chore: release v0.0.1

[MagicalTux]

🤖 New release

• nixvm: 0.0.0 -> 0.0.1

Changelog

0.0.1 - 2026-08-05

Fixed

• (wasm) remove */ from Terminal doc — it broke the generated pkg/nixvm.js

Other

• readlink on the mount root is EINVAL, not EISDIR — fixes realpath
• broken-pipe write end reports POLLOUT|POLLERR, not POLLERR alone
• encode a signal death as WIFSIGNALED, not WIFEXITED
• a pipe at EOF reports POLLHUP alone, not POLLHUP|POLLIN
• honor O_NONBLOCK on reads — empty non-blocking pipe returns EAGAIN
• F_GETLK reports the region unlocked; F_SETLK/OFD locks granted
• /proc/self/{comm,cmdline,stat,status,fd} report the live program
• prctl NO_NEW_PRIVS / PDEATHSIG / DUMPABLE round-trip; getpriority ABI
• raise SIGPIPE on writes to a reader-less pipe/socket
• honor EPOLLONESHOT (disarm after one event)
• map rt_sigqueueinfo/rt_tgsigqueueinfo (sigqueue) to delivery
• support AF_UNIX SOCK_DGRAM (syslog /dev/log, unix-dgram IPC)
• renameat2 flags, F_GETPIPE_SZ, O_TMPFILE fallback
• report non-zero uptime and free RAM
• honor O_APPEND — writes go to end-of-file
• implement chmod/fchmod (were no-ops) so mode changes stick
• access(X_OK) checks execute bits; implement SEEK_DATA/SEEK_HOLE
• page-round st_blocks (du/ls -s), matching tmpfs
• make /proc/self/{exe,cwd,root} resolve to live per-task state
• report nixvm's own core count, not the host's (sysfs leak)
• track FD_CLOEXEC (close on exec) and honor F_DUPFD's minimum
• deliver signals through the fd (was a never-readable stub)
• implement utimensat (set file mtime) instead of a no-op
• getitimer/setitimer honor which (VIRTUAL/PROF read back disarmed)
• implement alarm/setitimer/getitimer (ITIMER_REAL → SIGALRM)
• implement rt_sigtimedwait (sigwait/sigtimedwait/sigwaitinfo)
• honor the ppoll/pselect6/epoll_pwait signal mask
• POSIX process-group targeting (0 / -1 / -pgrp) + negative-pid fix
• fill rusage, honor the pid filter, accumulate children CPU
• real nanosleep + per-task CPU accounting (clock_gettime/getrusage/times)
• report real CPU time, consistent with clock_gettime
• full compliant clock set, correct per-clock semantics
• let the software backend reach the vDSO (any libc, not just KVM)
• interrupt every blocking syscall correctly (SA_RESTART vs EINTR)
• wire ISIG signal generation (^C/^/^Z → fg process group)
• pseudo-terminal support (/dev/ptmx + /dev/pts/N)
• virtual tty — forward terminal ioctls to the host tty
• general socket/network ioctls (SIOCGIF*, FIONREAD, SIOCOUTQ)
• implement the fd-flag ioctls (FIONBIO/FIOCLEX/FIONCLEX)
• edge-trigger EPOLLET so a signalled eventfd stops the loop spinning
• clock vDSO — serve clock_gettime/gettimeofday without a syscall
• service clock reads lock-free (no big sh lock)
• implement fcntl(F_SETFL) O_NONBLOCK — claude completes its request
• Revert "poll: honor EPOLLET (edge-triggered) in epoll_wait"
• poll for host I/O instead of declaring deadlock while a connection is live
• honor EPOLLET (edge-triggered) in epoll_wait
• install host egress in the nixvm run path (NIXVM_NET=host)
• fix two tests double-freeing map's data frames
• run signal handlers at CPL3 — fixes SMP fork/signal memory corruption
• don't double-apply the syscall result — fixes the interpreter
• serialize runs of a shared (CLONE_VM) address space — fixes SMP claude
• decouple guest virtual space from the physical RAM pool + lazy pool — claude runs
• async signal delivery to handlers + real rt_sigsuspend + SIGCHLD-on-exit fixes the & wait livelock
• B5 — peel eventfds/timerfds/epolls onto Kernel.pollfds
• B4 — peel pipes onto its own Kernel.pipes lock
• B3 — peel net onto its own Kernel.net lock
• B2 — peel mounts onto its own Kernel.vfs lock
• B1 — extract Shared behind one coarse Mutex; servicing takes &self
• refresh module doc for the ServiceCtx extraction (Phase A)
• extract per-task servicing state into a passed-in ServiceCtx (Phase A)
• per-address-space kernel stack — fixes SMP fork/exec corruption
• TLB shootdowns for host-side page-table edits (fork/CoW/munmap/mprotect)
• demand paging + faithful triple-fault decode; runs the full JSC engine
• retire the flat identity model — shared frame pool + per-process page tables
• real x86-64 AddrSpace over the Phase-1 frame pool
• physical RAM pool + 4 KiB frame allocator (MMU refactor, Phase 1)
• Send bounds on the fstool-backed mount so the feature build compiles
• service syscalls in place under a big kernel lock — kill the per-syscall hand-off
• run KVM_RUN lockless for true intra-process parallelism
• real time-based preemption for both vcpu backends
• resolve /proc/self/fd/ against the live fd table
• MADV_DONTNEED preserves file-backed (ELF segment) pages
• gated load-verify + open tracing debug aids
• preemptive time-slicing (NIXVM_SLICE) so a busy thread can't starve others
• guest-debug watchpoints + execute breakpoints (NIXVM_WATCHPOINT/NIXVM_EBP)
• NIXVM_NOWX debug gate to force uniform RWX
• accurate #PF handling, CPL3 restore, AVX state, grow-down stack
• enforce per-page W^X — guest page tables from the protection map
• enforce NX on instruction fetch (no more executing writable pages)
• synchronous signal delivery (SIGSEGV/SIGILL/SIGBUS to guest handlers)
• real mmap allocator, sched_yield, 8 MiB stack, 64 GiB KVM ceiling
• dependency-free soft-float — true 80-bit x87 + MXCSR directed rounding
• implement the x87 F-row transcendentals (FPREM/FSIN/FYL2X/…)
• share the anonymous-mmap arena per address space, not per task
• implement the SSE ops V8's JIT emits — TurboFan runs
• node runs on the software interpreter (POP r/m + IMUL/shift flags)
• add PALIGNR and PTEST (0F 3A 0F / 0F 38 17)
• syscall writes RCX←RIP and R11←RFLAGS, like hardware
• fix imm16 decode length + add SSE/stack ops V8 exercises
• share fd tables across threads + futex requeue + timer wakeups — node runs
• record the node event-loop limitation (futex/epoll busy-spin)
• bound the fd-cloexec loop — fcntl EBADF on closed fds + capped RLIMIT_NOFILE
• Phase 5 dynamic linking is done — record the two large-program fixes
• give upper and lower layers disjoint inode numbers
• flush writable MAP_SHARED file mappings back to their file
• record the broadened syscall coverage
• broad syscall coverage — ~60 new syscalls across both arches
• record host-egress (Phase 8) landing — apk works over the internet
• host egress — apk update && apk add work over the real internet
• surface a wasm panic instead of hanging at the prompt
• flock + sendmsg/recvmsg — apk update runs to a clean network failure
• don't panic on wasm32 — the browser terminal died on ls
• pick the guest architecture (arm64 / x86-64) before booting the VM
• x86-64 Alpine boots on the interpreter — xchg/AND/OR decode fixes
• boot a full Alpine userland — per-arch stat, legacy syscalls, KVM retry fix
• hardware backend for x86-64 guests — static glibc runs end-to-end
• run stock static glibc binaries — arch_prctl TLS, brk fix, decode gaps
• sync README/ROADMAP/Cargo for the HVF milestone + KVM readiness
• default-on select() with interp fallback + static program e2e
• implement HvfVcpu — EL0 execution, svc trap, syscall resume
• Hypervisor.framework FFI + process VM + bring-up test
• unify guest memory on one contiguous host-aligned region
• page-granular copy-on-write with fault-driven privatization
• export PWD=/ to match the shell's starting cwd
• fix vfork clobbering the parent's address space

---

This PR was generated with release-plz.