Date: Mon, 25 Jun 2012 19:05:07 +0300 From: Alexander Motin <mav@FreeBSD.org> To: Holm Tiffe <holm@freibergnet.de> Cc: freebsd-scsi@freebsd.org, "Kenneth D. Merry" <ken@freebsd.org> Subject: Re: Kernel panic in FreeBSD-8.3 from UFS Message-ID: <4FE88C33.1020008@FreeBSD.org> In-Reply-To: <20120625155316.GA37535@nargothrond.kdm.org> References: <B2FD678A64EAAD45B089B123FDFC3ED72B9F6C1F79@inbmail01.lsi.com> <20120601124338.GU2358@deviant.kiev.zoral.com.ua> <B2FD678A64EAAD45B089B123FDFC3ED72B9F6C1F7F@inbmail01.lsi.com> <20120601125824.GV2358@deviant.kiev.zoral.com.ua> <B2FD678A64EAAD45B089B123FDFC3ED72B9F6C21A7@inbmail01.lsi.com> <20120606170640.GA98428@nargothrond.kdm.org> <20120625135543.GA58915@beast.freibergnet.de> <20120625155316.GA37535@nargothrond.kdm.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 06/25/12 18:53, Kenneth D. Merry wrote: > On Mon, Jun 25, 2012 at 15:55:43 +0200, Holm Tiffe wrote: >> Kenneth D. Merry wrote: >> >>> On Tue, Jun 05, 2012 at 17:49:05 +0530, Desai, Kashyap wrote: >>>> Hi All, >>>> >>>> We found some potential area of memory leak in CAM layer. >>>> CAM XPT Memory leak is due to following function in scsi/scsi_all.c >>>> >>>> int >>>> scsi_command_string(struct ccb_scsiio *csio, struct sbuf *sb) >>>> >>>> >>>> In above function, CAM layer allocate memory for ccb device as below >>>> if ((cgd = (struct ccb_getdev*)xpt_alloc_ccb_nowait()) == NULL) >>>> >>>> >>>> _But_, unfortunately we never free the allocated memory and we see memory leak of 2K every time when someone is calling >>>> Scsi_command_string from kernel mode. >>>> >>>> >>>> Attached is a proposed patch for this issue. >>> >>> The patch looks good, I just committed it. >>> >>> Thanks! >>> >>> Ken >>> -- >>> Kenneth Merry >>> ken@FreeBSD.ORG >>> _______________________________________________ >>> freebsd-scsi@freebsd.org mailing list >>> http://lists.freebsd.org/mailman/listinfo/freebsd-scsi >>> To unsubscribe, send any mail to "freebsd-scsi-unsubscribe@freebsd.org" >> >> >> It looks that this patch or something related to it broke my tape backups. >> I do have two SCSI Tapes connected to my system: >> >> # camcontrol devlist >> <IBM-SSG S53D073 C61E> at scbus0 target 0 lun 0 (pass0,da0) >> <IBM-SSG S53D073 C61E> at scbus0 target 1 lun 0 (pass1,da1) >> <IBM-SSG S53D073 C61A> at scbus0 target 2 lun 0 (pass2,da2) >> <IBM-SSG S53D073 C61A> at scbus0 target 3 lun 0 (pass3,da3) >> <TANDBERG SLR5 4/8GB =09:> at scbus1 target 5 lun 0 (pass4,sa0) >> <COMPAQ DLT4000 D887> at scbus1 target 6 lun 0 (pass5,sa1) >> >> an with an 8.3 stable from Jun 14 both of them arent able anymore to do >> blocksizes over 8k and 8k are only working sometimes (huh?!). > > The change in the above email didn't get merged back to stable/8 until June > 20th. So it isn't that. > > There were no changes to the sa(4) driver in stable/8 from March 15th to > June 14th, but there were lots of other CAM changes. > >> # mt -f /dev/sa1 status >> Mode Density Blocksize bpi Compression >> Current: 0x1a:DLTapeIV(20GB) variable 81633 IDRC >> ---------available modes--------- >> 0: 0x1a:DLTapeIV(20GB) variable 81633 IDRC >> 1: 0x1a:DLTapeIV(20GB) variable 81633 IDRC >> 2: 0x1a:DLTapeIV(20GB) variable 81633 IDRC >> 3: 0x1a:DLTapeIV(20GB) variable 81633 IDRC >> --------------------------------- >> Current Driver State: at rest. >> --------------------------------- >> File Number: 0 Record Number: 0 Residual Count 0 >> >> # dd if=/dev/zero of=/dev/sa1 bs=1k count=1000 >> 1000+0 records in >> 1000+0 records out >> 1024000 bytes transferred in 4.330778 secs (236447 bytes/sec) >> # dd if=/dev/zero of=/dev/sa1 bs=2k count=1000 >> 1000+0 records in >> 1000+0 records out >> 2048000 bytes transferred in 3.252421 secs (629685 bytes/sec) >> # dd if=/dev/zero of=/dev/sa1 bs=4k count=1000 >> 1000+0 records in >> 1000+0 records out >> 4096000 bytes transferred in 2.933208 secs (1396423 bytes/sec) >> # dd if=/dev/zero of=/dev/sa1 bs=8k count=1000 >> 1000+0 records in >> 1000+0 records out >> 8192000 bytes transferred in 3.567864 secs (2296052 bytes/sec) >> # dd if=/dev/zero of=/dev/sa1 bs=16k count=1000 >> dd: /dev/sa1: Input/output error >> 1+0 records in >> 0+0 records out >> 0 bytes transferred in 0.000253 secs (0 bytes/sec) >> >> There is no error message from the kernel related to that. You may try to boot with verbose kernel messages and enable some CAM debug with `camcontrol debug -I all` and then repeat your tests. It should probably make kernel to tell more about errors. >> If I try to read an older backup tape (used 64k Tape Blocks for that): >> # dd if=/dev/sa1 of=/dev/null bs=64k count=10 >> dd: /dev/sa1: Input/output error >> 0+0 records in >> 0+0 records out >> 0 bytes transferred in 0.000824 secs (0 bytes/sec) >> # >> ... I get in /var/log/messages: >> >> Jun 25 14:56:05 unicorn kernel: (sa1:sym0:0:6:0): 65536-byte tape record >> bigger than supplied buffer >> >> Nice ehy? >> >> I've now booted kernel.old from Mar 15 and the problems are gone on both >> drives. > > It isn't obvious where the problem was introduced, unfortunately. > > Could you do a binary search to figure out which revision broke things for > you? -- Alexander Motin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4FE88C33.1020008>