[fjtrujy]

Scope

This PR now contains only the HostFS directory fixes, independently reviewable and mergeable into master:

• Return HOSTFS_CMD_DREAD rather than HOSTFS_CMD_READ in a DREAD response.
• Free scandir() entries on successful directory enumeration (not just on errors), preventing allocation leaks during repeated directory opens.
• Use lstat() so dangling symbolic links remain visible as directory entries.
• Clear the directory pointer after freeing it on an error path.
• Add one native C regression test covering the DREAD wire response, dangling symlinks, and repeated open/close cycles.

Fixes #57.

Verification

• make -C tests test — passed on macOS.
• HOSTFS_TEST_ITERATIONS=5000 make -C tests test — passed.
• make -C usbhostfs_pc -j4 — passed with only the pre-existing unrelated newpathlen warning.
• git diff --check — passed.

Other independent fixes split out of the original PR

• Validate PSP USB bulk transfers and HostFS replies #75 — PSP bulk-transfer submission, bounds checks and HostFS response validation.
• Rearm PSP asynchronous USB receives after errors #76 — PSP asynchronous receive rearming and lifecycle handling.
• Use the supplied libusb handle for host bulk reads #77 — Mac libusb bulk reads must use the supplied device handle.

The original combined PR commit (b45ad24) was preserved in a local backup branch before narrowing this PR. All four independent patches together reproduce the exact same Git tree as the original combined PR.

Out of scope

The intermittent multi-hour USB/pspsh hang is tracked separately in #74. Neither this directory PR nor the three extracted fixes alone claims to resolve the root cause. Experimental wake-recovery work is not part of these PRs.

[sharkwouter]

Could you still clean up the commit history and the files in this one still? It still looks a bit unfinished.

[fjtrujy]

@sharkwouter Thanks for the review. I've cleaned this up and force-pushed the rewritten branch (b45ad24):

• Squashed the four incremental/WIP commits into one focused commit, rebased on the updated master.
• Reduced the diff from 8 files (~688 additions) to 4 files (176 additions, 33 deletions): the PSP/PC HostFS fixes plus one native-C directory/DREAD regression test.
• Removed the long investigation report, mock pspsh, and standalone hardware smoke-test sources from this PR. The original work is preserved locally, and the long-uptime investigation is tracked in PSPLinkUSB / pspsh stops responding after many hours of mostly idle USB connection #74 rather than mixed into this patch.
• Updated the PR title and description to specify the verified fixes and explicitly state that the overnight hang is not resolved by this PR.
• Verified the focused C test, PSP kernel module and host binary builds, and git diff --check. CI is running again on the rewritten head (Linux/macOS/Docker already green at time of writing).

Would you mind taking another look at the revised, smaller diff? Thanks!

[fjtrujy]

I've split this PR into four independently reviewable changes, preserving this PR's original discussion and reviews:

• This PR (Fix HostFS directory leaks, dangling symlinks and DREAD replies #72): HostFS directory enumeration, memory cleanup, correct DREAD response and native C regression test.
• Validate PSP USB bulk transfers and HostFS replies #75: PSP bulk-transfer and HostFS reply validation.
• Rearm PSP asynchronous USB receives after errors #76: PSP asynchronous receive/rearm and lifecycle handling.
• Use the supplied libusb handle for host bulk reads #77: Correct libusb handle passed to Mac bulk reads.

The original combined commit b45ad24 is backed up locally. I verified that applying the four standalone patches to the same master commit gives the exact same Git tree as b45ad24; nothing was dropped.

The extracted PSP PRs build independently with the PSP toolchain; the Mac host PR and the directory-only PR also build, and the directory regression test passes at 5,000 iterations.

Thanks for reviewing the smaller, independently mergeable changes. The unrelated long-idle investigation stays in #74.