65 lines
2.8 KiB
Plaintext
65 lines
2.8 KiB
Plaintext
|
#! /bin/sh /usr/share/dpatch/dpatch-run
|
|||
|
|
## 003-Fix_disk_enumeration_loop.dpatch by Pat Suwalski <pat@suwalski.net>
|
|||
|
|
##
|
|||
|
|
## DP: When my controller has a failed disk, megasasctl fails.
|
|||
|
|
## DP:
|
|||
|
|
## DP: I have tracked the problem down to a bad assumption, that the various
|
|||
|
|
## DP: disk enumeration loops would use the number of currently online drives
|
|||
|
|
## DP: as the basis for the iterator.
|
|||
|
|
## DP:
|
|||
|
|
## DP: This number decreases when a disk fails. It is not noticeable in many
|
|||
|
|
## DP: cases, because on many some controllers pinfo->pd_present_count is
|
|||
|
|
## DP: one higher than the actual number of disks.
|
|||
|
|
## DP: However, the same situation would be encountered in a RAID1 system with
|
|||
|
|
## DP: two failed disks.
|
|||
|
|
## DP:
|
|||
|
|
## DP: The attached one-line patch uses the number of ports
|
|||
|
|
## DP: on the controller as the basis.
|
|||
|
|
|
|||
|
|
@DPATCH@
|
|||
|
|
|
|||
|
|
This patch fixes a segfault based on an incorrect assumption that the number
|
|||
|
|
of physical drives to loop over is the number of currently online drives
|
|||
|
|
connected to the controller. This change plays it safe and makes it use the
|
|||
|
|
number of ports on the controller, with a fallback for the worst case
|
|||
|
|
scenario.
|
|||
|
|
|
|||
|
|
Pat Suwalski <pat@suwalski.net>
|
|||
|
|
|
|||
|
|
diff -ur megactl-code/src/adapter.c megactl-code2/src/adapter.c
|
|||
|
|
--- a/src/adapter.c 2014-01-20 11:13:48.114598462 -0500
|
|||
|
|
+++ b/src/adapter.c 2014-01-20 14:33:32.950851825 -0500
|
|||
|
|
@@ -706,7 +706,31 @@
|
|||
|
|
}
|
|||
|
|
qsort (a->channel, a->num_channels, sizeof (*a->channel), cmpChannel);
|
|||
|
|
|
|||
|
|
- a->num_physicals = pinfo->pd_present_count;
|
|||
|
|
+ /* Some notes:
|
|||
|
|
+ Different meanings on different models.
|
|||
|
|
+ - FC_MAX_PHYSICAL_DEVICES used on older controllers, which is 256
|
|||
|
|
+ disks (overallocation)
|
|||
|
|
+ - pd_disk_present_count is number of working drives, not counting
|
|||
|
|
+ missing drives
|
|||
|
|
+ - pd_present_count is unclear. It is pd_disk_present_count + 1 on some
|
|||
|
|
+ controllers
|
|||
|
|
+ - device_interface.port_count contains number of physical ports on the
|
|||
|
|
+ controller
|
|||
|
|
+
|
|||
|
|
+ pd_present_count was used here, but in some controllers causes segfaults
|
|||
|
|
+ when there is a failed drive, and not enough space is allocated.
|
|||
|
|
+
|
|||
|
|
+ Since there cannot be more devices than there are ports, that is a safe
|
|||
|
|
+ number to set without going overboard.
|
|||
|
|
+ */
|
|||
|
|
+ a->num_physicals = pinfo->device_interface.port_count;
|
|||
|
|
+
|
|||
|
|
+ /* On some controllers, namely the PERC6e, the controller does not know
|
|||
|
|
+ how many ports there are in the enclosure. Fall back to the worst case
|
|||
|
|
+ scenario. */
|
|||
|
|
+ if (a->num_physicals < pinfo->pd_disk_present_count)
|
|||
|
|
+ a->num_physicals = FC_MAX_PHYSICAL_DEVICES;
|
|||
|
|
+
|
|||
|
|
if ((a->physical = (struct physical_drive_info *) malloc (a->num_physicals * sizeof (*a->physical))) == NULL)
|
|||
|
|
return "out of memory (physical drives)";
|
|||
|
|
memset (a->physical, 0, a->num_physicals * sizeof (*a->physical));
|