The kernel now checks owner/group/other on open, create, unlink, rename, exec
and path search; honours set-user-id and set-group-id; restricts kill, mount,
reboot and mknod; keeps supplementary groups; and implements chown. devtmpfs
nodes carry owners and modes that chmod/chown keep. Terminals run login, the
image has root and schos, X starts as the ordinary user.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EVRhxi9zpdW2uDUN4FVcao
Caps, Num and Scroll Lock now light their LEDs on PS/2 and USB keyboards,
share one state between keyboards, and stop toggling on key repeat. X sets
them through evdev writes while it holds the keyboard.
Unmapping memory in a process with threads on other CPUs now flushes their
TLBs with an IPI. Without it a freed page stayed reachable through stale
translations, which corrupted Epiphany's heap on SMP.
ACPI batteries and AC adapters appear under /sys/class/power_supply.
/proc/<pid>/maps lists file mappings, and a process killed by a fault
names the file and offset it died in. r8169 gets a 256-entry RX ring.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
A thread is a struct proc that shares three counted objects with others of
its group: the address space (struct mm: page tables, heap top, VMAs and a
lock for changing them), the descriptor table, and the signal dispositions.
clone() shares what its flags say and fork() copies.
- clone(): pthread_create's shape (VM, FILES, SIGHAND, THREAD, SETTLS,
PARENT_SETTID, CHILD_CLEARTID) and posix_spawn's (VM, VFORK, a stack).
- futex: WAIT, WAIT_BITSET with timeouts, WAKE, REQUEUE, CMP_REQUEUE, keyed
by the word's kernel address so shared pages meet. set_tid_address keeps
the word that exit zeroes and wakes, for pthread_join.
- exit ends a thread; exit_group, a fatal signal, or the leader exiting ends
them all. reapd frees exited threads; the leader is reaped by its parent
once the rest are gone. getpid is the group, gettid the thread.
- Stops and resumes reach every thread; other process signals the leader.
- exec kills the other threads and unshares the dispositions.
- No TLB shootdown, so pages unmapped from a shared address space are freed
only once every CPU has reloaded CR3 or is idle.
- procfs: /proc lists processes, task/ lists threads, comm, Threads:, Tgid.
- The console honours VMIN 0 / VTIME (btop's input drain hung on it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
- SIGSTOP/SIGTSTP/SIGTTIN/SIGTTOU stop, SIGCONT resumes; the parent gets
SIGCHLD and wait4 reports WUNTRACED/WCONTINUED; ps shows T. Ctrl-Z works,
and mc no longer hangs waiting for its subshell to stop.
- Blocking reads (console, pty, pipe) and sleeps take a pending stop and
carry on after SIGCONT; a caught signal without SA_RESTART ends them
with EINTR.
- nanosleep reads tv_nsec (sleep 0.5 used to return at once), writes the
remainder, and clock_nanosleep understands TIMER_ABSTIME.
- rt_sigaction uses the k_sigaction layout for Linux-ABI processes: the old
action's restorer was written 112 bytes past the caller's struct. Flags
are kept; SIGKILL and SIGSTOP cannot be caught.
- F1..F12 reach programs, as the Linux console sends them (PS/2, USB, I2C).
- TIOCLINUX subcode 6 answers "no modifiers" instead of echoing the 6 back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
- /dev/ptmx and /dev/pts/N: Unix98 pairs with a full line discipline,
window sizes, job control and hangup (kernel/dev/pty.c).
- The console speaks scroll regions, insert/delete line, SU/SD, RI/IND/NEL,
DECSC/DECRC; swallows charset designations and private CSI sequences;
and LF only returns the carriage under ONLCR.
- setsid() drops the controlling terminal, so a pane's pty becomes its ctty.
- pipe2() honours O_NONBLOCK and O_CLOEXEC (libevent's signal pipe).
- File modes are stored and enforced: chmod, fchmod(at), umask, creation
modes; /tmp is 1777.
- /proc/stat has real user/system/idle; /proc/<pid>/task; kthread flag.
- SO_PEERCRED, IOV_MAX 1024.
- /proc/strace: name a program and its syscalls go to the kernel log.
- World gains htop, btop, tmux, mc, ncdu, fastfetch, games, X toys, NetSurf.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
alarm() and setitimer() stored a countdown nothing counted down, so no
timer has ever fired; setitimer and getitimer were also registered with
no Linux number. alarmd now posts SIGALRM on the tick, while any process
has a timer armed. busybox's ping, /bin/ping on the Alpine root, sends
every request after the first from its SIGALRM handler.
It also opens SOCK_RAW, which was EPROTOTYPE. A raw ICMP socket is now a
ping socket whose reads start at the IP header. And a blocked socket
call gives up for a signal with a handler, not only for a kill: the
alarm woke recvfrom, and the wait put it straight back to sleep.
For Xorg: AF_UNIX abstract names, which read as the empty path and so as
EADDRINUSE ("server already running"); and getsockname and getpeername,
which did not exist, so libxcb sent no cookie and the server refused
every client.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
Found by running Alpine's apk on SchOS, one failure at a time:
- chroot(2) was a stub returning EPERM. Each process now has a root:
where an absolute path starts and where .. stops. Mounts remember
whose root they were made under, and a walk leaves a mount by the way
it came in, so one filesystem can be mounted twice.
- devtmpfs and proc mounted only on /dev and /proc, whatever the target;
a chroot got "already mounted".
- mmap never reused freed address space, so musl's malloc used up the
256 mapping records while 780 MB were free. A mergeable mapping now
goes straight after an anonymous one whose end has room.
- unlinkat returned -1 (EPERM) where apk expects ENOENT, and renameat
was an old copy of rename: no replacing the target, and v1 dirents
written into SchFS. Both now share rename's implementation.
- execve runs #! scripts, and says ENOEXEC for what it cannot run.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
Sixty-five headers sat side by side in include/, so `#include "irq.h"`
said nothing about where irq.h was or what it belonged to, and errno.h,
stat.h and fcntl.h at the top shadowed the system's for any host tool
built with -I./include. They are in folders now -- arch/ core/ fs/
dev/ gfx/ net/ uapi/ (and ldls/ for the Linux driver layer) -- and
every include names the folder, as Linux's do: "core/proc.h".
Done mechanically and checked the hard way: the kernel built from this
tree disassembles identically to the one built from the tree before
it. The full image boots, and userland is the same set of programs.
Paths in docs and comments follow; docs/guides/layout.md has a table
of what is where, and tools/fix-includes.py rewrites old-style
includes from the folders in the tree, for code written before this.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4
