diff options
author | Binh-Minh Ribler <bmribler@hdfgroup.org> | 2020-05-26 20:39:39 (GMT) |
---|---|---|
committer | Binh-Minh Ribler <bmribler@hdfgroup.org> | 2020-05-26 20:39:39 (GMT) |
commit | 7a4d7da14a745333c4e8f3955a0afd9db70969c0 (patch) | |
tree | fb09ecf511f936cfa5f6d67d94bafd98bf7d3bee /fortran/src/H5Gff.F90 | |
parent | 32f8fed120df107b46ba73f60429f14f71b30012 (diff) | |
download | hdf5-7a4d7da14a745333c4e8f3955a0afd9db70969c0.zip hdf5-7a4d7da14a745333c4e8f3955a0afd9db70969c0.tar.gz hdf5-7a4d7da14a745333c4e8f3955a0afd9db70969c0.tar.bz2 |
Fix HDFFV-11053
Description:
A superblock pointer was set to a temporary structure, as a kluge,
and meant to be reset back to NULL after the processing was completed.
However, in the case of failure before the completion, this setting
caused the superblock pointer to be unpinned twice in the error
recovery process. As a result, that generated a segfault in h5clear.
The fix resets the superblock pointer to NULL after H5F__super_ext_remove_msg
fails so that h5clear can fail properly.
Note: After the fix, when built in debug mode, there will be an
assertion failure with the user's file, which appeared to be corrupted:
H5MM_final_sanity_check: Assertion `0 == H5MM_curr_alloc_bytes_s' failed.
This did not happen on a good file or on the corrupted file with the
library built in production mode. The un-freed memory were allocated
during error recovery.
Platforms tested:
Linux/64 (jelly)
Diffstat (limited to 'fortran/src/H5Gff.F90')
0 files changed, 0 insertions, 0 deletions