Date: Fri, 4 Sep 2009 22:17:31 GMT From: Brad Huntting <huntting@glarp.com> To: freebsd-gnats-submit@FreeBSD.org Subject: ports/138540: sysutils/smartmontools Message-ID: <200909042217.n84MHVi4089170@www.freebsd.org> Resent-Message-ID: <200909042220.n84MK1vu023820@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
>Number: 138540
>Category: ports
>Synopsis: sysutils/smartmontools
>Confidential: no
>Severity: non-critical
>Priority: medium
>Responsible: freebsd-ports-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Fri Sep 04 22:20:00 UTC 2009
>Closed-Date:
>Last-Modified:
>Originator: Brad Huntting
>Release: 7.2-PRERELEASE
>Organization:
>Environment:
FreeBSD PharLap-00005045b076 7.2-PRERELEASE FreeBSD 7.2-PRERELEASE #0: Mon Aug 31 14:58:08 MDT 2009 * amd64
>Description:
smartctl likes to read SMART data into 512 byte buffers which it allocates on the stack. While this _should_ work just fine, I have found that sometimes, when using SATA over SAS (-d sat) with an MPT Fusion HBA, smartctl's buffer gets broken into 2 Scatter Gather Elements (probably because the 512 byte buffer stradles a page boundary). Sometimes, only the first of these gets filled before the mpt driver records a timeout and subsequently resets the SAS bus.
I have not fully diagnosed the problem or had time to write a sample code'let that can reproduce it. But in the mean time, this patch moves the data buffer from the stack to the heap where it will be aligned to avoid a page boundary.
Hopefully I will have time to diagnose the real problem and either fix it (if it's in the driver) or send in a bug report to LSI (if it's in the MPT controller).
thanx,
brad
>How-To-Repeat:
pray. Then on a 64bit system, with an 8gig LSI MPT SAS controler
Mpi Version 1.5.20.0
FWVersion 1.28.0.0
or FWVersion 1.28.2.0
connected to a SATA drive, turn on all available debugging in CAM and the MPT driver.
run smartctl -d sat -r ioctl -a /dev/da<foo>
sometimes, if you're 'lucky', you'll see a bus reset, and smartcl will fail to get some of it's SMART data.
>Fix:
Attached is a patch to the port that adds a patch that makes shure the data buffers used to read smart attributes and logs do not straddle page boundaries.
Patch attached with submission follows:
diff -Nru /usr/ports/sysutils/smartmontools/Makefile /home/huntting/src/Copan/smartmontools/Makefile
--- /usr/ports/sysutils/smartmontools/Makefile 2009-07-08 13:56:07.000000000 -0600
+++ /home/huntting/src/Copan/smartmontools/Makefile 2009-09-04 14:58:21.000000000 -0600
@@ -25,7 +25,7 @@
MAN5= smartd.conf.5
MAN8= smartd.8 smartctl.8
-CFLAGS:= ${CFLAGS:S/-O2/-O/}
+CFLAGS:= ${CFLAGS:S/-O2/-O/} -Wno-write-strings
post-patch:
@${REINPLACE_CMD} -e 's| install-initdDATA | |' ${WRKSRC}/Makefile.in
diff -Nru /usr/ports/sysutils/smartmontools/files/patch-scsiata.cpp /home/huntting/src/Copan/smartmontools/files/patch-scsiata.cpp
--- /usr/ports/sysutils/smartmontools/files/patch-scsiata.cpp 1969-12-31 17:00:00.000000000 -0700
+++ /home/huntting/src/Copan/smartmontools/files/patch-scsiata.cpp 2009-09-04 14:28:13.000000000 -0600
@@ -0,0 +1,22 @@
+--- scsiata.cpp.orig 2007-12-02 19:14:20.000000000 -0700
++++ scsiata.cpp 2009-09-04 14:27:48.000000000 -0600
+@@ -380,14 +380,14 @@
+ return 1, else 0 */
+ int has_sat_pass_through(int device, int packet_interface)
+ {
+- char data[512];
++ int ret;
++ char *data = (char *)malloc(512); // (alignment)
+ smart_command_set command;
+
+ command = packet_interface ? PIDENTIFY : IDENTIFY;
+- if (0 == sat_command_interface(device, command, 0, data))
+- return 1;
+- else
+- return 0;
++ ret = ( 0 == sat_command_interface(device, command, 0, data) ? 1 : 0 );
++ free(data);
++ return ret;
+ }
+
+ /* Next two functions are borrowed from sg_lib.c in the sg3_utils
>Release-Note:
>Audit-Trail:
>Unformatted:
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200909042217.n84MHVi4089170>