- a timer change with EV_ADD unset and data 0 keeps the period; one that set
fflags was silently ignored, where FreeBSD 11's kqueue_register copies it
onto the note: it now throws with the other re-adds, waiting for the
measured rules
- EV_ENABLE on a disabled timer reports the expiries missed while disabled in
one step and keeps the note's phase, as FreeBSD 11's filt_timerexpire counts
a disabled note and keeps its p_nexttime; TimerDisableEnableRearm pins it
(5 serial runs and 3 rounds of 8 parallel copies pass)
- the next-wait abort from the review round was observed before main's
5cbc46029b wait refactor and does not reproduce
Carries the additive pieces of #6510 onto main after its feature was
superseded by #5847: the deleteEventFunc invocation when an EV_ONESHOT
note retires in CollectTriggered, the Windows $Deleted path detection
in FileEvent (IsDeleted transitions), the relaxed timer-modify and
SetVnodeNotes guards, and the compatible kevent tests.
In payload processes on a PS5 (firmware 12.70), re-adding an existing
periodic EVFILT_TIMER with EV_ADD, fflags 0 and a positive period kept
the armed expiry and the expirations already counted, and applied the
new period after that expiry, as FreeBSD 11.0 does. EV_ONESHOT in the
re-add left the timer periodic. Here such a re-add threw; it now does
the same. Re-adding a one-shot or disabled timer, a re-add with nonzero
fflags, a period of 0 or less, a period that overflows the armed expiry
or flags other than EV_ONESHOT, and the other changes to an existing
timer except EV_DELETE still throw.
There, EV_ADD failed with EPERM for EVFILT_AIO, EVFILT_LIO and
EVFILT_SENDFILE at every ident and flag set tried, and EV_DELETE and
EV_ENABLE of them without EV_ADD failed with ENOENT. EVFILT_PROCDESC
failed with EBADF on a descriptor that is not open and with EINVAL on
every kind of open descriptor tried, and EV_DELETE of it on a socket
failed with ENOENT. These filters threw here; they now answer the same
way. EV_ADD of EVFILT_PROCDESC fails with EINVAL on the kinds of
descriptor tried and on shared memory, and throws on other kinds and on
descriptors 0 to 2 that are not sockets or kqueues; a change without
EV_ADD fails with ENOENT, after the descriptor check for
EVFILT_PROCDESC.
A new kevent measurements debt entry, which the kqueue and kevent entry
points to, records these measurements and the answers for NOTE_LOWAT
marks and EVFILT_READ on another kqueue, which still throw here. The
kqueue and kevent entry also stops listing a NOTE_NSECONDS period above
one second with a fraction among the cases that throw, since d71a1c577
made it work.
FreeBSD 11.4 kqueue_scan (sys/kern/kern_event.c) puts a reported level
knote back at the tail of the queue and takes an EV_CLEAR or EV_DISPATCH
knote off it, and knote_enqueue queues that one at the tail when it fires
again. kevent moved a reported knote to the tail only while it stayed
triggered, so a level socket or pipe knote and a re-fired EV_CLEAR or
EV_DISPATCH knote kept their place and could starve later knotes when
nevents is smaller than the number of ready ones. Now every reported
knote that is not deleted moves to the tail.
Merge the file repoll into GatherInterests through WaitSet, so a wait on a
queue that watches another queue still polls regular-file events every
millisecond.
TakeSnapshot read the file information and then the path. A delete
between the two left a snapshot whose path was already under
\$Extend\$Deleted\ but whose link count and delete flag were not, so
IsDeleted was false and the watch thread asked for a parent watch of
\$Extend\$Deleted\. FindFirstChangeNotification refuses it with
ERROR_ACCESS_DENIED, and the exception on the detached watch thread
ended the process. guest_kqueue hit this when FileWatch removed its
file while a watch thread was still polling.
The path is now read first. A path under \$Extend\$Deleted\ is only
returned after the delete, so the information read after it reports
the delete and the parent watch is skipped.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
FreeBSD 11 kqueue_scan (sys/kern/kern_event.c) re-inserts a reported
knote without EV_CLEAR or EV_DISPATCH at the tail of kq_head, and
knote_enqueue puts a knote that fires again at the tail as well, so the
next scan starts with the events that were not reported yet.
sceKernelWaitEqueue scanned the event list in place, so with num smaller
than the number of ready events the first ones were reported on every
call and the later ones starved. Each reported event that stays
registered now moves to the end of the list, in the order reported.
Culprit: none in 892a4f1328..b85d972f42; race in the Windows vnode watch (50b06110b guarded only the loop). Run 39473930499, guest_kqueue: a watched file deleted before registration is snapshotted under $Extend\$Deleted, WatchParent got ERROR_ACCESS_DENIED and threw.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
kevent threw NotImplemented for EVFILT_WRITE with NOTE_LOWAT above 1 on a socket, although the socket readiness check already reports free send-buffer space (SO_SNDBUF less SIOCOUTQ on Linux). The write event now fires once that space reaches the mark, as FreeBSD 11's filt_sowrite compares sbspace with kn_sdata, and stays silent while the space is below it. The debt entry no longer says a write mark is limited to 1. Reference: https://github.com/freebsd/freebsd-src/blob/release/11.0.0/sys/kern/uipc_socket.c
kevent threw for EVFILT_WRITE on any descriptor that is not a socket. On the write end of a pipe it now follows FreeBSD 11's filt_pipewrite: data is the free buffer space, and once the read end is closed the event reports EV_EOF with data 0. Adding the event after the read end is closed fails with EPIPE, as in pipe_kqfilter. EVFILT_WRITE on the read end still throws, because FreeBSD's pipe ends are bidirectional and host pipes are not.
kevent threw for EV_CLEAR on a socket or the read end of a pipe; Boost.Asio's kqueue reactor, among others, registers every socket that way. On Linux such a knote now watches its host descriptor through an edge watch. It is armed when added, on every later EV_ADD or EV_ENABLE while enabled, and by each host wakeup; when armed it is checked with the level-triggered readiness test and reported with the current data and EV_EOF, or disarmed if no longer ready, as kqueue_scan re-runs the filter; reporting it disarms it until the next edge. Waits block on the watch's epoll descriptor, so a socket that stays readable does not wake them. Windows and macOS still throw. GuestSockets::NativeDescriptor exposes the host socket behind a guest one for the watch.
FreeBSD activates an EV_CLEAR knote on each state change the object reports through KNOTE (sorwakeup, sowwakeup, pipeselwakeup), not on its level, so polling the level cannot tell an EV_CLEAR knote when to fire again. EdgeWatch::Watch registers the host descriptor with a private epoll instance in edge-triggered mode: every wakeup of the descriptor with a matching mask (data arriving, buffer space freed, a hang-up) queues one edge, which TakeEdge consumes, and the epoll descriptor itself is readable while an edge is pending, so a wait can block on it. Off Linux opening a watch throws.
WaitForDescriptors always handed the knote's own descriptor to the host wait. A filter that tracks readiness through a descriptor of its own, such as an edge watch, needs the wait to block on that descriptor instead, or a descriptor that stays ready keeps waking the wait. KernelFilter gets a waitDescriptor, and the filter's owner is held for the length of the wait so the descriptor cannot be closed under it.
A kqueue descriptor is readable while one of its events is triggered and
is never writable or exceptional, as FreeBSD's kqueue_poll answers, on
both hosts. On Windows select threw for it, and on Linux its /dev/null
placeholder was reported readable and writable.
A descriptor that stops being a socket, a kqueue or a pipe during the
wait now fails the call with EBADF when it was closed and is selected
again with the rest of the timeout when its number was reused, as
FreeBSD's kern_select looks every descriptor up again in each pass,
instead of throwing. On Windows a host descriptor that is not a pipe
throws when select starts, before waiting.
Depends on #7139.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
PPSA34015 (EA SPORTS FC 27) registers EVFILT_READ for descriptor 0 in a
new kqueue at start-up, ignores the result, and stopped with "kevent:
EVFILT_READ on a non-socket descriptor not implemented". A read event on
the host stream on descriptor 0 is now accepted and never triggers, as
select never reports descriptor 0 readable.
On Linux, FIONREAD on a datagram socket reports the next datagram only,
where FreeBSD's sbavail counts every queued byte, so a read mark above 1
there could never fire. kevent now throws for it as before; stream
sockets and marks of 1 are unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
kevent threw for NOTE_LOWAT. On a socket's EVFILT_READ it now sets the
low-water mark as FreeBSD 11's filt_soread does: the event fires once the
buffered bytes reach it, or at end of file or on a listening socket, and
a later change of the knote replaces it. EVFILT_WRITE takes a mark of 1,
where the host's writability stands in for a free byte; other write marks
and marks below 1 still throw. A pipe ignores the mark, as filt_piperead
does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
kevent threw for any descriptor that is not a guest socket. The read end
of a pipe now gets the descriptor knote, following FreeBSD 11's
filt_piperead: data is the number of buffered bytes and EV_EOF is set
once the write end is closed. GuestSockets::Ready falls back to the new
PipeReadState for a descriptor that is not a socket, and WaitAny waits on
the pipe with ppoll on Linux or in 1 ms WSAPoll slices on Windows, only
when the queue holds a pipe. On Windows the equeue waker now starts
Winsock itself, since a queue can wait on a pipe before any socket exists.
On a new knote EV_FORCEONESHOT is dropped from the flags; on an
existing one it sets EV_ONESHOT and fires the knote, as
kqueue_register does at l.1339-1341. A modify no longer ORs
EV_ONESHOT, EV_CLEAR or EV_DISPATCH into the knote.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
A knote registered or modified with EV_DISABLE stays registered but is
not polled, fired or reported until an EV_ENABLE change re-arms it, as
kern_event.c does with KN_DISABLED. EV_DISPATCH reports an event once
and then disables the knote, and EV_FORCEONESHOT registers as oneshot.
A negative nevents fails with EINVAL instead of throwing.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The Windows error for a refused FindFirstChangeNotification was one string shared
by WatchParent and WatchContents, so the log could not tell which handle or path
was denied. Build it from the watcher, the path the API was given and
GetLastError instead.
Still throws: the EVFILT_VNODE attach path is vfs_kqfilter, which returns only
EINVAL, ENOMEM or 0, with EBADF from fget before it, and consults no directory
permission, so no FreeBSD errno stands for a host that refuses to notify.
Windows 10 1903 deletes a file with open handles by moving it into
$Extend\$Deleted until the last handle closes, so the next snapshot of a
watched file that a title (or the test) deleted saw a new path, tried to
watch that directory's changes, got ERROR_ACCESS_DENIED and threw from
the watch thread, which aborts the process; the path change was also
reported as NOTE_RENAME next to NOTE_DELETE. A snapshot with a pending
delete or no links now reports NOTE_DELETE alone and drops the parent
directory watch instead of replacing it, as the file has no directory
to be renamed or removed from any more.
libkernel numbered guest sockets from GuestSockets::FirstDescriptor
(0x10000000), so a title that puts a socket in a 1024-bit fd_set writes
fd / 8 bytes past it. The Epic Online Services SDK does this on its
RtcNetworkThread and three titles stop there (#1794):
Palworld PPSA23126 (queeeef, main a11c8f7f7):
FATAL: unhandled exception 0xc0000005 on thread 37164 'RtcNetworkThread'
rip 0x00007ffcdb1ea2be eossdk-ps5-shipping.prx.guest.prx+0xdea2be
rcx 0000000010000005 rsi 0000000000400000 r12 0000000010000005
The Mound PPSA26223 (xXPBSXx, main 0a8b1d32b):
FATAL: unhandled exception 0xc0000005 on thread 'RtcNetworkThread',
rip eossdk-ps5-shipping.prx.guest.prx+0xefc6f4, rcx = 0x10000005,
rsi = 0x400000
Dying Light 2 PPSA02262 (tv7):
FATAL: unhandled exception 0xc0000005 on thread 'RtcNetworkThread'
rcx 0000000010000005 ... rsi 0000000000400000 ... r12 0000000010000005
On FreeBSD 11.4 (8c639945) FD_SETSIZE is 1024 (sys/sys/select.h:58-59)
and sockets take the lowest free entry of the one process descriptor
table that files use (kern_descrip.c:1643-1660 fdalloc,
uipc_syscalls.c kern_socket through falloc).
Every socket number is now a real host descriptor: a zero-access NUL
handle registered with _open_osfhandle on Windows (as directory
descriptors already are) and an F_DUPFD_CLOEXEC duplicate of the host
socket on Linux. The socket table maps that number to the socket, so
"is this a socket" is a table lookup instead of a range check at the 31
former FirstDescriptor comparisons. close releases both the socket entry
and the number, dup and dup2 duplicate the number through the host table
and keep the socket shared, dup2 of a file onto a socket number drops the
socket entry, and sceKernelClose now closes sockets too. sceNetSocket and
sceNetAccept take their ids from the same table instead of a counter from
32, so libkernel and libSceNet numbers never collide.
select polls guest sockets alongside libSceNet sockets, the random device,
stdio and (on Linux) host descriptors; poll throws only for an open
descriptor that is not a guest socket and reports POLLNVAL otherwise.
File calls that used to fail with EBADF on the socket range return
FreeBSD's answer for a socket: lseek ESPIPE (kern_lseek), ftruncate and
getdirentries EINVAL (invfo_truncate, getvnode), flock EOPNOTSUPP;
fstat on a socket throws. pread on a closed descriptor returns EBADF.
title failing on host descriptor numbers; these three titles are that
case, and this change keeps host numbering for files and only moves
sockets into it.
Tests: guest_socket_descriptors (new: socket below 1024, select on a
socket and a file together, close frees the number for a file, dup of a
socket, sceKernelClose), guest_select (socket and pipe together, Linux),
guest_poll, guest_dup, guest_socket_pair, guest_fstatfs,
guest_posix_filesystem and guest_sce_net updated.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
(cherry picked from commit 20378375b31eb5db922db6e55f45e44a4dae6445)
inotify queues IN_MOVED_FROM and IN_MOVED_TO of a rename one after the
other. The watcher thread could read the first before the second was
queued and report a rename inside the directory as a move to another
directory (NOTE_WRITE | NOTE_EXTEND instead of NOTE_WRITE), which made
guest_kqueue abort on loaded runners.
The watcher now reads the queue until it is empty and, when a move half
has no partner, calls getdents64 on the directory before reading again:
that takes the directory lock rename holds until both halves are queued.
Return the event's data when EV_ERROR is set in its flags and 0 otherwise, as KytyPS5 does and as FreeBSD 11 kevent stores the errno of a failed change. A null event throws like the other accessors.
FreeBSD's vfs_kqfilter checks neither the vnode type nor fflags, so the
registration is valid; it is only delivered for a move between directories,
which the Windows directory handle cannot tell from a removal plus an addition.
Also moves the TechnicalDebt entries to docs/dev/debt after main split the file.
Co-Authored-By: Claude <[email protected]>
Dean