| Bug #48284 | C99 aliasing violation, results in failures in the client library | ||
|---|---|---|---|
| Submitted: | 25 Oct 2009 6:57 | Modified: | 12 Mar 2010 17:13 |
| Reporter: | Davin McCall | Email Updates: | |
| Status: | Closed | Impact on me: | |
| Category: | MySQL Server: C API (client library) | Severity: | S2 (Serious) |
| Version: | 5.1.40 | OS: | Linux |
| Assigned to: | Jim Winstead | CPU Architecture: | Any |
| Tags: | aliasing | ||
[17 Nov 2009 10:33]
Tonci Grgin
Hi Davin. Verified just as described by looking into latest 5.1 sources.
[23 Nov 2009 22:02]
Bugs System
A patch for this bug has been committed. After review, it may be pushed to the relevant source trees for release in the next version. You can access the patch from: http://lists.mysql.com/commits/91356 3218 Jim Winstead 2009-11-23 Fix C99 aliasing violation due to mismatched types that were papered over with a cast. (Bug #48284)
[15 Dec 2009 9:43]
Magnus Blåudd
Patch is ok. It's effect is visual since it fixes a crash in mysql_client_test when release compiling with gcc 4.4.2
[17 Dec 2009 19:59]
Jim Winstead
Queued to the mysql-5.1-bugteam and mysql-pe trees.
[15 Jan 2010 8:58]
Bugs System
Pushed into 5.1.43 (revid:joro@sun.com-20100115085139-qkh0i0fpohd9u9p5) (version source revid:jimw@mysql.com-20091216204853-5clvhmihej25795n) (merge vers: 5.1.42) (pib:16)
[22 Jan 2010 1:12]
Paul DuBois
Noted in 5.1.43 changelog. An aliasing violation in the C API could lead to a crash. Setting report to NDI pending push to 5.5.x+.
[5 Feb 2010 11:50]
Bugs System
Pushed into mysql-next-mr (revid:alik@sun.com-20100204063540-9czpdmpixi3iw2yb) (version source revid:alik@sun.com-20091224075613-es9uswo4lidkm3tj) (pib:16)
[12 Mar 2010 14:20]
Bugs System
Pushed into 5.1.44-ndb-7.0.14 (revid:jonas@mysql.com-20100312135944-t0z8s1da2orvl66x) (version source revid:jonas@mysql.com-20100312115609-woou0te4a6s4ae9y) (merge vers: 5.1.44-ndb-7.0.14) (pib:16)
[12 Mar 2010 14:35]
Bugs System
Pushed into 5.1.44-ndb-6.2.19 (revid:jonas@mysql.com-20100312134846-tuqhd9w3tv4xgl3d) (version source revid:jonas@mysql.com-20100312060623-mx6407w2vx76h3by) (merge vers: 5.1.44-ndb-6.2.19) (pib:16)
[12 Mar 2010 14:51]
Bugs System
Pushed into 5.1.44-ndb-6.3.33 (revid:jonas@mysql.com-20100312135724-xcw8vw2lu3mijrhn) (version source revid:jonas@mysql.com-20100312103652-snkltsd197l7q2yg) (merge vers: 5.1.44-ndb-6.3.33) (pib:16)
[12 Mar 2010 17:13]
Paul DuBois
Noted in 5.5.3 changelog.
Description: There is a C99 aliasing violation in the source of Mysql 5.1.40 which causes client errors (specifically with MythTV but probably also with others) when Mysql is compiled with -O3 optimization (GCC 4.4.2). The problem is in in libmysql/libmysql.c, in the cli_stmt_execute() function. Line 2553 looks like: for (param= stmt->params; param < param_end; param++) store_param_type((char**) &net->write_pos, param); And store_param_type is declared as: static void store_param_type(char **pos, MYSQL_BIND *param) { uint typecode= param->buffer_type | (param->is_unsigned ? 32768 : 0); int2store(*pos, typecode); *pos+= 2; } However, the real type of net->write_pos is "uchar *" not "char *" and it is referenced as such elsewhere in the cli_stmt_execute function. The types are incompatible and it is not legal to access the net->write_pos value as if it were both a "uchar *" and "char *" (note that if the "char **" cast is removed GCC gives a warning that the types are incompatible). The eventual result is that GCC thinks "*pos+=2" can be moved outside the loop and collapsed to a single store. A bit silly, but that's what happens when you break the aliasing rules :) How to repeat: Compile with GCC 4.4.2, using the Mysql-recommended CFLAGS/CXXFLAGS (I added only -march=i686). CXX=gcc CFLAGS="-O3 -march=i686" \ CXXFLAGS="-O3 -felide-constructors -fno-exceptions -fno-rtti -march=i686" \ ./configure --prefix=/usr --localstatedir=/var/mysql --sysconfdir=/etc \ --mandir=/usr/share/man \ --enable-thread-safe-client --enable-assembler --with-gnu-ld \ --with-mysqld-user=mysql --with-ssl make make install Then run "mythfrontend"... Suggested fix: The fix is fairly trivial, two things should be changed: 1) change the declared type of the "pos" parameter in store_param_type from "char **" to "uchar **" 2) remove the "char **" cast from line 2554 in cli_stmt_execute. (Not strictly necessary, but GCC gives a warning if you don't). For a workaround with current version, it's easy just to add "-fno-strict-aliasing" to CFLAGS.