- Initialize PipelineContextSender.delete_suppressed=false: an uninitialized
true silently skipped the --delete keep-set manifest under -m/--threads,
so destination extras were never removed.
- Allocate/install the receiver deleted-path observer only when
report_deletes is set (--info=del / -i / --out-format under --delete), cap
the retained list at MAX_MANIFEST_ENTRIES, and free already-created lists
on the receiver-pipeline create failure path.
- Validate report_deletes/report_stats/report_dest_info on receive.
- Correct stale comments (config.h report_deletes, delete_plan.h deleted
count, multiprocessing.h stats locking, utils.h observer placement).
- Tests: sender delete_suppressed init, report_deletes gating (unit), and
-m/--delete default delete-after keep-set removal (integration).
Address findings from the four-agent review of PR #298:
- file_create: zero the new File.matched_bytes. It was uninitialized
malloc memory, so the receiver could sum a garbage value into
STATUS_STATS "Matched data" (nondeterministic --stats divergence and
an uninitialized-heap disclosure on the wire).
- send_append: load the source into memory before hashing/copying the
prefix and tail. Files >64 MiB without compression (and --sendfile
runs) are streamed without loading, so --append/--append-verify
dereferenced a NULL data->data and crashed.
- filter_file_append: clamp the rollback to the surviving rule count.
A "clear" rule in a merge file frees every rule including the
caller's; the old rollback rewound count to rules_before and
resurrected freed pointers for a double free / UAF. Also roll back
when set_rule_owner fails instead of leaving owner-less rules.
- filter_rule_parse: reject the xattr-name filter modifier (x), which
was parsed and silently reinterpreted as a filename rule (affecting
what --delete protects). The p modifier stays accepted (existing
grammar test).
- Remove two dead functions: compression_default_algo and
change_render_itemize_code.
- tests: free ctx->would_delete in the two test_multiprocessing manual
teardowns (ASan leak, 1648 bytes/run).
- docs: correct the RSYNC_COMPAT/HANDOFF tally (156 rows: 106/27/23),
downgrade --info/--debug to caveat with their silent categories, add
%C-vs-xxh64 and --delete-delay count caveats, refresh stale xattr
mode comments, and document -p special-bit (setuid/setgid/sticky)
parity plus its mitigations.
After the writer releases bytes, the blocked enqueuer may already have
admitted the next payload, so the transient queued_bytes==0 read is
scheduling-dependent (lost the race under ASan). The post-join state
(covers unblocking + budget re-limit) is deterministic.
The per-connection memory budget (MAX_CONNECTION_MEMORY, 256 MiB) only
charged wire buffers via receive_data_limited. Decompression buffers and
per-file chunk copies were not accounted for, and the multithreaded
receiver could enqueue up to 100 files (each up to 64 MiB uncompressed)
ahead of a slow disk writer, retaining ~6.4 GiB per connection. A client
sending highly compressible chunks with little bandwidth could OOM the
host while the reserve never tripped.
Bound the receive pipeline by aggregate payload bytes instead of item
count alone:
- Export MAX_CONNECTION_MEMORY from protocol.h.
- PipelineContextReceiver tracks queued_bytes (payload bytes received but
not yet released by the disk writer, i.e. queued or in the writer's
hand) under the existing mutex.
- receiver enqueue now blocks while the queue is full by count OR when
adding the file would push queued_bytes over the configured byte limit,
applying backpressure to the sender instead of failing the transfer.
- The disk writer releases the byte budget after each file is freed and
signals the not-full condition.
- The server sets the byte ceiling to
MAX_CONNECTION_MEMORY - 2*MAX_CHUNK_SIZE so that the queued payloads
plus the transient wire/decompression buffers of the one in-flight
chunk stay within the per-connection budget.
The single-threaded receive path is already bounded: it writes files to
disk before reading the next chunk, so its transient is at most one
chunk's wire + decompressed + copied payload (~3 * MAX_CHUNK_SIZE, below
the budget). Wire buffers remain charged exactly once by
receive_data_limited; this change does not double charge them.
Adds a deterministic unit test in test_multiprocessing.c proving that an
enqueue which would exceed the byte budget blocks until the writer
releases bytes.
- Fix transport_tcp.c: initialize server->ssl_ctx to NULL
(prevents SSL_CTX_free on garbage when server_create_tls fails)
- Fix test_transport_tcp.c: client_create now sets fd=-1, family=AF_UNSPEC
- Fix test_transport_tcp.c: server address family may be AF_INET or AF_INET6
- Fix test_file_sendfile.c: send_path=false protocol includes file_type prefix
TapTap