Fix a few Clang sanitizer warnings #1727

jhendersonHDF · 2022-05-05T00:02:22Z

Noticed while doing other testing

jhendersonHDF · 2022-05-05T00:02:50Z

src/H5Tconv.c

-                    b += b_stride;
+
+                    if (b)
+                        b += b_stride;


Don't do arithmetic on b if b is a NULL pointer.

jhendersonHDF · 2022-05-05T00:03:02Z

src/H5Tconv.c

@@ -3278,7 +3279,8 @@ H5T__conv_vlen(hid_t src_id, hid_t dst_id, H5T_cdata_t *cdata, size_t nelmts, si
                    else {
                        s = (uint8_t *)buf + (nelmts - safe) * (size_t)s_stride;
                        d = (uint8_t *)buf + (nelmts - safe) * (size_t)d_stride;
-                        b = (uint8_t *)bkg + (nelmts - safe) * (size_t)b_stride;
+                        if (bkg)
+                            b = (uint8_t *)bkg + (nelmts - safe) * (size_t)b_stride;


Don't do arithmetic on bkg if bkg is a NULL pointer.

jhendersonHDF · 2022-05-05T00:03:34Z

src/H5VLdyn_ops.c

            H5SL_destroy(H5VL_opt_ops_g[subcls], H5VL__term_opt_operation_cb, NULL);
+            H5VL_opt_ops_g[subcls] = NULL;


Set dynamic operations skip list array table entries to NULL to avoid a use after free during package termination.

jhendersonHDF · 2022-05-05T00:04:08Z

test/dsets.c

+    const hsize_t     shape[2]     = {8, 8};
+    const hsize_t     maxshape[2]  = {H5S_UNLIMITED, H5S_UNLIMITED};
+    const hsize_t     chunk[2]     = {8, 8};
+    const int         buffer[8][8] = {


Test was previously writing to an 8x8 dataset with an 8x1 array, underflowing into the following variables and causing a stack overflow.

jhendersonHDF · 2022-05-05T00:05:28Z

test/dsets.c

@@ -15267,7 +15276,7 @@ test_h5s_plist(void)
    /* Attempt to 'OR' block with invalid dimensions into the selection */
    H5E_BEGIN_TRY
    {
-        ret = H5Pset_dataset_io_hyperslab_selection(dxpl_id_copy, 2, H5S_SELECT_OR, &start, &stride, &count,
+        ret = H5Pset_dataset_io_hyperslab_selection(dxpl_id_copy, H5S_MAX_RANK + 1, H5S_SELECT_OR, &start, &stride, &count,


Specifying 2 for the dimensionality here caused an overflow since start, stride, count, block are single values, not 2-dimensional arrays. Since the test was trying to 'OR' an invalid block, use H5S_MAX_RANK + 1 for the dimensionality, which should fail early.

* Warnings fixes (#1680) * Clean stack size warnings in sio_engine (#1687) * Fixes stack size warnings in tcoords.c (#1688) * Address some warnings from casting away of const (#1684) * Fixes stack size warnings in dtransform (#1696) * Fixes stack size warnings in set_extent test (#1698) * Be a bit safer with signed arithmetic, thus quieting some signed-overflow warnings from GCC (#1706) * Avoid a signed overflow: check the range of `entry_ptr->age` before increasing it instead of increasing it and then checking the range. This quiets a GCC warning. * Avoid the potential for signed overflow by rewriting expressions `MAX(0, fwidth - n)` as `MAX(n, fwidth) - n` for various `n`. This change quiets some GCC warnings. * Change some local variables that cannot take sensible negative values from signed to unsigned. This quiets GCC warnings about potential signed overflow. * In a handful of instances, check the range of a signed integer before increasing/decreasing it, just in case the increase/decrease overflows. This quiets a handful of GCC signed-overflow warnings. * Committing clang-format changes Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com> * Fix object size warnings in cache.c test (#1701) * Fix some const cast and stack/static object size warnings (#1700) * Fix various warnings * Move HDfree_const to H5private.h for wider use * Print output from all ranks in parallel tests on allocation failure * Move const pointer freeing macro to h5test.h for now * Fixes a bug where t_cache fails due to a string size being too small (#1720) * Fixes a bug where t_cache fails due to a string size being too small Recent warning reductions led to an incorrect string size being passed to h5_fileaccess, causing the test to silently fail. In addition to fixing the bug, the test will now fail noisily on setup failures. * Updates the t_cache test to fail noisily on setup errors * Fix a few Clang sanitizer warnings (#1727) * Stop lying about H5S_t const-ness (#1209) Hyperslabs can be reworked inside several H5S callbacks, making H5S_t non-const in some places where it is marked const. This change switches these incorrectly const H5S_t pointer parameters and variables to non-const where appropriate. * Fix a few warnings after recent H5S const-related changes (#1225) * Adjustments for HDF5 1.10 * Hdf5 1 12 Miscellaneous warnings fixes (#1718) * Fixes const issues in the version 2 B-trees (#1172) The operations that were changed are fundamentally not const since the shadow operation can modify the node structure when SWMR is in use. * Quiets const warning in H5RS code (#1181) * Avoid calling H5Ropen_object with a misaligned H5R_ref_t: copy the (#1171) * Avoid calling H5Ropen_object with a misaligned H5R_ref_t: copy the raw H5R_ref_t bytes to a heap buffer that's known to have the right alignment. * Committing clang-format changes * Use an automatic H5R_ref_t instead of malloc'ing one. Go ahead and initialize the H5R_ref_t to all-0s so that arbitrary stack content doesn't foul things up. Bail out with an error if `size` exceeds `sizeof(H5R_ref_t)`. * Committing clang-format changes Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com> * Miscellaneous warnings fixes Co-authored-by: Dana Robinson <43805+derobins@users.noreply.github.com> Co-authored-by: David Young <dyoung@hdfgroup.org> Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com> * Fix several warnings (#747) Co-authored-by: Dana Robinson <43805+derobins@users.noreply.github.com> Co-authored-by: David Young <dyoung@hdfgroup.org> Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>

jhendersonHDF requested review from fortnern, soumagne, vchoi-hdfgroup, ChristopherHogan, gnuoyd, derobins, jrmainzer and qkoziol as code owners May 5, 2022 00:02

jhendersonHDF commented May 5, 2022

View reviewed changes

src/H5Tconv.c

b += b_stride;

if (b)

b += b_stride;

Copy link

Collaborator Author

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Don't do arithmetic on b if b is a NULL pointer.

jhendersonHDF commented May 5, 2022

View reviewed changes

Fix a few Clang sanitizer warnings

79cbfec

jhendersonHDF commented May 5, 2022

View reviewed changes

jhendersonHDF force-pushed the sanitizer_issues branch from df9d09d to 79cbfec Compare May 5, 2022 00:05

byrnHDF approved these changes May 5, 2022

View reviewed changes

derobins approved these changes May 5, 2022

View reviewed changes

derobins merged commit cdf837d into HDFGroup:develop May 5, 2022

jhendersonHDF added a commit to jhendersonHDF/hdf5 that referenced this pull request May 8, 2022

Fix a few Clang sanitizer warnings (HDFGroup#1727)

4baf21e

jhendersonHDF deleted the sanitizer_issues branch February 20, 2023 16:59

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix a few Clang sanitizer warnings #1727

Fix a few Clang sanitizer warnings #1727

jhendersonHDF commented May 5, 2022

jhendersonHDF May 5, 2022

jhendersonHDF May 5, 2022

jhendersonHDF May 5, 2022

jhendersonHDF May 5, 2022

jhendersonHDF May 5, 2022

		H5SL_destroy(H5VL_opt_ops_g[subcls], H5VL__term_opt_operation_cb, NULL);
		H5VL_opt_ops_g[subcls] = NULL;

Fix a few Clang sanitizer warnings #1727

Fix a few Clang sanitizer warnings #1727

Conversation

jhendersonHDF commented May 5, 2022

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment

jhendersonHDF May 5, 2022

Choose a reason for hiding this comment