feat(libkernel): guest memory protection and VirtualQuery on macOS hosts

- mprotect: the guest rounds a range to its 16 KiB pages. A Mach-O image
  is loaded at a multiple of the host page size only (4 KiB under
  Rosetta), so the pages are counted from where the ELF address 0 of the
  image holding the range was loaded, and a rounded edge smaller than a
  guest page that lies in the gap between two segments, where no host page
  exists, is left out. The write watch is told the new protection.
- sceKernelVirtualQuery reads the region with mach_vm_region.
- sceKernelMunlockall unlocks the ranges libkernel locked, one by one:
  macOS has no munlockall, which fails there with ENOSYS and would
  overwrite errno.
- The stack protect test reads page protections with mach_vm_region.

This is mugurc's f66a4df741 from his macOS series, without the mlock
check that #10889 adds.

Co-authored-by: mugurc <[email protected]>

docs/dev/debt/libkernel.md

@@ -259,6 +259,7 @@
 - [connect to AF_UNSPEC](../../../core/libs/prx/libkernel/Socket/src/Udp.cpp) (libkernel) dissolves a datagram association as FreeBSD 11.4's `udp_disconnect()` does, which resets the local address to the wildcard address and keeps the port. On Linux and on Windows a socket bound to an explicit port keeps its local address as well (`127.0.0.1:<port>` stays `127.0.0.1:<port>`), because a bound socket cannot be bound again, and `getsockname` reports that host address. On Windows a `sendto` to a non-loopback address from such a socket then fails with `ENETUNREACH` (Winsock 10051, measured on Windows 10), where FreeBSD sends it from the wildcard address. One bound to port 0 is bound again to the wildcard address with the port it had
 - [connect](../../../core/libs/prx/libkernel/Socket/src/Udp.cpp) (libkernel) - `connect` to port 0 on an `AF_INET` or `AF_INET6` stream or datagram socket fails with `EADDRNOTAVAIL` (49), and on a connected datagram socket it disconnects first, as on a PS5 (firmware 13.60) and in FreeBSD 11.4's `soconnect()`. A connected stream socket keeps its host `EISCONN` answer. Known gap: `sendto` and `sendmsg` to port 0 on an `AF_INET` datagram socket fail with 49 (measured for an unbound socket) unless the socket is bound to a specific address and port and the destination is neither `INADDR_ANY` nor `INADDR_BROADCAST`, which is the path where `udp_output()` skips `in_pcbconnect_setup()` and FreeBSD sends to port 0. There the call returns the length without the host, as Linux's `udp_sendmsg` refuses port 0 with `EINVAL`; an `AF_INET6` socket still goes to the host. [sceNetConnect](../../../core/libs/prx/libSceNet/Export.cpp) is not changed
 - [msync](../../../core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp) (libkernel) follows FreeBSD 11's `kern_msync` and `vm_map_sync`: the address is rounded down to the page, a range that wraps is `EINVAL`, as is `MS_ASYNC` with `MS_INVALIDATE`, and a range whose start or any page is not mapped (as `sceKernelVirtualQuery` reports it, reservations included) is `ENOMEM`; a page unmapped from inside a larger mapping still counts as mapped there, since `sceKernelVirtualQuery` keeps reporting the whole mapping. The guest has no file mappings, so a successful call writes nothing back. `MS_INVALIDATE` on a mapped range fails with `EBUSY`, as it did in a payload process on a PS5 (firmware 13.60) for anonymous memory, locked or not, also with a hole unmapped inside it; a title process, reservations, direct memory and flexible memory were not measured
+- Guest memory protection on macOS ([mprotect](../../../core/libs/prx/libkernel/DirectMemory/DirectMemory.cpp), [sceKernelVirtualQuery](../../../core/libs/prx/libkernel/DirectMemory/Export.cpp)): the guest rounds a range to its 16 KiB pages, counted from where the ELF address 0 of the image holding it was loaded, which is a multiple of the host page size only (4 KiB under Rosetta). A rounded edge smaller than a guest page that lies in the gap between two segments of an image, where no host page exists, is left out of the request. `sceKernelVirtualQuery` reports the region `mach_vm_region` returns, so its extent is the host's. After a protection change the write watch is told the protection, which the host reports for a watched page as its own
 - [madvise](../../../core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp) (libkernel) accepts `MADV_PROTECT` (10) as a no-op hint without modifying process protection. FreeBSD 11's `kern_madvise` (`sys/vm/vm_mmap.c`) handles it before the address check, ignores the address and length, and calls `kern_procctl(..., PROC_SPROTECT, PPROT_SET)`, which requires `PRIV_VM_MADV_PROTECT` and fails with `EPERM` without it. Returning 0 unconditionally assumes the calling process is privileged. Console behaviour is unverified
 - [mincore](../../../core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp) (libkernel) follows FreeBSD 11's `sys_mincore` (`sys/vm/vm_mmap.c`), not measured on the console: the address is rounded down to the page, length 0 returns 0, a null vector over a mapped range returns `EFAULT`, an overflowing range or any unmapped page in the range (as reported by `sceKernelVirtualQuery`) returns `ENOMEM`. For valid ranges, a page reports `MINCORE_INCORE` (0x1) when `sceKernelVirtualQuery` reports it committed and 0 when it is only reserved; actual residency, referenced and modified bits are not modelled. Length 0 returns 0 when the start page is mapped and `ENOMEM` otherwise, and the range and mapping checks come before the null vector `EFAULT`, as in `kern_mincore` Java 25 LAN / DNS (PPSA99869) imports it
 - [mlockall, munlockall](../../../core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp) (libkernel) - `mlockall` returns `EINVAL` for zero or unknown flags. Valid `MCL_CURRENT` throws because it would lock host stacks, heaps, modules and driver mappings along with guest memory; `MCL_FUTURE` throws because later mappings are not tracked for locking. FreeBSD 11.4's `sys_munlockall` (`sys/vm/vm_mmap.c`:1133-1165) checks `priv_check(td, PRIV_VM_MUNLOCK)` first (returning `EPERM` without it) and unwires the user map (`vm_map_unwire` over `vm_map_min`...`vm_map_max` with `VM_MAP_WIRE_HOLESOK`). Returning 0 unconditionally assumes the calling process is privileged. `munlockall` unlocks memory locked by `mlock` (`VirtualUnlock` on Windows, tolerating `ERROR_NOT_LOCKED` for pages never locked, and `munlockall` on Linux). XPSemu (PPSA97358) imports `mlockall` and `munlockall`; their calls in QEMU `os-posix.c` execute only when `-overcommit mem-lock=on` is enabled. The call has not been observed in a run with this implementation.