- Add COUNTBRANCH() macro to help with micro-optimizations
- Make error / backtrace / asan / memory code more unbreakable
- Add fast perfect C implementation of μ-Law and a-Law audio codecs
- Make strtol() functions consistent with other libc implementations
- Improve Linenoise implementation (see also github.com/jart/bestline)
- COMPILE.COM now suppresses stdout/stderr of successful build commands
The following Lua APIs have been added:
- IsDaemon() → bool
- ProgramPidPath(str)
The following Lua hooks have been added:
- OnClientConnection(ip:int,port:int,serverip:int,serverport:int) → bool
- OnProcessCreate(pid:int,ip:int,port:int,serverip:int,serverport:int)
- OnProcessDestroy(pid:int)
- OnServerStart()
- OnServerStop()
- OnWorkerStart()
- OnWorkerStop()
redbean now does a better job at applying gzip on the fly from the local
filesystem, using a streaming chunked api with constant memory, which is
useful for doing things like serving a 4gb text file off NFS, and having
it start transmitting in milliseconds. redbean will also compute entropy
on the beginnings of files to determine if compression is profitable.
This change pays off technical debts relating to memory, such as relying
on exit() to free() allocations. That's now mostly fixed so it should be
easier now to spot memory leaks in malloc traces.
This change also fixes bugs and makes improvements to our SSL support.
Uniprocess mode failed handshakes are no longer an issue. Token Alpn is
offered so curl -v looks less weird. Hybrid SSL certificate loading is
now smarter about naming conflicts. Self-signed CA root anchors will no
longer be delivered to the client during the handshake.
The most exciting improvement is dynamic pages will soon be able to use
the executable itself as an object store. it required a heroic technique
for overcoming ETXTBSY restrictions which lets us open the executable in
read/write mode, which means (1) wa can restore the APE header, and (2)
we can potentially containerize redbean extension code so that modules
you download for your redbean online will only impact your redbean.
Here's a list of breaking changes to redbean:
- Remove /tool/net/ prefix from magic ZIP paths
- GetHeader() now returns NIL if header is absent
Here's a list of fixes and enhancements to redbean:
- Support 64-bit ZIP archives
- Record User-Agent header in logs
- Add twelve error handlers to accept()
- Display octal st_mode on listing page
- Show ZIP file comments on listing page
- Restore APE MZ header on redbean startup
- Track request count on redbean index page
- Report server uptime on redbean index page
- Don't bind server socket using SO_REUSEPORT
- Fix#151 where Lua LoadAsset() could free twice
- Report rusage accounting when workers exit w/ -vv
- Use ZIP iattr field as text/plain vs. binary hint
- Add ParseUrl() API for parsing things like a.href
- Add ParseParams() API for parsing HTTP POST bodies
- Add IsAcceptablePath() API for checking dots, etc.
- Add IsValidHttpToken() API for validating sane ASCII
- Add IsAcceptableHostPort() for validating HOST[:PORT]
- Send 400 response to HTTP/1.1 requests without a Host
- Send 403 response if ZIP or file isn't other readable
- Add virtual hosting that tries prepending Host to path
- Route requests based on Host in Request-URI if present
- Host routing will attempt to remove or add the www. prefix
- Sign-extend UNIX timestamps and don't adjust FileTime zone
Here's some of the improvements made to Cosmopolitan Libc:
- Fix ape.S indentation
- Improve consts.sh magnums
- Write pretty good URL parser
- Improve rusage accounting apis
- Bring mremap() closer to working
- Added ZIP APIs which will change
- Check for overflow in reallocarray()
- Remove overly fancy linkage in strerror()
- Fix GDB attach on crash w/ OpenBSD msyscall()
- Make sigqueue() portable to most UNIX distros
- Make integer serialization macros more elegant
- Bring back 34x tprecode8to16() performance boost
- Make malloc() more resilient to absurdly large sizes
