I've tried replacing hardware, using 60ns DRAM with 70ns timings,
using mem=nopentium on a step 7, and Benjamin LaHaise's second vfs
patch. No luck!
This pair of shell scripts reliably causes a "bad dir: ... no data
block" roughly half of the time when run as a regular (non-super)
user. I use a small (128MB) spare IDE EXT2 FS mounted on /foo for
testing to reduce fsck times. The test usually leaves the filesystem
not-unmountable due to one or more unkillable rmdir processes.
Note that ~news/active is used as a source of directory names. You
may need to make alternative arrangements.
--Mike
begin 644 stress_1.tgz
M'XL(`,YEH3(``^W4WVK",!0&\%[[%!]NK'I1DU1;+]S-WJ-,2HVEB$U)XA3V
MY]F7CLG&+B:[<")\OYO`.:<T;7*.\U8[MU31&2DIYUF&"#WY8PV4RH`\S62:
MS5.5AH#,Y2R"/.>FCG;.EQ:(K#'^M[I3^2NU-A8'-"T44DPQ6V!E,!JYSVNQ
M++'=K!H;PCN(M3$+?$O9;4AA?#?NGVKUX-(?0W_V=9KG>\>I_D^E.O9_ED_S
M?B`H)=G__^$&#UUGS2',7"D1VEE7WMA&NX&NK>YP_];JO1-EY9LGC?AQ5)EM
M-RXF,5Y0[C>(GSO;M![#?CB((6[5:Y]R>H4DU#M13$0A1/T1--:'Y5#:VH5"
B)!V2Y-(_@(B(B(B(B(B(B(B(B(B(B.C*O0.*<D6'`"@`````
`
end