Rust: IMFS Grate

[rennergade]

Summary

• Full Rust rewrite of the IMFS (In-Memory Filesystem) grate, replacing the C implementation
• Uses fdtables from lind-wasm/src/fdtables for all fd management instead of the C version's static g_fdtable[MAX_PROCS][MAX_FDS] arrays
• Uses grate-rs GrateBuilder for the grate lifecycle (fork/exec/wait/semaphore)
• Includes a C test binary with 11 test cases

How it works

fdtables integration:

• underfd = node index (which IMFS node this fd points to)
• perfdinfo = open flags (O_RDONLY, O_WRONLY, etc.)
• Only custom state: a HashMap<(cage_id, fd), offset> for per-fd read/write position — everything else is fdtables
• copy_fdtable_for_cage on fork, empty_fds_for_exec on exec, close_virtualfd on close

Filesystem:

• Arena-style node storage in a Vec with free list (no raw pointers)
• File data in linked chains of 1024-byte Chunks
• Directory tree with . and .. link nodes
• Path resolution walks the tree, following symlinks

Registered syscalls: open, close, read, write, lseek, fcntl, unlink, pread, pwrite, fork, exec

Files

examples/imfs-grate-rs/
Cargo.toml
src/
main.rs - GrateBuilder setup, preload support
handlers.rs - 11 extern "C" syscall handlers
imfs/
mod.rs - ImfsState, fs operations, chunk read/write, path resolution
node.rs - Node, NodeType, Chunk, DirEntry types
test/
imfs_test.c - C test binary (11 tests)

[Yaxuan-w]

Unnecessary file

[drapl0n]

header file sys/stat.h is required for mkdir() call

[stupendoussuperpowers]

Tested this, added comments here but also opened a PR which merges into this branch with the fixes for the issues I stated.

Take a look at that PR here: #50

• Additional work is needed for fork/exec as I've mentioned in these comments, specially vis-a-vis fdtables usage.
• There are bugs in the fork/exec handlers
• We need some PRs to go into Lind for other behaviours (Lind-Project/lind-wasm#967 & Lind-Project/lind-wasm#970)
• There were some minor error handling and quality of life changes for imfs/
• Looks like these files aren't properly linted? A lot of the diff on the new PR is just cargo fmt fixes.

[stupendoussuperpowers]

Fixed behaviour around fork and updated the tests.

Re-requesting reviews.