[filtede98]

Summary

This patch strengthens input validation in the thread pool (lib/common/pool.c) and adds NULL checks in the multithreaded compression helpers (lib/compress/zstdmt_compress.c).

1. Bounds in the thread pool:

   • POOL_create_advanced and POOL_resize_internal reject numThreads above POOL_MAX_THREADS (10000, defined in pool.h), as suggested in review.
   • POOL_create_advanced rejects queueSize >= (size_t)-1 / sizeof(POOL_job). Without it, queueSize + 1 can wrap around, and the worker and dispatch routines then do % ctx->queueSize with a zero queue size.

2. Pointer checks in the thread pool:

   • POOL_joinJobs returns early when ctx is NULL, like POOL_free and POOL_sizeof.
   • POOL_add and POOL_tryAdd now assert that function is non-NULL, in both the multithreaded and single-threaded implementations. The assert on ctx was already there.

3. NULL checks in the multithreaded compression helpers: ZSTDMT_sizeof_bufferPool, ZSTDMT_setBufferSize, ZSTDMT_sizeof_CCtxPool, ZSTDMT_getCCtx and ZSTDMT_releaseCCtx.

4. Tests: tests/poolTests.c checks that POOL_create rejects an overflowing queueSize and numThreads and values above POOL_MAX_THREADS, that POOL_resize rejects (size_t)-1 and POOL_MAX_THREADS + 1, and that POOL_joinJobs(NULL) is safe. The new asserts and the NULL checks in zstdmt_compress.c have no dedicated tests.

Signed-off-by: Filippo Tedeschi filippotedeschi98@gmail.com

[terrelln]

1. Lets put a much smaller limit on the number of threads, define it as a macro in the header. I'm thinking like 10K. No reason to approach 2^32.
2. It is a programming error to pass a NULL ctx or function. So these could be asserts.

[filtede98]

Thank you @terrelln for the review! Updated as requested:

1. Defined POOL_MAX_THREADS as 10000 in pool.h and bound numThreads against it in POOL_create_advanced and POOL_resize_internal.
2. Converted the checks on ctx and function in POOL_add and POOL_tryAdd into asserts (assert(ctx != NULL); assert(function != NULL);).
3. Updated unit tests in poolTests.c to test the POOL_MAX_THREADS limit.

[ceedot-rock]

@Cyan4973 — the author addressed terrelln's review (POOL_MAX_THREADS 10000 cap, asserts in POOL_add/POOL_tryAdd) back on Sep 30 and it's been waiting on re-review since. terrelln's out for October — could you take the re-review?