util: Address various issues in AllocateFileRange #33228

pull luke-jr wants to merge 5 commits into bitcoin:master from luke-jr:fix_preallocate changing 1 files +23 −8

luke-jr commented at 3:35 PM on August 20, 2025: member

Fixes several bugs in AllocateFileRange.

Not a real fix for #33128, but a workaround for the issue.

Potentially might fix #28552 too, but I don't have a way to reproduce that. (After digging into Apple's XNU code, it looks too buggy to try to use F_PREALLOCATE at all IMO; for reference: XNU kernel code and HFS driver code - notice the truncation risk from comparing to block-allocated size rather than file size and reachable code paths the comments assume to be dead code)

Bugfix: AllocateFileRange: Disable buggy macOS-specific implementation

Several bugs exist:
- F_PREALLOCATE is undone at close (which we do immediately)
- HFS implementation appears buggy (and we have unknown issues on macOS with exFAT - possibly related?)
- F_PEOFPOSMODE assumes offset is EOF exactly

cab5e8d861

Bugfix: AllocateFileRange: Only posix_fallocate the intended range 4a9f8c78f6
Bugfix: AllocateFileRange: Don't SetEndOfFile if SetFilePointerEx fails 8e9c114fae

AllocateFileRange: Avoid clobbering data in fallback implementation

While this data should never be live, in practice it can be right now, so better play it safe

476861713e

Bugfix: AllocateFileRange: Avoid truncating files on Windows 2c113ebc2d
DrahtBot commented at 3:35 PM on August 20, 2025: contributor



The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.



Code Coverage & Benchmarks

For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/33228.



Reviews

See the guideline for information on the review process. A summary of reviews will appear here.



Conflicts

No conflicts as of last run.
fanquake renamed this:
~~Bugfix: AllocateFileRange: Address various issues~~
util: Address various issues in AllocateFileRange
on Nov 10, 2025

in src/util/fs_helpers.cpp:216 in 4a9f8c78f6 outdated

 212 | @@ -213,8 +213,7 @@ void AllocateFileRange(FILE* file, unsigned int offset, unsigned int length)
 213 |  #else
 214 |  #if defined(HAVE_POSIX_FALLOCATE)
 215 |      // Version using posix_fallocate
 216 | -    off_t nEndPos = (off_t)offset + length;
 217 | -    if (0 == posix_fallocate(fileno(file), 0, nEndPos)) return;
 218 | +    if (0 == posix_fallocate(fileno(file), offset, length)) return;

sedited commented at 12:53 PM on December 15, 2025:

Is this really a bugfix? Might this not be intentional to allocate as large a chunk of the file as possible, even in the case where a non-contiguous offset is passed in?

in src/util/fs_helpers.cpp:230 in 476861713e outdated

 226 | +    clearerr(file);
 227 |      while (length > 0) {
 228 |          unsigned int now = 65536;
 229 |          if (length < now)
 230 |              now = length;
 231 | +        const size_t rlen = fread(buf, 1, now, file);

sedited commented at 1:14 PM on December 15, 2025:

Why not seek to the end of the file instead of reading?

fanquake commented at 1:18 PM on March 3, 2026: member

@luke-jr are you planning on responding to comments here?
fanquake commented at 1:28 PM on June 1, 2026: member

@luke-jr can you respond to the review?

Contributors

Linked (view graph)

#28552 Crashes constantly on Mac OS Sonoma when using an ExFAT external SSD #33128 *BSD, OmniOS: `-reindex` is broken

util: Address various issues in AllocateFileRange #33228

Code Coverage & Benchmarks

Reviews

Conflicts