Fix for segmentation fault when updating a record having a small
BLOB whose size didn't change. Fix for probably impossible problem in Recovery. mysql-test/r/maria.result: result for new test mysql-test/t/maria.test: testcase for a bug (used to segfault) storage/maria/ma_blockrec.c: When writing a record, we put BLOBs into the head part if there is room for them. "Is there room" was first decided by !(tmp_data + length > end_of_data) (line 1894) but then was tested again as *blob_lengths < (ulong)(end_of_data - data). We see that in case of equality, the first condition was true but the second was not, so it was inconsistent and crashed later. storage/maria/ma_recovery.c: When wondering if recovery should update the state (like state.records): if table was closed, its is_of_horizon was set to X, then table was reopened and a REDO was written. If this REDO had LSN X (as horizon is just a lower bound of the LSN of the next record), we have to apply it. In practice this equality probably could not happen because of LOGREC_FILE_ID would be written before the REDO.
Showing
Please register or sign in to comment