[Regression] MacOS: `/dev/urandom` misdetected as optical device post BD changes

[milodowling]

Details

Quick summary

Since #19233, /dev/urandom is misdetected as an optical raw device on macOS, so sys_ss_random_number_generator() fails and kills the main thread on any game that calls it.

Details

#19233 gave fs::is_optical_raw_device() a POSIX body; previously it returned false unconditionally off-Windows. On macOS is_optical_device_node() is:

#elif defined(__APPLE__)
    return S_ISCHR(file_info.st_mode);

/dev/urandom is a character device, so it is treated as an optical drive. fs::file::open() then calls get_raw_device_size(), which on macOS compiles neither the Linux (BLKGETSIZE64) nor the BSD (DIOCGMEDIASIZE) branch and falls back to lseek(fd, 0, SEEK_END) — 0 for a character device. Size 0 is read as "no disc mounted", so the fd is closed and m_file is never assigned:

if (!raw_device_size) { ::close(fd); g_tls_error = fs::error::noent; return; }

sys_ss.cpp:148 then sees !rnd and throws.

Verified by running that logic standalone on macOS 15.7.3 / M4 Max:

is_optical_raw_device("/dev/urandom") = TRUE
open() -> fd=3 (succeeds)
get_raw_device_size(fd) = 0   -> open aborted

Log — the guest request is an ordinary 64-byte call:

W sys_ss: sys_ss_random_number_generator(pkg_id=2, buf=*0xd007e688, size=0x40)
F ... SIG: Thread terminated due to fatal error:
  sys_ss_random_number_generator(): Failed to generate pseudo-random numbers
  (in file .../sys_ss.cpp:150[:3]) (errno=19)

(errno=19 is the "may be irrelevant" stale-errno print from StrFmt.cpp, not the cause.)

Likely also affects FreeBSD via the S_ISCHR || S_ISBLK branch. Linux is unaffected (requires S_ISBLK plus major 11/7).

Suggested fix: restrict the macOS check to real disk nodes (/dev/rdisk*), or probe with DKIOCGETBLOCKCOUNT, which fails on non-disk character devices.

Build with regression

• First bad: 0.0.42-19896 (a99fde6f, the Add BD Drive support for Linux, macOS and FreeBSD #19233 merge commit)
• Last good: 0.0.42-19894 — the identical call succeeds there
• Still present on latest master: 0.0.42-19909

All three attached logs use identical settings on the same machine. The last-good log shows 3 successful RNG calls and 0 failures; both broken logs fail on the first call.

System configuration

Apple M4 Max, macOS 15.7.3, native arm64 build. Seen with MotorStorm: Apocalypse (BCUS98242, v1.06), but not game-specific.

Last working build

v0.0.42-19894-d9799c7f

First broken build

v0.0.42-19896-a99fde6f

Attach 3 log files

1-last-working-0.0.42-19894.log.gz
2-first-broken-0.0.42-19896.log.gz
3-latest-main-arm-0.0.42-19909.log.gz

I have attached all 3 required log files

Yes

Attach the RSX (.rrc) or RenderDoc (.rdc) capture files for visual issues

No response

System configuration

Apple M4 Max, macOS 15.7.3, native arm64 build. Default config aside from
RSX FIFO Fetch Accuracy: Fast &
Multithreaded RSX: true
Which seems unrelated was just trying to wrangle motorstorm on m4