Some fsr/fsctl (reorganizer) verification and stress stuff.
Both of the following tests unearthed bugs during development.
Both of the following could use more polishing and more comprehensive
test cases.
Both tests must be run as root and require that the running kernel
be lbooted with 'fsctl' and that /dev/fsctl exist.
Both take an argument specifying the directory holding the unshipped
programs used in the scripts.
Both take an argument specifiying a directory in which to work.
Both will give a usage if invoked with no arguments.
fsrverify:
A brief test of reorganizer functionality for a subset
of efs and reorganizer corner cases.
fsrstress:
A more involved test which reorganizes a file system under
heavy usage. If this runs for 24 hours without the kernel crashing
or assert failing or fail in a way it itself detects.SHIP IT! :-)
(My scsi based IP4.5 machine does a tad worse than one pass per minute
on a file system with 6MB free).
exop.c: various ops at the extent level
fop.c: various ops at the file level
indir.c: reorganize indirect extents as specified
random.c: random range generator
rsf.c: disorganize a file into extents of a specified size
testfsctl.c: test the most critical /dev/fsctl entry points
assfail.c:
A distillation of a problem uncovered running fsrstress.
----------------------------------------------------------------------
The following are some notes on verifying and excercising the
reoganizer. The above tests represent a subset of these notes:
File System Reorganizer Test Plan
fsctl the driver
fsr the reorganizer utility program
A) testing fsctl
open
-verify exclusive open
-uid==0 only
close
-verify cleanup of
-inode (unlocked)
-file system can unmount
-other processes can use inode
-device (next open will succeed)
-unused BALLOC'ed extents
ioctl
ILOCK
-inode is locked
-should fail if another inode already locked
ICOMMIT
-inode is the one locked by ILOCK
-proper extents and indirect extents:
-valid data blocks
-data blocks allocated
-proper number of indirect extents
-valid offsets
-valid magic number
-total bbs in extents consistent with file size
BALLOC
-bn+len saved for bugout-close
BFREE
-arguments valid data blocks
-extent previously free/alloc (as appropriate)
-should fail if no active "ILOCK"
-should fail if in different dev than ILOCK'ed inode
TSTALLOC
TSTFREE
-args valid data blocks
B) testing fsr
"better" file system organization
file data correct
-fsr responsible for copying data blocks
file meta-data correct
-same times, size, mode (type,perms), uid, gid
-movex paths
-direct extents
-proper blending
-proper length
-proper offset
-indirect extents
-indirect to direct extents
-#direct stays <= EFS_DIRECTEXTENTS
-#direct decreased enough to free indirect block(s)
-#direct increased above EFS_DIRECTEXTENTS
-#direct increased to an indirect block(s)
-blocks
-should properly allocate direct blocks before ICOMMIT
-should properly free old blocks after ICOMMIT
-properly free any disused indirect blocks after ICMT
file system consistent
-on-disk:
-in-core:
fs->fs_tfree
fs->fs_fmod
cg->cg_dfree
cg->cg_firstdfree
Some numbers:
EFS constants
12 direct extents
64 extents per block (indirect extents)
248 maximum extent size
32 maximum indirect extent size
24576 maximum number of extents
file system parameters
firstcg
cgisize
cgfsize
ncg
Some tests:
-gun down user level process
-should leave:
-device open-able (cleaned up)
-inode unlocked and unchanged
-data valid
-file system valid
-reset system (hard)
-should leave data intact
-organize files with the following number of extents:
11,12,13
63,64,65
127,128,129
24575,24576
24576+ should fail
-verify the following changes in number of extents:
(checks out the bottom half of 'fsr')
11->12
12->13 0->1 indirbbs
13->12 1->0 indirbbs
12->1
13->1 1->0 indirbbs
63->64
64->65 1->2 indirbbs
65->64 2->1 indirbbs
128->129 2->3 indirbbs
129->1 3->0 indirbbs
-run a job mix on the file system being reorganized:
make clobber, make all
can you reorganize a running executible? the reorganizer itself?
-(artificially) organize a file system (or part of it) as follows:
-single extent file with 1st block at cg's first dbn
-multi extent file with 1st block at cg's first dbn
-multi extent file with middle extent ending at cg's last dbn
-multi extent file with middle extent ending at fs's last dbn
-non-linear sequence multi extent file
More things to test/verify/stress
-recycled/non-recycled inodes
(assfail.c tests fsctl's preparation of an in-core inode which
will be recycled by the inode cache and presented to EFS)
