Document Linux drive detection

CHANGELOG.md

@@ -45,6 +45,16 @@ Found by adding CI on Linux and macOS — the suite had only ever run on Windows
   file would have reported 34 million "frames" and a duration to match. The
   video track is now chosen by handler type. Found by running the parser over
   510 real clips, 2.84 TB, from two camera bodies.
+- **Linux cards never reached the drive panel.** Cards mount at
+  `/media/<user>/<LABEL>` or `/run/media/<user>/<LABEL>`, one level below
+  where the scan looked. It listed the empty `/media/<user>` directory and
+  another user's unreadable mount as fake removable drives instead. Volumes
+  now come from `/proc/self/mountinfo`, which names real mount points and
+  gives the filesystem type, so pseudo and system filesystems are filtered
+  by type while NFS and SMB shares stay listed as network volumes, outside
+  card detection. An overlay root (containers, live sessions) is still
+  treated as the system volume. Without `/proc` the old directory scan is
+  the fallback.
 
 ### Changed
 

CONTRIBUTING.md

@@ -37,7 +37,7 @@ pip install -e ".[dev]"
 `ffmpeg` and `ffprobe` on `PATH` are optional — the suite runs without them.
 
 ```sh
-pytest                      # ~400 tests, about 33s
+pytest                      # ~416 tests, about 33s
 pytest --fuzz               # property tests at 3000 examples each, about 2 min
 ruff check src tests
 pytest --cov=offloader --cov-report=term-missing
@@ -62,6 +62,7 @@ suite fakes all of it:
 | A failing destination | `tests/test_data_safety.py` returns handles that raise on write |
 | Corruption | flip a byte and re-verify; size stays identical, checksum does not |
 | An interrupted recording | write a BRAW with `include_moov=False` |
+| A Linux card mount | the `linux_mounts` fixture in `tests/test_volumes.py` writes a fake `mountinfo` and points the scanner at it |
 | A deep Windows path | build until `len(str(path)) > 260` — do not hard-code, the temp root's length varies |
 
 If you have real hardware, there are tests that use it and skip cleanly when it

README.md

@@ -244,7 +244,11 @@ Volumes that look like camera media are badged `CARD` and sorted to the top —
 detected by the marker directories cameras write (`DCIM`, `PRIVATE`, `XDROOT`
 and friends) or by a root full of camera originals, since a Blackmagic card
 writes clips straight to the root and a reader in a dock reports as a fixed
-disk.
+disk. On Linux the panel reads the kernel's mount table, so a card mounted under
+`/media/<user>/` or `/run/media/<user>/` appears with its filesystem type.
+Pseudo and system mounts stay out of the list; NFS and SMB shares are kept
+and classified as network volumes, so they can be chosen as destinations,
+but they are never offered as cards.
 
 Along the bottom is the **queue**. Jobs run one at a time — offloads are I/O
 bound, and running two at once against the same bus makes both slower and the
@@ -356,17 +360,17 @@ what makes the report layer testable without moving bytes.
 
 ```sh
 pip install -e ".[dev]"
-pytest                      # 409 tests, ~33s
+pytest                      # 416 tests, ~33s
 pytest --fuzz               # same suite, 3000 examples per property (~2 min)
 ruff check src tests
 pytest --cov=offloader --cov-report=term-missing
 ```
 
-409 tests at 82% line coverage. They cover formatting against the reference's
+416 tests at 81% line coverage. They cover formatting against the reference's
 exact strings, checksum vectors and streaming equivalence, copy/verify
 behaviour including simulated destination corruption, pause/resume/cancel
 concurrency, retry discrimination, BRAW container parsing, ffprobe parsing,
-preset and history persistence, card detection, PDF geometry read back with
+preset and history persistence, card and mount-table detection, PDF geometry read back with
 PyMuPDF, the CLI, and the GUI.
 
 The GUI tests run on Qt's offscreen platform and drive the real queue