PartFile: harden per-part hashing against gap/disk-state drift - #560
Merged
mrjimenez merged 2 commits intoMay 11, 2026
Merged
Conversation
CPartFileHashThread's main-thread enqueue point in FlushBuffer Phase 3 queues every dirty part regardless of whether the part's gap-list range is fully filled. The synchronous verify loop in ~CPartFile (line 286) already gates on IsComplete; the async path lost that gate when the hash thread was introduced. Without it, a write that lands inside a part marks the part dirty and queues HashSinglePart even when the part still has gaps. HashSinglePart then reads PARTSIZE bytes from the partfile, walking past the highest written offset on a partfile that is sized to "last write end + 1" under the default sparse-no-preallocation behaviour on Linux/macOS, and trips CEOFException -> SetStatus(PS_ERROR). Restore the gate, mirroring the sync path: clear the dirty bit unconditionally so the loop doesn't busy-spin on incomplete parts, and trust that future writes to the same part will re-set it once the part is gap-complete.
HashSinglePart trusts the gap-list unconditionally: it computes
offset = PARTSIZE * partnumber, length = GetPartSize(partnumber), and
hands them to CreateHashFromFile. The gap-list and the partfile's on-
disk length can diverge in several scenarios:
- A write that the gap-list believed succeeded actually failed
(ENOSPC / EIO / transient I/O), the .met got persisted before the
retry, and the daemon was killed unclean before retry succeeded.
- The partfile was truncated externally (FS hiccup, manual edit,
partial backup restore).
- aMule on Linux/macOS uses sparse-no-preallocation by default
(CreateSparseFile ignores the size argument; partfile size grows
only with writes, never reaches the target file size unless the
last write happens to extend it).
In each case CreateHashFromFile reads past the partfile's actual length
and throws CEOFException, which the existing handler converts to
SetStatus(PS_ERROR). Recovery requires the user to truncate -s <full
size> the partfile manually before the file can resume.
Detect the divergence pre-read: when partfileLen < offset+length,
reopen the gap via AddGap and return false (caller treats the part as
corrupt and the gap-list opens those bytes for re-fetch). Recovery
then happens transparently on the next download attempt instead of
demanding manual intervention.
This was referenced May 9, 2026
Closed
got3nks
pushed a commit
to got3nks/amule
that referenced
this pull request
Jul 24, 2026
* Fix alcc: preserve Unicode characters in ed2k links The `CleanFilename()` function in alcc used a regular expression `[^[:alnum:]_.-]` to replace every character not in the set of alphanumerics, dot, underscore, and hyphen with an underscore. This stripped all non‑ASCII characters (e.g., Chinese, Japanese, Cyrillic) from filenames, turning them into a series of underscores in the generated ed2k:// links, making them unreadable. This commit rewrites `CleanFilename()` to replace only the three characters that actually break the ed2k link format: `|` (field separator), `/` (end‑of‑link marker), and `\` (path separator on Windows). All other characters, including Unicode, are left untouched. As a result, `alcc` now produces fully readable ed2k links with the original filenames, while still preventing malformed links. The link integrity is maintained because the ed2k protocol identifies files by their hash and size, not by the filename string. This change has been tested with filenames containing Chinese characters under a UTF‑8 locale (LANG=zh_CN.UTF‑8) on Debian GNU/Linux. * Update ed2khash.cpp
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Surfaced in the follow-up thread on #302: when a partfile's logical length is shorter than what the gap-list claims is filled, per-part hash verification trips
CEOFException->SetStatus(PS_ERROR), requiring manualtruncate -s <full size>to recover. Two narrow changes to make per-part hashing robust against gap-vs-disk-state drift.Fix
1. Gate the async hash enqueue on
IsComplete(part)—CPartFileHashThread's main-thread enqueue point inFlushBufferPhase 3 queues every dirty part regardless of whether the part's gap-list range is fully filled. The synchronous verify loop in~CPartFilealready gates onIsComplete(line 286); the async path lost that gate when the hash thread was introduced. Without it, a write that lands inside a part marks the part dirty and queuesHashSingleParteven when the part still has gaps;HashSinglePartthen walks past the highest written offset and tripsPS_ERROR.2. Defensive bound-check in
HashSinglePart— pre-read, verifypartfileLen >= offset + length. On mismatch, log +AddGap(offset, offset+length-1)+ return false. The caller treats the part as corrupt and the gap-list opens those bytes for re-fetch, so recovery happens transparently on the next download attempt.Together these cover the scenarios where gap-list and on-disk state can legitimately diverge — transient write failure followed by
.metpersistence and unclean shutdown, external truncation, partial backup restore, FS corruption — all of which previously required manual intervention.Test
End-to-end live download of a 10 GB file from a LAN seeder, on macOS amuled built from this branch:
7686fedd790b58955633d46df8e56523).PS_ERROR, noEOF while hashingexceptions.