[Orca-dev] Re: One stray system with disk problems

[Dmitry Berezin]

Hi Jon,

I have to say I feel bad that I did not get a chance to reply with this sooner since I promised a patch for this problem. Well, here you go...

First of all, here is my understanding of the problem:
SE has a documented "feature" – "The MAX_DISK and MAX_IF constants will often have a larger value than the number of actual resources on the system." (SE User Manual); thus, GLOBAL_diskinfo_size, being set to MAX_DISK will also be greater then the actual number of disks in the system. In orcallator.se, RAWDISK code uses GLOBAL_disk_info[] array to map short disk names, reported by kstat, to long names. Raw_disk_map function implements a rather convoluted (sorry Alan) way of dealing with this – it uses multiple calls to raw_disk_short_name_cmp function and the assumption that disk names come before partition names in GLOBAL_disk_info[] to detect the end of the entries with data. Well, there are several of (much easier, IMHO) ways of doing this:

1. Do what disks.se does – 
  for(i=0; i<GLOBAL_diskinfo_size; i++) {
    if (GLOBAL_disk_info[i].short_name == nil) {
      break;
    }
  ...
  }
2. Use GLOBAL_disk_count instead of GLOBAL_diskinfo_size. GLOBAL_disk_count is initialized in live_rules.se (used by orcallator.se) and is equal to the actual number of disks.
2a. Use RAW_disk_count instead of GLOBAL_diskinfo_size. It was just initialized to the actual number of disks.
3. Instead of comparing short disk names in a loop, use find_inst access function, defined in diskinfo.se to get an index into GLOBAL_disk_info[].

Based on the SE debug output that you have posted, this is exactly the place where orcallator.se crashes on your system.

Try this -  replace raw_disk_map() function with this version:

raw_disk_map() {
  int  i;
  int  index;

  for (i=0; i<RAW_disk_count; ++i) {
    // Do not map st & fd devices.
    if (strncmp(RAW_disk[i].short_name, "st", 2) != 0 &&
        strncmp(RAW_disk[i].short_name, "fd", 2) != 0) {
      index = find_inst(RAW_disk[i].short_name);
      if (index != -1) {
        strcpy(RAW_disk[i].long_name, GLOBAL_disk_info[index].long_name);
      }
    }
  }
  RAW_disk_map = 0;
}

  Hope this will fix it.

  -Dmitry.

  P.S. Can you send me the full se debug output with the crash (orcallator 1.37). Based on the data in your other posts, you shouldn't have had this problem, so it would be very helpful for me to understand why _you_ ran into this issue.