xfsdump, xfsprogs distro builds and DEBUG=
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest5 patches 3 repliesIn review
Patches in v1 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/5 | xfsrestore: remove unused variable strctxprestore/ | +0 | 0 replies | |
| 2/5 | annotate variables only used for assertcommon/ common/ common/ +10 more | +50 | 0 replies | |
| 3/5 | xfsrestore: include TREE_DEBUG all buildsrestore/ restore/ | +21 | 0 replies | |
| 4/5 | xfsrestore: remove failing assert from noref_elim_recurserestore/ | +2 | 1 reply | |
| 5/5 | xfsrestore: assert suppression workaroundinclude/ man/ restore/ +4 more | +75 | 0 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.
Cover letter [PATCH 0/5] xfsdump, xfsprogs distro builds and DEBUG=
tldr; Fedora builds with -DDEBUG, is that what folk want or expect?
We see cases from the field where xfsrestore aborts on an assert(3). In each case there's a real problem that needs to be fixed, but these problems are preventing folk from restoring their data. With the asserts suppressed or removed xfsrestore usually does something sane enough, at worst leaving restored content in the orphanage where a user can find it.
This was notably seen with the original xfsdump bind mount fix, now we're seeing asserts fire during cumulative restores in noref_elim_recurse.
xfsrestore: tree.c:1421: noref_elim_recurse: Assertion 'isrealpr' failed
This appears to be caused by a rename to within the tree between restore levels, or a combination of modifications during occurring during the dump. Fixing it will likely require changes in noref_elim_recurse, or tree_post, to ensure elements of the tree are created for these edge cases.
But why are we seeing assert(3) active in Fedora based distro builds?
xfsprogs and xfsdump m4/package_globals.m4 shows builds should default to DEBUG and xfsdump doc/INSTALL talks about how to turn off asserts and create an optimized build, but only Debian has its build rules in-tree, where they set DEBUG=-DNDEBUG.
Building locally on Fedora or similar gets varied results with recent Fedora versions building as if DEBUG=-DNDEBUG was set within the realm of configure or autoconf, which may be misleading people.
But for Fedora release builds we can see that they're building with DEBUG=-DDEBUG resulting in gcc called with -DDEBUG in the build logs.
https://src.fedoraproject.org/rpms/xfsdump
https://kojipkgs.fedoraproject.org//packages/xfsdump/3.2.0/4.fc44/data/logs/x86_64/build.log
https://src.fedoraproject.org/rpms/xfsprogs
https://kojipkgs.fedoraproject.org//packages/xfsprogs/6.18.0/2.fc44/data/logs/x86_64/build.log
I don't think the intention was to have these build as DEBUG, but that seems to be the current result. But before I dig into this further can I confirm what people think this should do?
The following series is illustrative of the changes to build xfsdump cleanly with DEBUG=-DNDEBUG, or the type of downstream workarounds required to help end users hitting problems for the status quo.
Don
Donald Douwsma (5):
xfsrestore: remove unused variable strctxp annotate variables only used for assert
xfsrestore: include TREE_DEBUG all builds
xfsrestore: remove failing assert from noref_elim_recurse
xfsrestore: assert suppression workaround
common/drive_minrmt.c | 5 +-- common/drive_scsitape.c | 5 +-- common/drive_simple.c | 9 ++--- common/main.c | 6 +-- common/mlog.c | 6 +-- common/qlock.c | 18 ++++----- common/ring.c | 2 +- dump/content.c | 8 ++-- include/config.h.in | 4 ++ man/man8/xfsrestore.8 | 4 ++ restore/Makefile | 6 ++- restore/content.c | 36 +++++++----------- restore/dirattr.c | 3 +- restore/getopt.h | 4 +- restore/inomap.c | 9 ++--- restore/restore_assert.c | 39 +++++++++++++++++++ restore/restore_assert.h | 18 +++++++++ restore/tree.c | 82 ++++++++++++++-------------------------- restore/win.c | 24 ++++-------- 19 files changed, 148 insertions(+), 140 deletions(-) create mode 100644 restore/restore_assert.c create mode 100644 restore/restore_assert.h
Signature · 1 line
2.47.3