Split FastSync's single use_metadata bundle into four independent rsync-parity attributes: preserve_perms, preserve_times, preserve_owner, preserve_group. use_metadata is now a derived transport bit (config_derived_use_metadata).
CLI: real -p/--perms, -t/--times, -o/--owner, -g/--group plus --no-perms/--no-times/--no-owner/--no-group (short and long) and --no-preserve; -a is now rsync -rlptgoD; --preserve = -pt; -A implies -p; -X does not; --chmod implies -p; --usermap/--groupmap/--chown imply owner/group per side; --incremental/--delta still auto-preserve unless negated.
Receiver: per-attribute FileAttrPolicy gating for files, dirs (modes applied at end of transfer), symlinks and specials; rsync -E read-bit rule; new files get source_mode & ~umask sanitized (no group/other write); per-side identity resolution; deferred directory metadata; batch dir-metadata replay; daemon modules without 'client owner = yes' no longer refuse plain -a but force super off (no ownership) with a warning.
Wire: PROTOCOL_VERSION 2.21.0 -> 2.22.0 (four appended config bools, golden 653 / 95530566005420798). FileMetadata/chunk/batch framing unchanged. Docs/CHANGELOG/CMake updated to 2.22.0.
Follow-up to a237043 addressing three security/correctness re-review findings.
(1) MEDIUM: a server-contacting --dry-run with --compare-dest/--copy-dest/
--link-dest still read and hashed the basis file and compared it with the
client-supplied digest, a 1-bit content oracle. basis_match_find() gains a
hash_content parameter; the dry-run shortcut passes false and returns no
match without touching basis bytes, so an otherwise-matching entry is
reported as would-transfer. The real (non-dry-run) path is unchanged.
(2) LOW: xattr_capture_path() hardcoded preserve_acls=true, so the receiver's
hard-link copy fallback re-applied system.posix_acl_* even when -A was not
negotiated. The function now takes preserve_acls and members.* is
unaffected; scanner and receiver callers thread the negotiated flag.
(3) INFO: the --fsync --link-dest temp reopen now uses O_NONBLOCK and treats
a raced-in FIFO's ENXIO as a benign fsync-skip instead of blocking.
Tests: dry-run + basis unit test (asserts would-transfer, no content read) and
integration test; xattr capture ACL-filter test. Verified strict build, ASan,
clang-format, cppcheck, and the CI integration subset.
Address confirmed receiver security findings B1-B6:
B1 (HIGH): add O_NONBLOCK to the three receiver read-opens that opened an
existing destination/basis entry before the S_ISREG gate
(incremental_check_open_destination, basis_open_regular, hardlink_read_source)
so a client-planted FIFO can no longer block the receive thread forever while
the post-open type gate still rejects it.
B2 (HIGH/MED): --inplace now fstatat(AT_SYMLINK_NOFOLLOW)-probes the target and
refuses any existing non-regular entry, opens with O_NONBLOCK, and re-checks
S_ISREG on the opened fd. This stops a FIFO from hanging the open and stops a
char/block device from being written directly (bypassing --write-devices).
B3 (MED): under --dry-run the incremental quick-skip no longer reads/hashes the
destination file for --checksum/--delta; it decides from metadata only and
reports would-transfer when the comparison is inconclusive, closing the
read-only-module content-hash oracle.
B4 (LOW): xattr_name_appliable() now gates the two system.posix_acl_* names on
preserve_acls (--acls), not the derived use_xattrs (--xattrs OR --acls). The
receiver drops (never applies) ACL entries when -A was not negotiated while
keeping user.* working for -X.
B5 (INFO): receive_manifest_section() charges a per-entry overhead against
MAX_MANIFEST_BYTES and the aggregate entry count across all three sections is
capped at MAX_MANIFEST_ENTRIES.
B6 (MED): data_charge_session() reserves decompressed/chunk-copy bytes against
the owning ProtocolSession (MAX_CONNECTION_MEMORY) and records them on the Data
so data_destroy() releases them via the Data.owner path. Applied to the
whole-file/append/delta decompression sites and chunk_deserialize() per-file
copies; a missing session owner degrades to the previous uncharged behavior.
Tests: FIFO destination/basis non-hang (with alarm), --inplace FIFO/device
refusal, dry-run no-read oracle test plus updated metadata-only dry-run tests,
ACL-without--acls drop, manifest total-entry cap, and chunk session charging.
--dry-run --read-batch=FILE still wrote to the destination because
batch_read_apply -> file_save_to_disk_full bypassed the per-caller
!dry_run guards. Guard file_save_to_disk_full and manifest_delete_all
directly (return SKIPPED/no-op) so every save/delete path is mutation-free
in dry-run, and keep the per-caller guards. Reject --dry-run combined with
--read-batch/--only-write-batch at CLI validation with a clear error (a
dry-run of a local batch apply is not meaningful).
Today a server rejection sends a bare STATUS_ERROR and the reason only
reaches the server log, so the client cannot say why a transfer was
refused. Add an optional, bounded server->client error-detail frame:
- Status gains STATUS_ERROR_DETAIL appended LAST so existing wire
values are unchanged.
- send_error_detail(fd, msg) sends STATUS_ERROR_DETAIL followed by the
existing length-prefixed string primitive, slicing over-long messages
to MAX_ERROR_DETAIL_BYTES (4096).
- receive_status() (and the timed/keepalive status readers) always
consume the detail body and map the status back to STATUS_ERROR,
capturing the text into a thread-local buffer exposed by
protocol_last_error(); a bare STATUS_ERROR leaves it cleared. Every
existing call site keeps working and the stream cannot desync.
- Upgrade the daemon module gate / config validation (config.c), the
final transfer failure (server.c) and receiver-side path/node
validation (file_receive.c) to send a concrete reason; surface it on
the client in client_send.c/config.c.
- Bump PROTOCOL_VERSION to 2.21.0 (CMake VERSION, CHANGELOG, docs) and
update the pinned config wire golden hash / CLI-version tests.
- Add tests/test_protocol_error.c covering mapping+capture, the
over-long bound, bare-error clearing, and thread-locality.
The per-file STATUS_CHECK fast path was a single 534-line function that
was hard to review. Extract it into small static helpers called in order
by a short linear orchestrator:
- incremental_check_receive_request (receive/validate request frame)
- incremental_check_open_destination (secure open + stat)
- incremental_check_quick_skip (metadata/content skip decision)
- incremental_check_try_basis (compare/copy/link-dest)
- incremental_check_try_append_resume(--append tail resume)
- incremental_check_try_delta (block delta)
- incremental_check_try_fuzzy (--fuzzy basis)
- incremental_check_receive_full (STATUS_NEXT + whole file)
Pure refactor: the ordered sequence of wire operations
(send_status/send_n_data/receive_n_data/receive_status/receive_wire_str)
is byte-for-byte identical to the original, and every resource cleanup
is preserved (a unified idempotent cleanup replaces the duplicated
per-path close/free blocks). No functional changes.
--dry-run now handshakes with a remote/daemon receiver and reports what
WOULD transfer/skip based on receiver state, mutating nothing on either
side.
- Serialize Config.dry_run into the wire config frame and append
STATUS_DRY_RUN_TRANSFER to the status enum (no renumbering); bump
PROTOCOL_VERSION/CMake VERSION/CHANGELOG/golden wire to 2.21.0.
- Receiver: receive_incremental_check_ex runs the normal read-only
decision and answers STATUS_OK (skip) or STATUS_DRY_RUN_TRANSFER
(would transfer) with no basis materialization/append/delta/full
transfer. All mutation sites are guarded by !dry_run: file store,
manifest deletes, --mkpath root creation, --delay-updates staging,
publication, directory-time application, and outcome acks.
- Client: send_dry_run_remote connects, sends the config, checks each
regular file and prints the would-transfer set + trailer; no file data
or delete manifest is sent. Plain local destinations keep the
client-side manifest.
Deduplicate the repeated directory-time capture gate
(`config->use_metadata && !config->omit_dir_times`) used by the
sender-side (multiprocessing.c) and receiver-side (receiver.c) sinks
into a single predicate declared next to the DirTimeList machinery in
file_receive.h and defined in file_receive.c.
Behavior preserved: identical short-circuit condition and semantics,
no signature or protocol changes.
- A6: escape attacker-controlled file paths and the receive root in log
lines (file_receive, server, protocol DEBUG) with output_escape()
- A8: identity_set_active() returns bool and fails closed when a requested
usermap/groupmap cannot be deep-copied; handler refuses the connection
- remove the const cast and duplicate super_mode clamp from
server_module_gate via an explicit override the handler applies once
- release the identity snapshot on the queue_create failure path
- refactor identity_parse_copy_as to a single cleanup tail and drop the
duplicated group error format specifier
- file_ensure_directory_secure() now chowns a final directory it creates under
--copy-as and fails on error; the symlink parent-creation call site propagates
it. The is_dir branch fails when the confined parent cannot be opened under
--copy-as. Closes the residual wrong-owner gap for synthesized/symlink
parent directories.
- file_restore_symlink_metadata() early NULL return is copy-as-aware.
- Preserve errno across the implicit-parent failure cleanup.
- Neutral skip messages (the clamp, not --no-super, may be responsible).
- Daemon copy-as test tolerates the non-root privilege refusal; usage text lists
--copy-as.
- H3: a daemon module without 'client owner = yes' now also has super-user
device activity forced off (char/block mknod, --write-devices), so a root
daemon can no longer be made to create/write raw devices under AUTO. The
entries are skipped, preserving ordinary -a pushes.
- H1/H2: propagate a failed required --copy-as chown from symlink metadata
restore and implicitly-created parent directories, so the entry (and run)
reports failure instead of a wrong-owner success.
- Docs/help/headers updated for A2/A3 and the device clamp; startup warning
spells out the client-owner risk.
- Tests: daemon device clamp (skipped without opt-in, created with opt-in),
updated --super/--fake-super expectations.
A1: daemon refuses every client-chosen ownership/super-user request
(--numeric-ids/--chown/--usermap/--groupmap/--fake-super/--copy-as/--super)
unless the selected module opts in with 'client owner = yes'.
A2: fake-super owner replay requires an explicit ownership identity policy.
A3: --super no longer implies --numeric-ids (ownership stays opt-in).
A5: a failed --copy-as chown marks the entry failed instead of reporting
success with the wrong owner.
Add the receiver-side --super / --no-super tri-state (Config->super_mode)
under the safe-subset + clear-refusal privilege model: FastSync never
elevates privileges, it only permits super-user attempts that are already
confined fd-relative below the authorized receive root.
- identity: privilege_super_permitted() gate (OFF=false, ON=true, AUTO follows
geteuid()==0); identity_apply_ownership/_link become no-ops when not
permitted; --super with no explicit identity policy implies raw numeric-id
preservation (explicit usermap/groupmap/chown/numeric-ids still win); warn
exactly once when --super is requested by a non-root receiver.
- file_receive: gate char/block device-node creation on the gate; FIFO/socket
handling is unchanged.
- wire: trailing super_mode int after the --iconv spec, validated 0..2 in
receive_privilege_options and validate_received_config; PROTOCOL_VERSION
2.17.0 -> 2.18.0; version-sensitive tests and docs updated.
- CLI: --super/--no-super parsed explicitly before the generic --no-* branch
(malformed --super=x rejected); usage text added.
- tests: config wire round-trip + invalid-value rejection, privilege-gate mode
unit test, CLI parse test, integration transfer + root-gated ownership
suppression/appliance tests.
- docs: RSYNC_COMPAT --super row + Wave E note, protocol mentions, README.
Review fixes for Phase 7 Wave D.
#1 (HIGH): STATUS_DIR_TIMES entries no longer create directories. A new
receiver-only File.dir_time_only flag marks dir-time entries; file_save_to_disk_full
short-circuits them as FILE_SAVE_SKIPPED before any device/dir branch, so the sink
still accumulates metadata into the deferred DirTimeList but creates nothing. Empty
source dirs stay untransferred (-a), -m/--prune-empty-dirs semantics are preserved,
and a pre-existing regular file/symlink at an empty-dir mirror path no longer aborts
the transfer. dir_time_list_apply fstatat()s the leaf (AT_SYMLINK_NOFOLLOW) and skips
absent/non-directory paths QUIETLY; only a real existing directory is stamped.
Also initialize File.dir_time_only in file_create() (uninitialised garbage otherwise).
#2 (MED): send_dir_times() chunks entries into repeated STATUS_DIR_TIMES frames of at
most MAX_MANIFEST_ENTRIES, matching the receiver's per-frame bound; the tautological
> INT_MAX check is gone.
#3 (LOW): dir_time_list_add() assigns each grown array right after its realloc (no
dangling) and advances capacity only after both succeed.
#4 (LOW): RSYNC_COMPAT.md -- STATUS_MKDIR carries metadata, dir times are transmitted
via STATUS_DIR_TIMES and applied at the end, empty dirs are still never created; -m
rationale, -O row and Wave D notes updated. Summary counts untouched.
#5 (LOW): integration tests for the three #1 scenarios (empty-dir non-creation under
-a and -a -m, collision non-abort), scanner test now covers empty-dir capture, and
test_file_restore_symlink_metadata asserts the positive apply path when supported.
PROTOCOL_VERSION stays 2.17.0; config-frame layout unchanged.
Wave D of Phase 7. Make -O/--omit-dir-times and -J/--omit-link-times real by
preserving directory and symlink times, and mark --secluded-args as an explicit
Impossible/Divergence no-op.
Wire: PROTOCOL_VERSION 2.16.0 -> 2.17.0. Adds a terminal STATUS_DIR_TIMES frame
(int count + (wire path, metadata) pairs) sent after all file data and the
optional delete manifest. STATUS_MKDIR also carries metadata for --dirs entries.
Config-frame layout is unchanged.
Sender: the recursive scanner captures every traversed source directory (both
DirectoryScanner and the parallel scanner root + workers, appends mutex-guarded)
into a shared list; the single-threaded and -m paths transmit it last.
Receiver: a DirTimeList accumulates received directory metadata and applies it
with fd-relative no-follow utimensat only at the very end -- after all children,
after the commit-style --delete, and after --delay-updates publication -- in the
single-threaded success frame and in server.c after the -m threads join. -O skips
the application. Symlink metadata is applied at link creation with
utimensat/fchownat/fchmodat AT_SYMLINK_NOFOLLOW; -J suppresses only link times.
identity_apply_ownership_link shares the identity resolver with the fd path.
Docs: -O/-J rows -> Implemented; --secluded-args -> Impossible/Divergence;
--protocol accepted/rejected values and Phase-6/7 notes updated.
Tests: unit (scanner dir capture, DirTimeList apply, symlink metadata, protocol
version values) and integration (dir mtime round-trip + -O, symlink mtime
round-trip + -J, independent suppression), parameterized over single/multithread.
- write_all_sparse: skips all-zero runs >= 4096 bytes via lseek(SEEK_CUR) and
ftruncates the final size, wired into the atomic temp+rename and --inplace
paths with no wire change (full image already in memory).
- --partial retention: on a save failure after the temp held data, rename the
already-written temp to the destination path (best-effort; falls through to
unlink; never retains when --partial is off) so --append/--append-verify can
resume; tested by forcing futimens EINVAL with an out-of-range nsec.
- --block-size aliases --delta-block; verified config->delta_block_size is
honored by the delta engine end-to-end (unit + integration tests).
- fake_super_restore_fd: parses and re-applies user.fastsync.stat fd-relative
(fchown best-effort/non-root skipped, fchmod, futimens); a save under
--fake-super now both records and re-applies.
- -N/--crtimes and --stderr=client promoted to a new 'Impossible/Divergence'
status bucket (Summary: 136/2/4/3/2 = 147).
- cppcheck/clang-format clean; unit 37/37; integration 407 passed.
Capture+transmit source atime (pre-read stat; O_NOATIME sender guard) and birth
time (statx STATX_BTIME); receiver restores atime with mtime (crtime not settable
portably -> transmitted, explicitly not applied). --open-noatime is client-only.
-O/-J documented as accepted no-ops (FastSync never preserves dir/symlink times).
Wire: metadata frame gains atime/crtime val+sec+nsec; PROTOCOL_VERSION
2.11.0->2.12.0. Review fixes: gate atime capture to Linux (no epoch clobber on
non-Linux), close fd on fdopen failure, honest -O/-J status (Compat no-op).
Source files sharing (st_dev,st_ino) are recreated as hard links on the
destination; only the first member's data crosses the wire (siblings ride a
payload-less STATUS_HARDLINK frame). Ordering requires the single-FIFO-writer
receiver + forced sequential scan (documented). link()-failure falls back to a
byte-identical local copy. Rejects -s/--append. PROTOCOL_VERSION 2.11.0->2.12.0.
Review fixes: delete the dead HardLinkRegistry (ordering holds by FIFO writer),
and --existing no longer aborts when the first member is absent but the sibling
exists (leaves the sibling in place).
Receiver allocates the destination file's full size before streaming data
(posix_fallocate preferred, ftruncate fallback on EOPNOTSUPP/ENOSYS) so an
out-of-space transfer fails fast instead of partway. Additive config bool
crossing the wire; PROTOCOL_VERSION 2.10.0 -> 2.11.0. Threaded through all
store paths (atomic, inplace, partial/delay-updates staging, link-dest copy
fallback). Review hardening: explicit lseek(0) before the data write so
correctness does not depend on posix_fallocate leaving the fd offset unchanged.
BLOCKER: an absent mirror whose PARENT directory does not exist on the
destination (a deeper --files-from missing entry, -R or full-mirror layout) was
treated as a hard failure because file_open_secure_parent returned -1 when the
parent was missing. That aborted the whole run, skipped the --delete extras
walk, and tore down an early --delete-before/during connection, contradicting
'missing mirror = no-op / all-missing succeeds'. A parent-open failure is now a
no-op when errno is ENOENT/ENOTDIR (matching file_remove_tree_secure); only a
genuine I/O error fails the run.
Also: an already-absent mirror reached via unlinkat-ENOENT no longer prints
'Deleted: <path>' (a no-op dressed as a deletion); the per-path 'Deleted:' line
is printed only when an entry was actually removed.
Adds real algorithm selection (xxh64 default, plus md5 via OpenSSL EVP) and a
64-bit seed for the per-file whole-file digest used by the --incremental/
--checksum handshake and basis-dir content verification. The seed also feeds
the delta path's per-block xxHash32 strong checksum (low 32 bits) so an
explicit seed deterministically changes those digests too. Sender and receiver
hash identically: the algorithm id and seed cross the config wire frame and the
STATUS_CHECK handshake now carries a length-prefixed, bounded digest instead of
a fixed 64-bit value. Unsupported algorithm names are rejected at parse time
(never a silent no-op). PROTOCOL_VERSION bumped 2.9.0 -> 2.10.0; defaults
(xxh64, seed 0) preserve prior byte-for-byte behavior.
Implement rsync's append modes: when an existing destination file is SHORTER
than the source, the receiver negotiates a resume offset and only the tail is
transferred; the full file (retained prefix + tail) is rebuilt and installed
through the normal atomic store path, so the result is byte-identical to the
source whenever the prefix matches.
- --append: sends the tail without content-verifying the retained prefix
(rsync parity; the documented prefix-trust risk).
- --append-verify: verifies the retained prefix against the source's prefix
xxHash64 before appending and, on a mismatch, falls back to a clean full
transfer (never a corrupt prefix+tail blend).
New wire frames STATUS_APPEND / STATUS_APPEND_SIG / STATUS_APPEND_OK /
STATUS_APPEND_DATA; PROTOCOL_VERSION bumped 2.9.0 -> 2.10.0 (peers must match).
Both flags imply --incremental and are incompatible with -s (chunk
serialization) and --whole-file (rejected up front). Respects --inplace,
--partial/--partial-dir and --delay-updates via the shared store engine.
PROTOCOL_VERSION 2.9.0 -> 2.10.0. The STATUS_MANIFEST frame gains a third
section carrying destination-relative exact-delete paths (the missing
--files-from entries' mirrors); the config frame gains a delete_missing_args
bool (ignore_missing_args stays client-only). The receiver validates the third
section like the keep-set and commits it with manifest_delete_all():
manifest_delete_missing_args runs first (explicit user requests, never blocked
by protected-prefix exclusion protection; staging/basis protected; a non-empty
directory mirror removed only under --force/--delete, rsync parity) and then the
ordinary extras walk. Server --allow-delete gates it like --delete.
The suffix-trim loop using computed end offsets tripped cppcheck's
knownConditionTrueFalse value-range analysis (it unsoundly concluded the trims
always consume the whole middle). Rewrite it with explicit moving end indices
and add an inline suppression with a rationale for the residual false
positive; cppcheck --error-exitcode=1 is clean again. The trimming logic is
unchanged and was verified against a full DP reference over 200k random name
pairs.
Review follow-ups on the --fuzzy candidate scan:
- Allocate the two DP rows once per directory scan instead of once per
candidate (4096-entry directories no longer do thousands of malloc pairs).
- Pre-prune before the DP with two cheap lower bounds on the edit distance:
the name length gap and the count of characters of one basename absent from
the other; a candidate whose gate (distance*2 <= longer) already fails on
the max of those bounds is skipped without running the DP.
- Trim the common prefix and non-overlapping common suffix before the DP so
it only runs over the differing middles.
- Note that the 4096 readdir cap bounds iterations, not per-entry DP cost,
and that the seen set is filesystem-order dependent (winner stays
deterministic via the total comparator).
- Open the chosen candidate with O_NONBLOCK so a name raced to a FIFO cannot
block the receive thread forever in open(2); the existing fstat S_ISREG gate
still rejects non-regular files. (The pre-existing basis_open_regular has
the same latent FIFO pattern and is intentionally left unchanged.)
- Free old_data in receive_delta_file's defensive NULL guard.
The STATUS_MANIFEST frame now carries two count-delimited sections: the kept
paths and a protected-prefix list (excluded-on-source paths the walker must not
delete unless --delete-excluded opted out). The receiver's DeleteManifest is
passed through the commit/early paths unchanged. manifest_delete_extras
honors a client --max-delete (all-or-nothing) and produces a distinct error for
it versus the 100000-entry server bound. --force clears a non-empty directory
that blocks an incoming regular file (confined, symlink-safe) via a new
file_remove_tree_secure helper.
When a file must be transferred and the destination holds no usable content
at the exact path (file absent, or outside the delta engine's size bounds),
the receiver now searches the target's own destination directory for an
existing regular file with a similar basename and uses it as the delta basis
via the existing receiver-driven STATUS_DELTA_SIGNATURE handshake. The
sender never learns the basis was another file, so no wire change beyond the
new config flag was required.
Heuristic (deterministic, simpler than rsync's, documented): candidates are
sibling entries confined below the root and opened O_NOFOLLOW (symlinks are
never followed, nothing outside the destination root is read); dotfiles,
directories, the target's own name and stage/temp names are excluded; size
gate is delta_should_attempt; name gate is a Levenshtein distance <= half
the longer basename; the closest candidate (size tie-break, then lexical) is
loaded; the scan is capped at 4096 entries.
Byte-exactness is independent of the basis: block matches are verified by
Adler-32 + xxHash32, delta_apply validates every reference, and a basis that
shares nothing makes the sender reply with a whole-file transfer. When no
candidate qualifies the normal whole-file transfer runs unchanged.
- copy-dest basis hits leaked the heap-allocated basis path: BASIS_DEST_COPY
did not transfer it (only LINK does) and returned before the basis cleanup.
basis_match_free is now called on every materialization return path (success
and send-failure) after content/link ownership is transferred.
- the --delete walker regression: the delay-updates staging name must be
protected only as a DIRECT child of the receive root, while basis dirs may
be skipped at any depth. delete_extras_limited now takes DeleteSkipEntry
entries carrying a top_level_only flag instead of a flat prefix list, so a
nested destination directory named .fastsync-stage is ordinary content again
(its extras are deleted) and a basis tree is still never removed.
Deletion timing is now real and selected by the four rsync flags plus the
plain --delete default. Wire protocol bumps to 2.8.0: two new config
booleans (delete_during, delete_delay) are serialized and validated, joining
the existing delete_before/delete_after.
- Early modes (--delete-before, --delete-during/--del): the sender pre-scans
the whole tree (paths only), transmits the keep-set manifest BEFORE any
file data, and the receiver removes extras and acks STATUS_OK; the sender
only streams data after the deletion committed. Deletion is thus performed
even if a later transfer phase fails (rsync delete-before/during are
destructive by definition). FastSync streams in a single scan so it cannot
interleave per-directory like rsync delete-during; --delete-during selects
the same engine mode as --delete-before (documented divergence).
- Late/commit modes (plain --delete, --delete-after, --delete-delay): the
manifest closes the data stream and deletion is committed only after
STATUS_FINISHED proves the whole transfer succeeded, preserving FastSync's
commit-style safety. --delete-delay converges with --delete-after because
FastSync never snapshots the destination during data flow (documented).
- The STATUS_MANIFEST frame is now self-delimiting and position-independent.
Single-threaded receivers delete before the success frame; the -m receiver
hands the keep-set to server.c, which commits the deletion only after the
disk writer thread has drained (fixes a delete-vs-in-flight-temp race).
- Every timing flag implies --delete; at most one timing flag is allowed.
- Each timing flag implies --delete, matching rsync; conflicts are rejected.
The receiver's per-file incremental check now consults the ordered basis-dir
list whenever the destination is not already up to date. An exact basis match
requires equal size, equal mtime (unless --size-only; --ignore-times disables
basis matching like rsync), and an equal content xxHash64 -- the sender sends
its xxHash for every file whenever basis dirs are configured (not only under
--checksum), so a hard link or local copy is only ever made from byte-identical
content.
On a match:
- compare-dest: reply STATUS_OK and skip data only when the destination does
not already hold the file (sparse, rsync parity). A destination that holds
a DIFFERENT version falls back to a normal transfer instead of rsync's
delete, keeping the mirror complete.
- copy-dest: reply STATUS_OK and hand a synthetic File (bytes read from the
basis file, basis metadata) to the normal store sink, so the file is
installed as a real local copy through the existing atomic temp+rename
engine and honors --existing/--ignore-existing/--update/--backup/
--delay-updates/--partial-dir unchanged.
- link-dest: same, but File.basis_link records the basis path and the store
engine calls the new file_to_disk_secure_link(): an atomic temp hard link +
rename. Cross-filesystem/refused links fall back to a byte-identical local
copy (never a corrupt or partial file); the copy fallback applies metadata,
while a successful link keeps the basis inode's own attributes so the basis
file is never mutated.
Basis-materialized files carry File.skip so they are not acknowledged to a
--remove-source-files sender (the sender already saw STATUS_OK and keeps the
source). The no-match path is byte-for-byte identical to the existing delta /
full-data transfer.
TapTap