Bug#339740: major bug in debians mysqld on em64t
Hi, > Hm, hard to track down. But are you sure that your converting is correct and > that you do not produce a buffer overflow somewhere? For example the lines > > case MYSQL_TYPE_LONG: > printf("L:%u(%p), ", *(unsigned long int*)stmt->params[j].buffer, > stmt->params[j].buffer); > > do give warnings with -Wall. Of course this is only the debugging statement... Sorry, change the %u to a %lu and the warning is fixed. Also the earlier debug printf have %d when %u or %lu would be more correct. fmt_xfields_bind() on line 478 sets up the MYSQL_BIND structure setting the pointer and MYSQL_TYPE. Which I have combed through many times. I do have printf %p all over the place to insure that the buffer pointers are correct. If its usefull the url to see what MYSQL_TYPE_ goes with what C type is: http://dev.mysql.com/doc/refman/5.0/en/c-api-prepared-statement-datatypes.html fmt_xfields_bind() sets the MYSQL_TYPES and on line 163 down you can see the C types. > Can you shrink the program down to only one query? You can use the "log = > /var/log/mysql/mysql.log" option or tcpdump/ngrep to verify what the server > receives. I will try to simplify it. I can't garentee my code is bug free but I beleive I have killed most in trying to solve this one, plus there is the fact that mysql's binary server and 32bit run ok. To reduce the number of queries sent change the for() on line 368, eg - for (i=0; i<34235; i++) { + for (i=0; i<1; i++) { that should just send one. But note because 1 is less then the BULK_INSERT_COUNT (=2) it will be the 'insert remainder' starting line 413 that will send the query. you can reduce BULK_INSERT_COUNT to 1 or have a loop that does at least 2. I was planning on editing the test code to do larger BULK_INSERT_COUNT (limit to 2 by the pre filled array on line 216) so that I can look at things that cause mysqld and the client to die. (my app does 100 bulk inserts and ~12 million records, thats 120'000 iterations) I will let you know when I find out more. > > I did originally posted this bug to MySQL, until I found that their binaries > > work fine: http://bugs.mysql.com/bug.php?id=14988 > Which binaries, the libraries you link against or the mysql server? Try > linking statically against mine or their library to see if the bug is in the > prepare statement code or in the server. /usr/local/mysql/ is not in any library path so my app is linking with /usr/lib/libmysqlclient.so.15 from the debian package. infact all the debian packages are still installed. All I did was stop the debian mysqld_safe and start mysql's, so just the running server differs. Thank you for looking at this. Thorben
Bug#339740: major bug in debians mysqld on em64t
Package: mysql-server-5.0 I have a major bug with mysql-server-5.0 (both etch, 5.0.13rc-1 and sid, 5.0.15) on em64t (Quad Xeons) that either kills my app with a various strange mysql errors, seg faults mysqld or mysqld simple exits with no error logged at all =/ None of these problems happen with the Debian packages on a 32bit systems or with MySQL's own community edition 64bit binaries on em64t. So I am lead to believe the problem is just with Debians 64bit packages. This includes source builds (apt-get source ..;cd ...; dpkg-buildpackage -rfakeroot) So maybe its to do with Debian build environment, options, patches, etc I have written a version of flow-export that uses the prepared statement API with bulk inserts to greatly improve its speed. It runs fine except on my em64t box with debians mysql packages. The data that is sent to the server is not what appears in the tables, which appear to be junk memory values. It dies with various error messages, most times after running a few iterations it gives: mysql_stmt_execute(): Incorrect arguments to mysql_stmt_execute even thought the MYSQL_STMT passed to is the same one last iteration. mysqld itself sometimes seg faults I have stripped down my app to just the mysql prep stmt code path to make it stand alone so that it can be used as a test case. the test app does not stop with an error or kill mysqld (if you want that increase the number of iterations done and/or the size of the bulk insert) but the values that appear in the table are not the ones sent. I did originally posted this bug to MySQL, until I found that their binaries work fine: http://bugs.mysql.com/bug.php?id=14988 Also I have a db with a number (~300+) large (~5-10GB) tables which I merge using MyISAM_MERGE tables. When ever I change the UNION of these merge tables using ALTER TABLE it tends to kill mysqld with no error logged mysqld just disappears. This again is true just for debians 64bit packages, others are fine. My installation: Debian Linux amd64 etch (with bits from sid, eg mysql-server-5.0 5.0.15) Quad 64bit Xeon em64t, 16GB RAM Please ask if you need any more information. Thank you Thorben Jändling /* This is a version of the app I am working on that has been stripped down to just the mysql prep stmt code path. Hence the code looks bizare, since function calls and (if) branches have been removed or hard coded. This app works fine on 32bit systems but for 64bit random numbers turn up in the table. To try, compile: gcc -lmysqlclient -I/usr/include/mysql mysql-64bit-test.c -o mysql-64bit-test and run: ./mysql-64bit-test Here is the table schema, please also edit the DB_DEFAULT_* defines below. CREATE TABLE IF NOT EXISTS !TABLENAME! ( ID int(10) unsigned NOT NULL auto_increment, UNIX_SECS int(32) unsigned NOT NULL default '0', UNIX_NSECS int(32) unsigned NOT NULL default '0', DPKTS int(32) unsigned NOT NULL default '0', DOCTETS int(32) unsigned NOT NULL default '0', FIRST int(32) unsigned NOT NULL default '0', LAST int(32) unsigned NOT NULL default '0', SRCADDR int(32) unsigned NOT NULL default '0', DSTADDR int(32) unsigned NOT NULL default '0', NEXTHOP int(32) unsigned NOT NULL default '0', SRCPORT int(16) unsigned NOT NULL default '0', DSTPORT int(16) unsigned NOT NULL default '0', PROT int(8) unsigned NOT NULL default '0', TCP_FLAGS int(8) unsigned NOT NULL default '0', SRC_MASK int(8) unsigned NOT NULL default '0', DST_MASK int(8) unsigned NOT NULL default '0', SRC_AS int(16) unsigned NOT NULL default '0', DST_AS int(16) unsigned NOT NULL default '0', COLLECTOR enum('UNKNOWN','JIM','BOB','JANE','MARY','WHO','STEVE','TOM') NOT NULL default 'UNKNOWN', PRIMARY KEY (ID) ) ENGINE=MyISAM DEFAULT CHARSET=utf8; */ #include #include #include #include #include #define DB_DEFAULT_DBHOST "localhost" #define DB_DEFAULT_DBNAME "fillme" #define DB_DEFAULT_DBPORT "" #define DB_DEFAULT_DBTABLE "editme" #define DB_DEFAULT_DBUSER "tellme" #define DB_DEFAULT_DBPWD "setme" /*the following was folded in from flow-tools */ /* possible fields in export */ #define FT_XFIELD_UNIX_SECS 0x0001LL #define FT_XFIELD_UNIX_NSECS 0x0002LL #define FT_XFIELD_SYSUPTIME 0x0004LL #define FT_XFIELD_EXADDR 0x0008LL #define FT_XFIELD_DFLOWS 0x0010LL #define FT_XFIELD_DPKTS 0x0020LL #define FT_XFIELD_DOCTETS 0x0040LL #define FT_XFIELD_FIRST 0x0080LL #define FT_XFIELD_LAST0x0100LL #define FT_XFIELD_ENGINE_TYPE 0x0200LL #define FT_XFIELD_ENGINE_ID 0x0400LL #define FT_XFIELD_SRCADDR 0x1000LL #define FT_XFIELD_DSTADDR 0x2000LL #define FT_XFIELD_NEXTHOP 0x0001LL #define FT_XFIELD_INPUT 0x0002LL #define FT_XFIELD_OUTPUT 0x0004LL #define FT_XFIELD_SRC
Bug#300214: Wifi map script.
Package: ifupdown Hi, I wrote a wifi map script to map available ap/essid to logical interfaces. I thought it maybe useful to others, and that mailing through bugs is preferable to the maintainer then mailing them directly. eg: mapping eth1 script /get-wifi.sh map 00:11:22:33:44:55 eth1-here map 66:77:88:99:AA:BB eth1-there or with essid map mywifi eth1-mywifi then you can just ifup eth1 ofcourse some use things like whereami, but in some case thats overkill. And i don't like autodetect and almost every thing can be configured from interfaces Reguards Thorben Script: #!/bin/sh # Uncomment which ever method you prefere. ## Match based on ap Mac more secure (can hide essid, avoid twins) wlist=$(iwlist $1 scan | sed -e '/Address/! d' -e 's/[ \t]*Address: \(.*\)/\1/') ## Match based on essid more easier? (ap essid can't be hidden) #wlist=$(iwlist $1 scan | sed -e '/ESSID/! d' -e 's/[ \t]*ESSID:"\(.*\)"/\1/') while read wifi iface; do if echo $wlist | sed -n -e "/$wifi/ q 0" -e "/$wifi/! q 1"; then echo $iface; exit 0; fi done exit 1 -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#302626: Plugins missing (eg xineplug_vo_out_opengl.so)
Package: libxine1 Hi, plugins such as xineplug_vo_out_opengl.so appear to be missing from the binary packge. For me opengl o/p is usefull since xv is not available on ATI Dual head second display, but opengl is. I have not looked in the debian source, but they are in the 1.0.0 source package from sf. I compiled this source and then just copied the .so into the /usr/lib/ dir where the debian package has the other plugin .so and it seems to work ok. (some bugs like opengl plugin don't seem to like switching from fullscreen to window or vice versa) I search for that file on packages.debian.org, package content search, it was not found so i guess its not in debian? Thank you Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#333090: MySQL big table support not compiled in?
Package: mysql-server-5.0 Hi, It seems that the mysql server is not built with the --with-big-tables option. See: http://bugs.mysql.com/bug.php?id=10023 Please could this be changed so that (MERGE) tables with more then 2^32 rows can be supported. This option maybe useful in other versions too. Thank you Thorben
Bug#292634: Missing srp support
Package: ppp Version: 2.4.2+20040428-6 My ISP has switched from chap to EAP. To set EAP up, ppp's srp-entry untility is required. pppd's man page referes to it, but it is missing from the package. I got ppp source and found it to be there, please can you included it in the package. FYI and anyone else googling this... To compile srp-entry you will need to package/build srp from: http://srp.stanford.edu/download.html (use a pre v2, or srp-entry will segfault, i used 1.7.5) and edit SRCDIR/pppd/Makefile to uncomment USE_SRP=y build it. Now you can create entries in /etc/ppp/srp-secrets for EAP auth. Thank you Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#296277: start ifplugd via ifupdown instead of at boot
Package: ifplugd Hi, instead of starting ifplugd from init, couldn't there be an option in /etc/network/interfaces like use-ifplugd then when you ifup ethX (or boot ifupdown runs) the interface is started with ifplug and when you ifdown, the interface and ifplug is stopped. ofcourse this would require ifupdown to know if its being called from ifplugd or not (ie should it start ifplugd or has cable been plugged in and ifplugd is asking for the interface to come up) easiest way (i think) is to add an --ifplug to ifupdown. ifplugd can call ifupdown with it, and it will simply bring the interface up and down. with out it it will also start stop ifplugd (and bring the interface down or up). assuming the use-ifplug is in config else --ifplug is ignored (I thought another way is to see if a ifplugd is already running for given interface, but the only works for bringing and interafce up, ifupdown wont know what to do in down case (bring iface down, and kill or leave ifplugd?)) you find this interdependancy bad. but when i ifdown ethX i want it down and ifplug stopped and not bringing it back up again. Also i would like ifplugd to not start particular interfaces until i ifup it. Thank you Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#398948: i810fb not available at boot-time if compiled as a module
I've been investigating this issue on two notebooks that have i815 chipsets, with Ubuntu Edgy 6.10 (2.6.17-10). The reason i810fb (and most other framebuffers) isn't available at boot-time when compiled as a module is that the framebuffer drivers have to call drivers/video/fbmem.c's fb_get_options() function, but when built as a module the code to make that call and parse the command-line isn't compiled in drivers/video/i810/i810_main.c: #ifndef MODULE static int __devinit i810fb_init(void) { char *option = NULL; if (fb_get_options("i810fb", &option)) return -ENODEV; i810fb_setup(option); return pci_register_driver(&i810fb_driver); } #endif I looked at the source for several other fb drivers and they all have the same structure. So, the answer seems to be, if you need specific framebuffer support at boot-time you'll need to build the kernel with i810, intel_agp, agpgart, drm, i810 all configured to statically link using the kernel build tool: make config TJ.
Bug#398948: i810fb not available at boot-time if compiled as a module
I've been investigating this issue on two notebooks that have i815 chip-sets, with Ubuntu Edgy 6.10 (2.6.17-10). I've documented my research in this Ubuntu bug report: https://bugs.launchpad.net/ubuntu/+source/fbset/+bug/64666 The reason i810fb (and most other framebuffers) isn't available at boot-time when compiled as a module is that the framebuffer drivers have to call drivers/video/fbmem.c's function fb_get_options(), but when built as a module the code to make that call and parse the command-line isn't compiled in drivers/video/i810/i810_main.c: #ifndef MODULE static int __devinit i810fb_init(void) { char *option = NULL; if (fb_get_options("i810fb", &option)) return -ENODEV; i810fb_setup(option); return pci_register_driver(&i810fb_driver); } #endif I looked at the source for several other fb drivers and they all have the same structure. So, the answer seems to be, if you need specific framebuffer support at boot-time you'll need to build the kernel with i810, intel_agp, agpgart, drm, i810 all configured to statically link using the kernel build tool: make config TJ.
Bug#398948: i810fb not available at boot-time if compiled as a module
On Sat, 2007-01-20 at 11:21 -0800, Matt Zimmerman wrote: > The argument handling is done differently for the modular case; elsewhere in > the file you'll find module_param macros for that purpose. > In that case (compiled as a module) the i810fb driver uses module_param(). One of the circumstances that this bug discusses is the driver being loaded as a module with command-line options by the boot-loader. My observations suggest that in the case of a module the boot-time command-line isn't being used, for whatever reason. I'm trying to determine at what point in the boot process the driver module is loaded, and where/how the options are being passed to it. A reference to a kernel source file where that occurs would assist me tremendously. TJ. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#398948: i810fb not available at boot-time if compiled as a module
On Sat, 2007-01-20 at 12:59 -0800, Matt Zimmerman wrote: > The options are passed to it by modprobe, based on command-line parameters > and its configuration files, either in early userspace (initramfs) or from > an init script. > It seems that it doesn't parse the kernel command line as a module. According to the last parapgraph of "Processing kernel command line" at the end of "Building the Linux Kernel Image" of "Kernel 2.4 Internals" (not sure if things have changed mcuh for 2.6, but can't find any references) http://www.faqs.org/docs/kernel_2_4/lki-1.html#ss1.9 "Note that __setup() does nothing for modules, so the code that wishes to process boot commandline and can be either a module or statically linked must invoke its parsing function manually in the module initialisation routine. This also means that it is possible to write code that processes parameters when compiled as a module but not when it is static or vice versa." Looking at the i810fb.c code I don't see where it is doing kernel command-line parsing in its module form. I've just built a custom kernel with i810fb, intel_agp, drm, etc. statically linked and that worked as expected, switching into graphic mode, displaying the Ubuntu usplash screen, and giving me a full-screen 1024x768 tty. TJ. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#408159: LOAD DATA INFILE from a pipe now broken!
Package: mysql-server-5.0 Version: 5.0.30-3 Also version 5.0.32-2 I would say this problem is very critical This has seriously broken one of our systems, and I would be greatful for a URL to previous versions so I can downgrade mysql (they are no long available in the pool archive). Now the problem: When doing LOAD DATA INFILE 'pipe' INTO TABLE table..; as soon as the writing process starts to write, the mysql process just stops with no errors, just: Query OK, 0 rows affected (0.19 sec) Records: 0 Deleted: 0 Skipped: 0 Warnings: 0 I know that many months ago mysql 5.0 (or 4.1) were patched upstream to support unix pipes, however this now appears to have regressed. try: mkfifo pipe cat > pipe in another shell mysql use db; LOAD DATA INFILE '/full/path/to/pipe' INTO TABLE table ... (...); mysql should not stop loading until after cat is closed/killed. If you need any more information please let me know. Thank you Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#408159: LOAD DATA INFILE from a pipe now broken
Hi, LOCAL means that the local client reads the pipe, and then sends the data to the server, as opposed to the server reading the data. Due to the qualities involved this is not that practical. (I would write the myisam table files directly if I know how.) However in my search for previous mysql-server-5.0 packages I downloaded the etch_di_rc1 iso images. These have mysql 5.0.24a-9. I can confirm that LOAD DATA INFILE 'pipe' works for mysql-server-5.0 version 5.0.24a-9. So the bug was introduced some between 5.0.24a-9 and 5.0.30-3. I hope this helps. Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#414872: PDNS recursor fails to start on mips, and bug in package/init scripts
Package: pdns-recursor Version: 3.1.3-2 Hi, The pdns_recursor does not start on mips. running it directly with --daemon=no it outputs:- " Mar 14 10:20:05 Done priming cache with root hints Mar 14 10:20:05 Enabled 'epoll' multiplexer swapcontext in schedule: Function not implemented " return is 1 I'm running Debian Etch on an SGI O2 (IP32) MIPS box. Tried with kernels 2.6.16 to 2.6.19 Also there is a bug in the package remove scripts. They call the pdns recursor init script with stop, which returns an error if no daemon was stopped (eg/ie if it was not running to begin with). The remove scrips then fails and package removal is aborted. All because a daemon that wasn't running anyway could not be stopped. Regards, Thorben -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#429754: Bug
Package: Eterm Version: 0.9.4 When i open eterm, and i use transparancy, then the pseudo transparency is bad positioned. If i move the window then it jumps the correct place. (I hope that you could understand that.) I am using Debian etch unstable, 2.6.18-4 kernel and fluxbox. Jozsef Kelemen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#417910: Bug
Package: opera Version: 9.10-20061 I have debian unstable. I updated with aptitude yesterday, and then i can't start the program. If i try to launch from command line then i get the next message: ERROR: ld.so object 'libjvm.so' from LD_PRELOAD cannot be preloaded: ignored ERROR: ld.so object 'libawt.so' from LD_PRELOAD cannot be preloaded: ignored I reported the bug to Opera too because i didn't know where i should send. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#356853: activated specific schedules
A third problem was the primitive tools available to assembly language programmers. AN ALLE FINANZINVESTOREN! DIESE AKTIE WIRD DURCHSTARTEN! FREITAG 20. APRIL STARTET DIE HAUSSE! REALISIERTER KURSGEWINN VON 400%+ IN 5 TAGEN! Symbol: G7Q.F Company: COUNTY LINE ENERGY 5 Tages Kursziel: 0.95 Schlusskurs: 0.21 WKN: A0J3B0 ISIN: US2224791077 Markt: Frankfurt LASSEN SIE SICH DIESE CHANCE NICHT ENTGEHEN! G7Q WIRD WIE EINE RAKETE DURCHSTARTEN! UNSERE ERWARTUNGEN WIRD G7Q.F UBERTREFFEN! I tried several times to read them, but had to give it up. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#72140: WebsiteOur
Press Release selected http://img444.imageshack.us/my.php?image=fshc6.jpg products address developer needs -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]