Skip to content

Latest commit

 

History

History
185 lines (122 loc) · 3.85 KB

README.md

File metadata and controls

185 lines (122 loc) · 3.85 KB
Raw

Currently, pushing the code. SQLDebug will be completed soon.

Overview

alt text

Prerequisite

Compile and Install DynamoRIO

  • Clone DynamoRIO repository
$ git clone https://github.com/DynamoRIO/dynamorio.git
  • Patch file for compiling tracing library
# copy library
$ cp dynamorio_patch/*.c dynamorio/api/samples/

# patch dynamorio/api/samples/CMakeLists.txt: insert one line
add_sample_client(cbrtrace    "cbrtrace.c;utils.c"    "drmgr;drx")
+ add_sample_client(cbrtrace2    "cbrtrace2.c;utils.c"    "drmgr;drx")
  • Compile the dynamorio
# compile
$ cd dynamorio
$ mkdir build
$ cd build
$ cmake ..
$ make -j 4

# check compiled library (libcbrtrace.so and libcbrtrace2.so)
$ ls -al api/bin/*.so
  • Example command to capture the trace
$ dynamorio/build/bin64/drrun -c dynamorio/build/api/bin/libcbrtrace.so -- YOUR_SQLITE YOUR_DB.db <  YOUR_QUERY.sql

Debugging SQLITE3

Setup fossil Sqlite for bisect

1) Clone fossil repo

mkdir sqlite_fossil
cd sqlite_fossil
fossil clone https://www.sqlite.org/src sqlite.fossil
fossil open sqlite.fossil
  • If we do regression fuzzing on sqlite 3.23.0 and sqlite 3.27.0, we should find commit ID (you can use the following commands to find out)
    • sqlite 3.23.0: 736b53f57f
    • sqlite 3.27.0: c9af2f71bf

2) Some useful commands (if you need a manual analysis)

; dump fossil log
$ fossil timeline -n 30000 > sqlite_log
$ cat sqlite_log|grep -v "\=\=\=" > sqlite_log2
$ cat sqlite_log2|grep "\:" |grep "\[" |grep "\]" > sqlite_log3

$ python
import sys
import fileinput
for line in fileinput.input('sqlite_log3', inplace=True):
    if line.startswith(' '):
        continue
    if "ticket [" in line.lower() or 'edit [' in line.lower():
        continue
    else:
        sys.stdout.write(line)

; working with tag
$ fossil tag list
$ fossil update tag:version-3.23.0

; diff commits
$ fossil diff --from f08c1731b0 --to 736b53f57f

; go to specific commit
$ fossil checkout 27e20d6998
$ fossil checkout 736b53f57f

Bisect problematic commit

./bisect.py -db sqlite -i sq_ex -r sqlite_fossil -l bisect_log

  • -db or --database: Required, target database (e.g., sqlite or postgres)

  • -i or --indir: Required, input directory where regression-inducing SQL queries are located

  • -r or --repodir: Required, repository directory (either git or fossil)

  • -l or --logfile: Optional, log filename

  • Important: You SHOULD know which commit is related to the performence regression bug. After identify the commit number, you should compile two DBMS with debug symbol.

    • #1: DBMS which initially introduces performance regression (e.g., SQLite_bad)
    • #2: DBMS which lastly showed good performance (e.g., SQLite_good)

Extract execution trace thru DBI

  • Capture the execution trace using
# manually capture one sql file
$ dynamorio/build/bin64/drrun -c dynamorio/build/api/bin/libcbrtrace.so -- YOUR_SQLITE YOUR_DB.db <  YOUR_QUERY.sql
  • Using provided script for user's convenience
# to be updated

Statistical debugging

Please use the demo script and see how it works

# run and get the important predicates
$ cd sq_demo
$ python statdbg_sqlite.py sqlite3

# NOTE: we didn't provide the original binary yet so displayed report does not include source code information

Debugging PostgreSQL

Setup test repo

To be updated

Bisect problematic commit

To be updated

Dynamic Binary Instrumentation

To be updated

Statistical debugging

To be updated

However, we added demo to show how it works.

$ cd {REPO}/src/sqldebug/demo
python statdbg.py trace 898

Experimental

Delta LDA algorithm

To be updated

Utility commands

To be updated