the "tag_list" to find out the "corked" status of an object.
The fix:
(1) Add a counter "num_objs_corked" in the cache structure to track the number
of "corked" objects.
(2) Skip the search of "tag_list" if the counter is zero i.e. no "corked" objects.
* commit '055208b71c573be3d4b1047b840d4d8b255bc25e':
Minor tweak to CMake preadwrite entry.
Updated sec2, log, and core VFDs to use pread/pwrite when available (can be controlled via a configure/CMake option)
* commit '5f1f231310bcb3278f11a51e1ea47782f6ee3235':
Updated H5Dio.c comment.
Fixed a valgrind issue with the unprotect log message. The logging call was placed after the point where the cache entry was freed. The fix was to just use the entry's address directly in the log call instead of the entry itself.
Fixed a bug revealed by the dtransform test. A datatype transform buffer needed to be calloc'd.
Add script raybsub to handle submitting .lsf files on ray with bsub <
script.lsf syntax that we couldn't handle with CMake.
Add hpc-cmake-tgz option for bin/release.
call was placed after the point where the cache entry was freed.
The fix was to just use the entry's address directly in the log
call instead of the entry itself.
Description:
The test for HDFFV-10588 has revealed a bug in H5Ewalk.
H5Ewalk did not stop midway even when the call back function returns
H5_ITER_STOP. This is because a condition is missing from the for
loops in H5E__walk causing the callback functions unable to stop until
all the errors in the stack are iterated. Quincey advised on the final
fix. In this fix, "status" is switched to "ret_value" and HGOTO_ERROR
to HERROR, and the for loops won't continue when "ret_value" is not 0.
Platforms tested:
Linux/64 (jelly)
Linux/64 (platypus)
Darwin (osx1010test)
Description:
HDFFV-10586 CVE-2018-17434 Divide by zero inh5repack_filters
Added a check for zero value
HDFFV-10588 CVE-2018-17437 Memory leak in H5O_dtype_decode_helper
This is actually an Invalid read issue. It was found that the
attribute name length in an attribute message was corrupted,
which caused the buffer pointer to be advanced too far and later
caused an invalid read.
Added a check to detect attribute name and its length mismatch. The
fix is not perfect, but it'll reduce the chance of this issue when a
name length is corrupted or the attribute name is corrupted.
Platforms tested:
Linux/64 (jelly)
Linux/64 (platypus)
Darwin (osx1010test)
* commit '90d13bef33f9e2e80b23996a0c39f16f7c34ecf8':
Fixed typo Platforms tested: Darwin (osx1010test)
Refixed HDFFV-10578 Description: Applied Neil's fix for this issue after removing previous attempt. The resources are now released in init_objs() when failure occurs there. Neil will fix HDFFV-10676 separately. Platforms tested: Linux/64 (jelly) Linux/64 (platypus) Darwin (osx1010test)
Removed previous change in H5O__chunk_deserialize().
Removed the previous change in H5O__chunk_deserialize()
Removed previous change in table_list_add().
Removed the previous change in table_list_add()
Updated per review Description: HDFFV-10676 - CVE-2018-13873 Changed the new assert to if statement, per Dana's comment. Platforms tested: Linux/64 (jelly)
HDFFV-10578 and HDFFV-10676 Description: HDFFV-10578 - CVE-2018-17234 The file has some issue, however, there was a bug in h5dump that caused memory leaks after the problem in the file was encountered. The bug was that an if statement was missing in the function table_list_add() resulting in the memory not being freed at a later time. After the fix had been applied, there were no more leaks after h5dump detected the issue in the file and reported the error.
* commit 'e27df5a5fe365480b0809ce523b0e9cd1096a43d':
Add variables to set up module switch between configure and build for cross compiling on CrayXC40.
Remove install-exec-hook from hl/fortran/Makefile.am. Add check for existing links in hl/fortran/src/Makefile.am.
Configure build directory in batch scripts.
Add libhdf5_hl_fortran symlinks for libhdf5hl_fortran files. Set ctest*.lsf.in.cmake scripts for sierra. Correct src/CMakelists.txt syntax error.
Correct syntax error in CMakeLists.txt.
Add .lsf batch scripts and insert the build directory into the scripts during configuration.
Description:
Applied Neil's fix for this issue after removing previous
attempt. The resources are now released in init_objs() when
failure occurs there.
Neil will fix HDFFV-10676 separately.
Platforms tested:
Linux/64 (jelly)
Linux/64 (platypus)
Darwin (osx1010test)
Avoid potential invalid read when decoding unknown object header message.
* commit '685288beabc4f4383d69ec07be05c067fab78544':
Delay checking if decoded message's "shareable" flag is appropriate for the message type until we've verified we understand the message type. Reduce size of H5O_msg_class_g to *not* include space for H5O_BOGUS_INVALID. Make bogus messages shareable. Add new bogus message test with shareable messages to cover the formerly problematic code. Re-run gen_bogus.c to add this test case and also to fix the bogus_invalid messages that were no longer H5O_BOGUS_INVLAID due to a new message class being added in a previous commit. Added comment to remind developers to run gen_bogus.c when adding a new message class.
the message type until we've verified we understand the message type.
Reduce size of H5O_msg_class_g to *not* include space for
H5O_BOGUS_INVALID. Make bogus messages shareable. Add new bogus
message test with shareable messages to cover the formerly problematic
code. Re-run gen_bogus.c to add this test case and also to fix the
bogus_invalid messages that were no longer H5O_BOGUS_INVLAID due to a
new message class being added in a previous commit. Added comment to
remind developers to run gen_bogus.c when adding a new message class.
* commit '7bf6d847ae36940f1b1147820356795c7764dbeb':
Remove incorrect comment.
Updated error stacks for h5dump testing.
Corrected comment in src/H5VLint.c, fixed pass-through info size in src/H5VLpassthru.c, switched to stashing VOL connector ID & info in API context (in src/H5CX.c, src/H5CXprivate.h, src/H5F.c, src/H5Fint.c, and src/H5Fefc.c), patched up all sorts of issues in the tests, to make them work with 'check-vfd' (and 'check-vol' again).
Clean up style issues, make function callbacks static, correct debug statements, etc. No functionality changes at all.
* commit 'f0656a57dc3baf02709e1ad5e496cf63cf820877':
Remove incorrect comment.
Updated error stacks for h5dump testing.
Corrected comment in src/H5VLint.c, fixed pass-through info size in src/H5VLpassthru.c, switched to stashing VOL connector ID & info in API context (in src/H5CX.c, src/H5CXprivate.h, src/H5F.c, src/H5Fint.c, and src/H5Fefc.c), patched up all sorts of issues in the tests, to make them work with 'check-vfd' (and 'check-vol' again).
Description:
HDFFV-10578 - CVE-2018-17234
The file has some issue, however, there was a bug in h5dump that caused
memory leaks after the problem in the file was encountered. The bug
was that an if statement was missing in the function table_list_add()
resulting in the memory not being freed at a later time.
After the fix had been applied, there were no more leaks after h5dump
detected the issue in the file and reported the error.
In H5O__chunk_deserialize, replaced an assert with an if statement
and reporting error, per Neil's recommendation
HDFFV-10676 - CVE-2018-13873
Also in H5O__chunk_deserialize, added an assertion to detect
out of bound ids
src/H5VLpassthru.c, switched to stashing VOL connector ID & info in
API context (in src/H5CX.c, src/H5CXprivate.h, src/H5F.c, src/H5Fint.c, and
src/H5Fefc.c), patched up all sorts of issues in the tests, to make them work
with 'check-vfd' (and 'check-vol' again).
* commit '0e34f0feaaeb6d2286f87f695f25fae45c070a42': (26 commits)
HDFFV-10664 update reference file
HDFFV-10664 update reference files
HDFFV-10664 add missing function and check for restriction
Remove "bad" statements at end of TEST_ERROR. Formerly arguments to FAIL_PUTS_ERROR
Add RELEASE.txt entry for HDFFV-10596.
Add semicolon at end of TEST_ERROR() invocation for consistency.
Uninstall should remove the libhdf5_hl_fortran links.
Add symlinks named libhdf5_hl_fortran* to libhdf5hl_fortran in Makefile.am to match cmake install and the name pattern for other hl lib files.
Change FAIL_PUTS_ERROR to TEST_ERROR.
Cleaned up the parallel flush test and set t_pflush1 to always fail. Also set CMake to handle this.
Update calls: H5Dcreate to H5Dcreate2, H5Acreate to H5Acreate2.
Removed a comment.
Fix comment
Align H5Lcreate_ud behavior with documentation for NULL udata pointer
Update toolchain comment
Use c99 standard cmake variable
Add PGI toolchain
Correction based on code review.
HDFFV-10664 add check for state before set call
HDFFV-10546 refactor variable name
...
