Don t work too hard Use SunSolve

Don't work too hard! Use SunSolve!

Instead of trying to debug UNIX internals code without the proper tools (Source!), let's jump into research mode to see if this is a known problem. Yes, we hear a few of you mumbling "It's about time!", but think of all that you learned.

Using Sun's SunSolve program, let's search the bugs and patches databases for the following keywords:

• fifo_rdwr

• vno_rw

• alignment

• 4.1.3

Picking search keywords is sometimes an art of its own, so let's explain why we chose the keywords we did. The first two are from the stack traceback. Alignment is a good word to use to search for memory alignment panics. Specifying the operating system will help eliminate non-4.1.3 bug reports . The results in this search: Bug 1050077.

Bug 1050077 reports a problem with the following stack traceback shown as an example.

 physmem bf3  _panic(0xf81514e9,0xf825acbc,0x0,0x0,0xf825b,0xf7fffc78) + 6c  _trap(0x7,0xf825acbc,0x0,0x0,0x0,0x0) + 184  fixfault(?)  _uiomove(0x0,0x0,0x1000,0x2,0x1002,0x2) + 50  _fifo_rdwr(0xff0b3600,0xf825aeac,0x0,  0xff0d5757  ,0x2,0x0) + 5ac  _vno_rw(0xf81b39e0,0x1,0xf825aeac,0x1000,0xff0b3604,0x0) + a4  _rwuio(0xf81b39e0,0xf825aeac,0xf825aea4,0x1000,0x1000,0xf825aeac) + 2b0  _write(0xf825afe0,0x20,0xf813af88,0xf813afa8,0xf825b000,0xf813afa8) + 34  _syscall(0xf825b000) + 3b4 

Interesting, isn't it? You couldn't ask for a much nicer match!

Next, we search the patch database for bug 1050077 and get a match on patch number 100347. The patch description reads as follows :

"Problem Description: Under a heavy load, writing to a named pipe (fifo) can cause a kernel panic, if one doesn't have the definition in fifo.h set up correctly. Specifically, the definition for FIFOBUF must be greater than FIFOBSZ. Otherwise, you end up using a union instead of a structure, and the union pointer gets overwritten."