Properly clean up cache when failing to load an object header #4477
Conversation
This file contains 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
fortnern
requested review from
lrknox,
derobins,
byrnHDF,
jhendersonHDF,
qkoziol,
vchoi-hdfgroup,
bmribler,
glennsong09,
mattjala and
brtnfld
as code owners
hyoklee
approved these changes
May 11, 2024
derobins
added
Merge - To 1.14
Priority - 1. High 🔼
|
H5HL_protect() is similar (w/two cache components), do you think it needs a similar change? |
qkoziol
approved these changes
May 14, 2024
From a quick reading of the HL code, it looks like, while it doesn't remove the prefix from cache in case of datablock decoding failure, it should handle such a failure gracefully and leave the heap struct in a consistent state. It might still be a good idea to remove the prefix from cache. |
jhendersonHDF
approved these changes
May 14, 2024
byrnHDF
pushed a commit
to byrnHDF/hdf5
that referenced
this pull request
May 14, 2024
…up#4477) * Properly clean up cache when failing to load an object header * Don't check message type a second time in H5G__open_oid if the first attempt returns error * Add more asserts to H5O__assert() to avoid segfaults --------- Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
qkoziol
pushed a commit
to qkoziol/hdf5
that referenced
this pull request
May 14, 2024
…up#4477) * Properly clean up cache when failing to load an object header * Don't check message type a second time in H5G__open_oid if the first attempt returns error * Add more asserts to H5O__assert() to avoid segfaults --------- Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
lrknox
pushed a commit
to lrknox/hdf5
that referenced
this pull request
May 20, 2024
…up#4477) * Properly clean up cache when failing to load an object header * Don't check message type a second time in H5G__open_oid if the first attempt returns error * Add more asserts to H5O__assert() to avoid segfaults --------- Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
lrknox
added a commit
that referenced
this pull request
May 23, 2024
* win32defs: Fix Wundef warning (#4467) * Refactor error handling code to eliminate internal ID calls (#4453) All calls to the H5I routines are now made in API routines (sometimes in FUNC_ENTER/LEAVE_* macros), except for some calls to H5E_clear_stack() within the library, but I'm planning to remove those over time. Also, made all the library internal error messages into static const variables, instead of malloc'ing them, which means that they can just be referenced and not copied. Several new and updated auto-generated header files were necessary to enable this. * CMake: Fix mingw/fortran build (#4466) * Update for blosc2 in plugins and prefix hdf5 cmake varnames (#4468) * Fix an issue where compound datatype member IDs can be leaked during conversion (#4459) Also fixes issues with handling of partially initialized datatypes during library shutdown * H5Group: Fix operator= (#4473) Closes #4472 * Fix github issue #2523: doxygen -- fix grammatically incorrect sentence alias (#4474) * Remove env step not used by CI in testing (#4476) * Add H5fortkit dependecy for H5Rff.F90 (#4482) * Properly clean up cache when failing to load an object header (#4477) * Properly clean up cache when failing to load an object header * Don't check message type a second time in H5G__open_oid if the first attempt returns error * Add more asserts to H5O__assert() to avoid segfaults --------- Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com> * Add a missing image from the original document (#4490) * Disable EOF checks for SWMR readers in more cases. (#4489) Fixes a race condition where the reader opens the file and sets its EOF from the file's size (from the stat() call in the driver open callback). Then, before the reader can read the file's superblock, a SWMR writer races in, extends the file, and closes the file, writing an updated superblock with the 'writer' and 'SWMR writer' flags in the superblock off (appropriately). Then the reader proceeds to read the superblock, and flags the EOF as wrong. Taking out the check for the 'writer' and 'SWMR writer' flags will cause SWMR readers to avoid flagging the file as incorrect. * Remove unnecessary fortran install (#4498) * Only one version of binaries is produced for platforms (#4496) * Fix for github issue #2220. (#4497) Document the limitation in the Passthrough Conncector section of the VOL Connector Author Guide. The limitation is posted by Neil in the github issue on Dec 22, 2022. * Release asset tarballs with no version filenames (#4494) * Improve spec. reading superblock into cache (a little) by using v2 size (#4491) * Improve spec. reading superblock into cache (a little) by using v2 size Instead of reading the absolute minimal possible, use the likely value of a v2+ superblock w/8-byte addresses & lengths. * Fix for github Issue #1388 can't delete renamed dense attribute with corder tracking enabled (#4462) * Fix for github issue #1388: can't delete renamed dense attribute with corder tracking enabled The problem occurs in step 3(b) below which will delete the attribute with corder x from the creation order index v2 B-tree. The rename sequence in H5A__dense_rename() occurs in the following order: 1) The old attribute with corder x was removed from the creation order index v2 B-tree 2) The new renamed attribute was inserted via H5A__dense_insert(): (a) insert the attribute with new name j into the name index v2 B-tree (b) insert the attribute with corder x into the creation order index v2 B-tree 3) The old attribute was removed via H5A__dense_remove(): (a) remove the attribute with old name k from the name index v2 B-tree (b) remove the attribute with coorder x from the creation order index v2 B-tree Fix: deactivate the "corder_bt2_addr" field so that H5A__dense_remove() won't delete the attribute with corder x from the creation order index v2 B-tree. * Fix/revert a libtool sed hack (#4501) * Revert "Remove Autotools sed hack (#3848)" This reverts commit 8b3ffde. * Fix libtool sed cleanup on MacOS Convert sed -i line to sed > libtool.bak && mv libtool.bak libtool to avoid non-portable -i option. * Update src/H5public.h * Set H5 specific vars immediately if legacy find (#4512) * Correct find process vars (vs in-line build) * Correct SZIP find * Everything is libaec 1.0.6 or newer * Correct option help text
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Addresses #4433
Since an object header is split into multiple cache objects but treated as a single object in the object header code, it is possible, if the cache load of the object header fails, for some of the object header objects to remain in cache pointing to an incomplete/inconsistent H5O_t struct. Modified H5O_protect to expunge all chunks and delete the main object header chunk if it fails for any reason. Also modified H5G__open_oid() to avoid the second call to H5O_msg_exists() if the first returns an error.