Attachment 144217 Details for Bug 193543 – Brief analysis of code paths and possible approaches

Brief analysis of code paths and possible approaches

notes.txt (text/plain), 2.30 KB, created by Dave Wysochanski on 2006-12-21 21:03:26 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Dave Wysochanski

Created: 2006-12-21 21:03:26 UTC

Size: 2.30 KB

patch

obsolete

>Problem here is that report_init sets the report type to "PVS" if it sees
>pv on the option line.  The code that changes it is here:
>(gdb) n
>1228            if (!_parse_options(rh, format))
>(gdb) p *rh
>$4 = {cmd = 0x80e9344, mem = 0x80ffbe0, type = VGS,
>  field_prefix = 0x80d8ff8 "vg_", flags = 29, separator = 0x80cc4de " ",
>  keys_count = 0, field_props = {n = 0x810107c, p = 0x810107c}, rows = {
>    n = 0x8101084, p = 0x8101084}}
>(gdb) n
>1231            if (!_parse_keys(rh, keys))
>(gdb) p *rh
>$5 = {cmd = 0x80e9344, mem = 0x80ffbe0, type = 6,
>  field_prefix = 0x80d8ff8 "vg_", flags = 29, separator = 0x80cc4de " ",
>  keys_count = 0, field_props = {n = 0x8103670, p = 0x8103690}, rows = {
>    n = 0x8101084, p = 0x8101084}}
>(gdb) p format
>$6 = 0xbf9ceebc "pv_name,vg_name"
>(
>
>Basically 
>report_init
>  _parse_options
>    _field_match
>
>
>But we make a decision on the report type before we determine if we have
>any vg's on the cmdline (see switch stmt right after call to report_init in
>_report function).  This seems like the fundamental issue.
>
>
>I tried setting rh->type back to 4 value (VGS) but I got a segfault in
>_dev_name_disp, called from report_object.
>
>Next step I think is to grok report_object and subtleties of different
>reports....
>
>Ok, some code in report_object that is somewhat odd...
>		switch (_fields[fp->field_num].type) {
>		case LVS:
>			data = (void *) lv + _fields[fp->field_num].offset;
>			break;
>		case VGS:
>			if (!vg) {
>				skip = 1;
>				break;
>			}
>			data = (void *) vg + _fields[fp->field_num].offset;
>			break;
>		case PVS:
>			data = (void *) pv + _fields[fp->field_num].offset;
>			break;
>		case SEGS:
>			data = (void *) seg + _fields[fp->field_num].offset;
>			break;
>		case PVSEGS:
>			data = (void *) pvseg + _fields[fp->field_num].offset;
>		}
>
>
>The reason I'm segfaulting when I try to set the type back to VGS is because
>this code sets the data pointer based on the field type and 'pv' variable, 
>which is NULL when it gets passed into this function.  Apparently for report
>type "VGS" we don't need any detailed pv info, so pv is NULL.
>
>
>So maybe it is easier to stay with the PVS report_type and try to use some
>sort of filter logic...  ...but... we are calling into process_each_pv and
>in there we assume the cmdline has pvs on it....  *sigh*
>
>Ok, this clearly requires some thought to fix - maybe agk has some insight.

Actions: View

Attachments on bug 193543: 144217 | 146730