* possible auto-partition bug? (linux-2.4.20-8)
@ 2003-08-15 17:55 Pete Nishimoto
2003-08-15 20:23 ` Andries Brouwer
0 siblings, 1 reply; 3+ messages in thread
From: Pete Nishimoto @ 2003-08-15 17:55 UTC (permalink / raw)
To: andrea; +Cc: linux-kernel, Peten
Hi,
My name is Pete Nishimoto and I work for Sun Microsystems
as a linux device driver developer, currently working with
RedHat 9.0 (2.4.20-8) and I believe I have found a problem
with the partitioning logic and the pager, which I've
detailed below. I look forward to any replies/comments
and thanks in advance to all who review/read this.
Pete (Peter.Nishimoto@sun.com)
------------------------------------------------------------------------------
I am writing a virtual disk driver for linux (a container
file within which resides the "linux" disk). To that end,
I am able to freely to specify the disk geometry in any terms,
specifically we can manipulate the maximum number of heads.
I believe the I have uncovered a problem with the
autopartitioning logic used during installation.
The result is that there are partitions created that cannot be
successfully scanned by badblocks. Note that even though I
found this on our virtual disks, when applying the same
installation criteria to real PC harddisks, about half of
the PC harddisks exhibited the same problem (badblock scan fails).
At the end of this messages is appended some results of
real disks on real PCs for which badblocks fails.
This could be considered a non-problem, since it appears that
only badblocks fails, and it only fails on the last
block of the partition. One might consider saying that
these are just true "bad" blocks and can be discarded/disregarded.
Further, these "bad" blocks would only be see if something
tried to read (or write) to the last block of the partition.
However, the problem (as I had uncovered it). is that the
badblock utility is used during install, and, if badblocks
reports an error during install, the installation will halt.
Again, this might be explained away because it requires a
"non-standard" installation sequence. The exact installation
sequence must be:
- start with a disk < 6 GBytes
- autopartition the entire disk
- check that a badblocks scan should be done
- select *all* packages are to be installed. Note
that since this is a small disk, not all packages
will fit
- watch the first badblock scan succeed
- installation will then see that not all packages
will fit, so the user is prompted to respecify the
packages to install.
- once a new (smaller) package list has been specified,
installation will run badblocks again.
- this 2nd pass of badblocks will report and error and
the installation fails.
The specifics are that the autopartitioning logic will allocate
partitions on cylinder boundaries. By definition, each cylinder
contains an integral number of sectors (0x200 bytes/sector).
However, during partitioning, mke2fs (and probably others like
mke3fs) will set the blocksize for the filesystem. For smaller
partitions, this is usually set to 0x400 bytes. However, for larger
partitions, this is set to 0x1000 bytes. This is effectively
the page size for filesystem. The problem occurs because
when the partitions were generated, they may not contain
an integral number of blocks (pages). Thus, as badblocks
traverses the partition and requests reads only of 0x1000 size
blocks, at the end of the partition, since the partition is
not an integral number of blocks/pages, the paging read
fails (generic_make_request() in ./drivers/block/ll_rw_blk.c).
While this may not seem to be a problem, as the blocks size is
stored in the superblock, during installation the kernel has
retained some data structures (some inodes) generated prior
to the change in blocksize.
During autopartitioning (see parted-1.6.5/libpared/disk.c)
the partitioning logic uses a granularity of
(max sectors * max head ), which is the number of sectors per
cylinder. Therefore, each partition will be allocated on
cylinder bondaries, reguardless of overall disk size.
A following is an actual example from our disk driver (kdev ==
0x1602):
Geometry ( 3.29 Gbytes, 6426000. (620d90) Sectors, 3290112000. (c41b2000)
Bytes)
nCyls :0x00000190 ( 400.)
nHeads :0x000000ff ( 255.)
SectorsPerTrack :0x0000003f ( 63.)
BytesPerCylinder :0x007d8200 ( 8225280.)
BytesPerTrack :0x00007e00 ( 32256.)
StartingOffset :0x00000400 ( 1024.)
Installation autopartitioning will allocate in the following manner:
/boot = 208782. (0x32f8e)
/ = 5172930. (0x4eeec2)
/swap = 1044255. (0xfef01)
---------
6425967.
Now, the linux pager read/writes the disk in units of
diskblocks. This blocksize is stored
in a couple of places, one of which is the superblock, but
it is also stored in a global array blksize_size[], which is
indexed per device major/minor number.
The sequence order for the file system generation is:
- create the partitions (via a set of 'parted' libraries)
- create the filesystem (via mkfs)
The autopartitioner will divvy up the disk as best as possible
into 3 partitions, using the (above mentioned) granularity as
a limiting factor.
When generating the filesystem for ext2 (via mke2fs),
mke2fs sets the blocksize depending upon
the size of the partition. Specifically, the code path
is (where size is in MBytes):
if (size >= 0) blocksize = 0x1000
if (size <= 0x200) blocksize = 0x400
if (size <= 0x3) blocksize = 0x400
Of particular note, when the partitions are created, the
blocksize isn't taken into account (in fact, it hasn't even been
set yet). Thus, we have partitions that aren't an integral
multiple of the blocksize.
Note that:
sector size == 0x200 bytes
initial block size == 0x400 bytes
linux page size == 0x400 or 0x1000 bytes (blocksize)
For our default disk case, note that:
4eeec2 % 3ec1 == 0
4eeec2 % 400/200 == 0
4eeec2 % 1000/200 != 0
This means, that the partition, which is on a cylinder boundary,
is an integral number of sectors (which is obvious), but it
IS NOT an integral number of 1000/200 units, where 0x1000 is
a valid page blocksize.
When badblocks traverses the disk, it causes the pager
to sequence through the partition, and it does this in
blocksize blocks (the blocksize specified for badblocks is for
the usermode badblocks program only, and does not have anything
to do with the pager's blocksize). So, since the partition
is NOT an integral multiple of blocksize, in the above
partition, the end of the partition cannot be read by
a single page, since it is less than a page in size.
Thus, badblocks will die.
The blocksize for our drivers are initially set at driver startup.
However, during autopartitiong, when the partition size has been
determined, libparted (invoked from autopart.py during
installation) DECIDES TO CHANGE the blocksize, from 0x400
(which is our default) to 0x1000. This information is then
written to the disk (the superblock) and blksize_size[]
has been updated, but some extant
internal kernel data structures (inodes) have not been updated,
thus there is data skew for the blocksize within the kernel.
Now, the block size is determined by the value
i_blksize stored in the inode. This is initially set by
alloc_inode() (./fs/inode.c), but is later modified by
bdget() (./fs/block_dev.c). However, the problem is that
bgget() does not check to see what the current blocksize of
the filesystem is. Note that there is no extant superblock
in the kernel superblock table for lookup at the time that
bgget() is reassociating the inode. Thus, the i_blksize
of an inode is that of the RAMDISK_DEVICE (the "template
device that was used to create the inode - i,e., the superblock
of the caller of alloc_inode() - 0x400), a default which is a kernel
system default and NOT the default supplied by the driver.
Thus, there is no way for an extant inode to get updated after
the new partition/filesystem has been generated, and a new blocksize
has been determined/set.
The precise inode sequence is as follows:
- mke2fs is invoked, setting the blocksize of the
filesystem to 0x1000
- alloc_inode() generates an inode based on the
RAMDISK superblock (sb->s_dev == 0x1). blocksize
is set to 0x400
- bdget() is called with our disk driver (kdev == 0x1602).
No mods for the blocksize. Note:
- the blksize_size[] entry for 0x1602 is now 0x1000
- there is no superblock entry in the global superblock
table for 0x1602, so a call to get_super()
(./fs/super.c) or even copy-coding the inline
routine find_super() (./fs/super.c) locally produces
no superblock for any 0x1600 device.
Now, for the first pass of badblocks, the inode->i_blkbits
is set at 0x400. This is fine as the number of sectors in
the partition is an integral number of 0x400. However, upon
the 2nd pass of badblocks, the inodes are recreated and the
"new" blocksize of 0x1000 is used. Since the partition is
not an integral number of 0x1000 blocks, the kernel
(generic_make_request() in ./drivers/block/ll_rw_blk.c) will
fail as there will be a paging request to read a page
(of blocksize) starting at sector 0x4eeec0.
Thus, there are 2 problems:
1) the autopartition "granularity". This is hardcoded to
be the max number of cylinders and heads (255. * 63.).
2) the changing paging blocksize
Problem #2 is directly related to problem #1 in that the granularity
must be an integral number of the paging blocksize. Since the
paging blocksize changes for the current implementation, then
we should set the paging blocksize to the maximum (0x1000)
so that it will be later changed from 0x400. Thus, the
granularity should be an integral mutiple of 8, since 0x1000
and 0x400 are multples of 0x8 and such a granularity will
be an integral mutiple of 0x400 and 0x1000 as well. This
value is then should be 248. * 63.
Attempts:
1) change default blocksize for our drivers.
result: doesn't affect partitioning, and we still get
badblocks because of block indexing.
2) force mke2fs to always use 0x400 as the blocksize
result: works
issue : do this for ext3 and other file systems?
issue : how to get this version in at install?
3) change superblock blocksize on fly
result: unknown
issue : when is a good time for this? There is no
kernel callback notification, and, when this is
done, it already has an inode active, etc.
4) change initial allocation so that we get good numbers
(e.g., granularity now 248 * 63, which is divisible by 8)
result : paritioning works, but 2nd pass of badblocks
still fails. And, if no badblocks check is made
during install, badblocks still fails after
install because the blocksize for the partition
has been set to 0x1000 (use debugfs to show).
5) change initial allocation so that we get good numbers
(e.g., granularity now 248 * 63, which is divisible by 8)
AND change default blocksize for our drivers to 0x1000.
This is to combine the autopartition compensation (granularity)
and to account for superblock blocksize override after
the file system is made.
"New" default redhat disk geometry:
Geometry ( 3.26 Gbytes, 6374592. (6144c0) Sectors, 3263791104. (c2898000)
Bytes)
nCyls :0x00000198 ( 408.)
nHeads :0x000000f8 ( 248.)
SectorsPerTrack :0x0000003f ( 63.)
BytesPerCylinder :0x007a1000 ( 7999488.)
BytesPerTrack :0x00007e00 ( 32256.)
StartingOffset :0x00000400 ( 1024.)
result: partitions fine, BUT mke2fs fails because
while the blocksize is now 0x1000, the kernel
still retains inodes for which the blocksize
is 0x400. Thus, when mke2fs makes a call to
zero_blocks() to zero out the "last" block,
it will obtain the inode with the wrong blocksize,
and this results in a seek failure. See appended log.
issue : can we change the inode->i_blkbits at the right time?
6) change initial allocation so that we get good numbers
(e.g., granularity now 248 * 63, which is divisible by 8)
AND change default blocksize for our drivers to 0x1000
AND put a kernel fix in bdget() to update the blocksize
result: partitions fine. installs fine.
badblocks works (does not fail).
-------------------------------------------------------------------------------
Successful tests of Attempt #6
Geometry ( 3.26 Gbytes, 6374592. (6144c0) Sectors, 3263791104. (c2898000)
Bytes)
nCyls :0x00000198 ( 408.)
nHeads :0x000000f8 ( 248.)
SectorsPerTrack :0x0000003f ( 63.)
BytesPerCylinder :0x007a1000 ( 7999488.)
BytesPerTrack :0x00007e00 ( 32256.)
StartingOffset :0x00000400 ( 1024.)
Geometry ( 5.44 Gbytes, 10624320. (a21d40) Sectors, 5439651840. (1443a8000)
Byte
s):
nCyls :0x000002a8 ( 680.)
nHeads :0x000000f8 ( 248.)
SectorsPerTrack :0x0000003f ( 63.)
BytesPerCylinder :0x007a1000 ( 7999488.)
BytesPerTrack :0x00007e00 ( 32256.)
StartingOffset :0x00000400 ( 1024.)
Geometry (13.38 Gbytes, 26123328. (18e9c40) Sectors, 13375143936.
(31d388000) By
tes):
nCyls :0x00000688 ( 1672.)
nHeads :0x000000f8 ( 248.)
SectorsPerTrack :0x0000003f ( 63.)
BytesPerCylinder :0x007a1000 ( 7999488.)
BytesPerTrack :0x00007e00 ( 32256.)
StartingOffset :0x00000400 ( 1024.)
-----------------------------------------------------------------------------
-----------------------------------------------------------------------------
"Real PCs" Autopartition Failure
On a real PC, the following is allocated:
==================================================================
Disk /dev/hda: 4303 MB, 4303272960 bytes
255 heads, 63 sectors/track, 523 cylinders, total 8404830 sectors
Units = sectors of 1 * 512 = 512 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 63 208844 104391 83 Linux
/dev/hda2 208845 7630874 3711015 83 Linux
/dev/hda3 7630875 8401994 385560 82 Linux swap
Command (m for help): u
Changing display/entry units to cylinders
Command (m for help): p
Disk /dev/hda: 4303 MB, 4303272960 bytes
255 heads, 63 sectors/track, 523 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 1 13 104391 83 Linux
/dev/hda2 14 475 3711015 83 Linux
/dev/hda3 476 523 385560 82 Linux swap
==================================================================
Now:
7630874
-208845
-------
7422029+1 => 7422030 sectors = 71404d
71404d % 8 != 0
Badblocks -sv /dev/hda2 fails.
Disk label specs:
8894 Cyl
4303 MB
15 Heads
63 Sectors
-------------------------------------------------------------------------------
Another PC failure:
==================================================================
Disk /dev/hda: 20.4 GB, 20493656064 bytes
255 heads, 63 sectors/track, 2491 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 1 13 104391 83 Linux
/dev/hda2 14 2443 19518975 83 Linux
/dev/hda3 2444 2491 385560 82 Linux swap
Command (m for help): u
Changing display/entry units to sectors
Command (m for help): p
Disk /dev/hda: 20.4 GB, 20493656064 bytes
255 heads, 63 sectors/track, 2491 cylinders, total 40026672 sectors
Units = sectors of 1 * 512 = 512 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 63 208844 104391 83 Linux
/dev/hda2 208845 39246794 19518975 83 Linux
/dev/hda3 39246795 40017914 385560 82 Linux swap
Command (m for help):
==================================================================
badblocks -sv /dev/hda2 fails.
Disk label specs:
16383 Cyl
16 heads
63 sectors
-------------------------------------------------------------------------------
Another PC failure:
==================================================================
Disk /dev/hda: 40.0 GB, 40020664320 bytes
255 heads, 63 sectors/track, 4865 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 1 13 104391 83 Linux
/dev/hda2 14 4817 38588130 83 Linux
/dev/hda3 4818 4865 385560 82 Linux swap
Command (m for help): u
Changing display/entry units to sectors
Command (m for help): p
Disk /dev/hda: 40.0 GB, 40020664320 bytes
255 heads, 63 sectors/track, 4865 cylinders, total 78165360 sectors
Units = sectors of 1 * 512 = 512 bytes
Device Boot Start End Blocks Id System
/dev/hda1 * 63 208844 104391 83 Linux
/dev/hda2 208845 77385104 38588130 83 Linux
/dev/hda3 77385105 78156224 385560 82 Linux swap
==================================================================
badblocks -sv /dev/hda2 fails.
No Disk label specs
-----------------------------------------------------------------------------
------------------------------------------------------------------------------
Debug Log of mke2fs failure for "Attempt #5"
#### mke2fs invoked
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: built on Aug 11 2003 at 10:17:05
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: argc 2.
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: arg 0. = /usr/sbin/mke2fs
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: arg 1. = /tmp/hdc2
Mon Aug 11 10:47:53 2003 : (mke2fs) PRS: entry - param 805abc0
param.s_log_block
_size 0 blocksize 0 blocks_count 0
Mon Aug 11 10:47:53 2003 : (lib/ext2fs/getsize.c) ext2fs_get_device_size: entry
- file /tmp/hdc2
Mon Aug 11 10:47:53 2003 : (32) alloc_inode: inode c277d480 sb c1aa0400 s_dev 1
s_blocksize_bits a
Mon Aug 11 10:47:53 2003 : (32) bdget: inode c277d480 old i_dev 1 new
i_dev/i_rd
ev 1602 bdev 0
Mon Aug 11 10:47:53 2003 : (lib/ext2fs/getsize.c) ext2fs_get_device_size:
exit -
BLKGETSIZE size 4e3240 blocksize 400 retblocks 271920
Mon Aug 11 10:47:53 2003 : (mke2fs) PRS: ext2fs_get_device_size results -
dev_si
ze 271920
#### mke2fs/set_fs_defaults() called, which determines the blocksize
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: entry - super 805abc0
fs_ty
pe 0 blocksize 0 super->s_log_block_size 0 s_blocks_count 271920
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: megs 9c6 p->size 0
p->b
locksize 1000 type default
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: NOT SKIPPED
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: setting blocksize -
sett
ings type default p->blocksize 1000
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: set
s_log_block_size 2 b
locksize 1000
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: megs 9c6 p->size
200 p-
>blocksize 400 type default
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: megs 9c6 p->size 3
p->b
locksize 400 type default
Mon Aug 11 10:47:53 2003 : (mke2fs) set_fs_defaults: exit - super 805abc0
blocks
ize 0 super->s_log_block_size 2
Mon Aug 11 10:47:53 2003 : (mke2fs) PRS: exit - param 805abc0
param.s_log_block_
size 2 blocksize 0
#### alloc_inode 0xc277d480 with superblock of s_dev == 0x1, blocksize 0x400
Mon Aug 11 10:47:53 2003 : (32) alloc_inode: inode c277d480 sb c1aa0400 s_dev 1
s_blocksize_bits a
#### bdget() called with inode 0xc277d480 and dev == 0x1602 (sundska2)
#### no change to blocksize (s_blocksize_bits == 0xa). The blocksize
#### for 0x1602 is 0x1000, which is equivalent to blocksize_bits of 0xc
Mon Aug 11 10:47:53 2003 : (32) bdget: inode c277d480 old i_dev 1 new
i_dev/i_rd
ev 1602 bdev 0
#### continue execution (debug left here for elucidation)
Mon Aug 11 10:47:53 2003 : (32) SunDskOpen: entry - Inode ccf9bdc0 Filep
cd13c84
0
Mon Aug 11 10:47:53 2003 : (32) SunDskOpen: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5bfc4 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5bfc4 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5bfc4 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 16c
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) __add_to_page_cache: page c108b690 offset 5c
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err 6c 108.
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: entry - inode c277d480
*ppos
400 i_rdev 1602 i_blkbits a
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: pos 400 limit 0 count
ffffff
ff
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (32) __add_to_page_cache: page c10896a0 offset 0
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: entry inode c277d480
i_bl
kbits a block 0
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: rdev 1602
inode->i_blkbits a b
h ccc8ecc0 iblock 1
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: rdev 1602
inode->i_blkbits a b
h ccc8e240 iblock 2
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: rdev 1602
inode->i_blkbits a b
h ccf07c40 iblock 3
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: exit - err c00 3072.
Mon Aug 11 10:47:53 2003 : (32) alloc_inode: inode c277d080 sb c1a7ec00
s_dev 10
0 s_blocksize_bits a
Mon Aug 11 10:47:53 2003 : (32) do_generic_file_read: entry - ppos 0 index
0 off
set 0
Mon Aug 11 10:47:53 2003 : (32) block_read_full_page: entry - iblock 0 page
c108
96a0 page->index 0
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: rdev 1602
inode->i_blkbits a b
h ccc8ed40 iblock 0
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) block_read_full_page: calling submit_bh
ccc8ed40
Mon Aug 11 10:47:53 2003 : (32) submit_bh: entry - b_blocknr 0 bh->b_size
400 co
unt 2
Mon Aug 11 10:47:53 2003 : (32) submit_bh: exit
Mon Aug 11 10:47:53 2003 : (32) block_read_full_page: submit_bh complete
Mon Aug 11 10:47:53 2003 : (32) SunDskRequest: entry - Queue c0314238
Mon Aug 11 10:47:53 2003 : (32) SunDskIpcReadWrite: entry - Device d00ccc20
Read
Write 0 StartSector 0 NumSectors 2 Buffer c2742000
Mon Aug 11 10:47:53 2003 : (32) CalcCHS: Device d00ccc20 DiskNum 0 PartitionNum
2 LBA 0
Mon Aug 11 10:47:53 2003 : (32) CalcCHS: exit - Cylinder d Head 0 Sector 1
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c030 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c030 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c030 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err 128 296.
Mon Aug 11 10:47:53 2003 : (32) SunDskIpcReadWrite: exit - status 0
Mon Aug 11 10:47:53 2003 : (32) SunDskRequest: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c158 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c158 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c158 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err 61 97.
Mon Aug 11 10:47:53 2003 : (32) file_read_actor: entry - page c10896a0 offset 0
size 1000 count 200
Mon Aug 11 10:47:53 2003 : (32) file_read_actor: exit - size 200
Mon Aug 11 10:47:53 2003 : (32) do_generic_file_read: called actor - ret
200 off
set 0 index 0
Mon Aug 11 10:47:53 2003 : (32) do_generic_file_read: new offset 200 index 0
Mon Aug 11 10:47:53 2003 : (32) do_generic_file_read: exit - ppos 200 index
0 of
fset 0
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: entry - inode c277d480
*ppos
0 i_rdev 1602 i_blkbits a
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: pos 0 limit 0 count
ffffffff
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: entry inode c277d480
i_bl
kbits a block 0
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: exit - err 400 1024.
#### mke2fs makes call to zero_blocks()
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: rsv 10 blocks 9c648 start 9c640
Mon Aug 11 10:47:53 2003 : (mke2fs) Main: calling zero_blocks - start 9c630
bloc
ks 9c648 blocks-start 18
Mon Aug 11 10:47:53 2003 : (mke2fs) zero_blocks: entry blk 9c630 num 18
Mon Aug 11 10:47:53 2003 : (mke2fs) zero_blocks: blk 9c630 (640560.)
Mon Aug 11 10:47:53 2003 : (32) alloc_inode: inode c277d280 sb c1aa0400 s_dev 1
s_blocksize_bits a
Mon Aug 11 10:47:53 2003 : (32) bdget: inode c277d280 old i_dev 1 new
i_dev/i_rd
ev c800 bdev 0
Mon Aug 11 10:47:53 2003 : (32) alloc_inode: inode c277d280 sb c1aa0400 s_dev 1
s_blocksize_bits a
Mon Aug 11 10:47:53 2003 : (32) bdget: inode c277d280 old i_dev 1 new
i_dev/i_rd
ev 1600 bdev 0
Mon Aug 11 10:47:53 2003 : (32) SunDskOpen: entry - Inode cdad3980 Filep
cd13c94
0
Mon Aug 11 10:47:53 2003 : (32) SunDskOpen: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c1b9 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c1b9 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c1b9 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err 6c 108.
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: entry - Inode cdad3980 Filep
cd13c9
40 Command 534400 Argument 8055ced
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: SunDebugPrintForceToBridge
turned O
N
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: SunDebugPrintForceToBridge
turned O
N
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c225 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c225 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c225 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err cf 207.
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: entry - Inode cdad3980 Filep
cd13c9
40 Command 534400 Argument bfffe390
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: SunDebugLevel set to ffffffff
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: SunDebugLevel set to ffffffff
Mon Aug 11 10:47:53 2003 : (32) SunDskIoctl: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c2f4 i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c2f4 limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c2f4 s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err c9 201.
Mon Aug 11 10:47:53 2003 : (32) SunDskOpen: entry - Inode c277d280 Filep 0
Mon Aug 11 10:47:53 2003 : (32) SunDskRelease: i_rdev 1600 MINOR 0
Mon Aug 11 10:47:53 2003 : (32) SunDskRelease: diskNum 0 partitionNum 0
Mon Aug 11 10:47:53 2003 : (32) SunDskRelease: exit - status 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: entry - inode cf7cb340
*ppos
5c3bd i_rdev 0 i_blkbits a
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c3bd limit 0 count
ffff
ffff
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: pos 5c3bd s_maxbytes 0
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: entry inode cf7cb340
i_bl
kbits a block 170
Mon Aug 11 10:47:53 2003 : (10) __block_prepare_write: exit
Mon Aug 11 10:47:53 2003 : (10) generic_file_write: exit - err d1 209.
#### pager works on mke2fs/zero_blocks() write request with inode 0xc277d480
#### However, inode->i_blkbits == 0xa (0x400 blocksize), but the
#### blocksize for 0x1602 is really 0x1000 (i_blkbits should be 0xc)
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: entry - inode c277d480
*ppos
9c630000 i_rdev 1602 i_blkbits a
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: pos 9c630000 limit 0
count f
fffffff
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: ready for write
Mon Aug 11 10:47:53 2003 : (32) __add_to_page_cache: page c10894e0 offset 9c630
#### __block_prepare_write() (./fs/block_dev.c) sets up the write by first
#### "getting" the block. It computes the block number using inode->i_blkbits,
#### which is essentially 0x400. Note that in this case, it is asking to
#### write the "last" block of the partition.
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: entry inode c277d480
i_bl
kbits a block 2718c0
#### blkdev_get_block() (./fs/block_dev.c) is invoked to get the "last" block
#### 0x2718c0, based on a blocksize of 0x400. First thing that it does
#### is a check to see if we exceed the partition size by getting the
#### size of the partition via max_block(). max_block() (./fs/block_dev.c)
#### uses the blksize_size[] global arrays which says that the blocksize
#### is 0x1000. Thus, the result that max_block() returns is in terms of
#### 0x1000 blocksized blocks, which is numerically less than the block
#### index passed into blkdev_get_block() as the "last" block. Thus,
#### blkdev_get_block() returns an error
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: rdev 1602
inode->i_blkbits a b
h ccc8e740 iblock 2718c0
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) max_block: entry - dev 1602
Mon Aug 11 10:47:53 2003 : (32) max_block: size 1000 blocks 271923
BLOCK_SIZE_BI
TS a sizebits c retval c0000000
Mon Aug 11 10:47:53 2003 : (32) max_block: exit - retval 9c648
Mon Aug 11 10:47:53 2003 : (32) blkdev_get_block: max_block(inode->i_rdev)
9c648
failure
Mon Aug 11 10:47:53 2003 : (32) __block_prepare_write: get_block failure -
err 2
718c0 block ccc8e740
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: prepare_write failure -
stat
us fffffffb
Mon Aug 11 10:47:53 2003 : (32) generic_file_write: exit - err fffffffb -5.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: possible auto-partition bug? (linux-2.4.20-8)
2003-08-15 17:55 possible auto-partition bug? (linux-2.4.20-8) Pete Nishimoto
@ 2003-08-15 20:23 ` Andries Brouwer
2003-08-16 2:15 ` jw schultz
0 siblings, 1 reply; 3+ messages in thread
From: Andries Brouwer @ 2003-08-15 20:23 UTC (permalink / raw)
To: Pete Nishimoto; +Cc: andrea, linux-kernel, Peten
On Fri, Aug 15, 2003 at 01:55:59PM -0400, Pete Nishimoto wrote:
> My name is Pete Nishimoto and I work for Sun Microsystems
> as a linux device driver developer, currently working with
> RedHat 9.0 (2.4.20-8) and I believe I have found a problem
> with the partitioning logic and the pager, which I've
> detailed below. I look forward to any replies/comments
> and thanks in advance to all who review/read this.
Hi. You sent a long story, but at first sight it seems not relevant
for this linux-kernel mailing list.
A disk is made by a manufacturer, and has a number of sectors that we
must regard as given. If a filesystem is created on this disk then
often the disk size will turn out not to be precisely an integral
number of filesystem blocks.
Many people first partition the disk in some more or less arbitrary way.
Partitions may belong to other operating systems. Again we have no control.
In short, absolutely nothing is wrong if a disk, or a partition, has a size
that is not an integral number of filesystem blocks.
You talk about badblocks, but that is a userspace utility. If something
is wrong with it, that is not a kernel matter. Moreover, this utility
allows one to specify blocksize and last block to test.
So - the relevance to the kernel is not clear to me.
Concerning "RedHat 9.0 (2.4.20-8)" - discussion about vendor specific kernels
is probably best done on vendor lists.
Andries
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: possible auto-partition bug? (linux-2.4.20-8)
2003-08-15 20:23 ` Andries Brouwer
@ 2003-08-16 2:15 ` jw schultz
0 siblings, 0 replies; 3+ messages in thread
From: jw schultz @ 2003-08-16 2:15 UTC (permalink / raw)
To: linux-kernel
On Fri, Aug 15, 2003 at 10:23:04PM +0200, Andries Brouwer wrote:
> On Fri, Aug 15, 2003 at 01:55:59PM -0400, Pete Nishimoto wrote:
>
> > My name is Pete Nishimoto and I work for Sun Microsystems
> > as a linux device driver developer, currently working with
> > RedHat 9.0 (2.4.20-8) and I believe I have found a problem
> > with the partitioning logic and the pager, which I've
> > detailed below. I look forward to any replies/comments
> > and thanks in advance to all who review/read this.
>
> Hi. You sent a long story, but at first sight it seems not relevant
> for this linux-kernel mailing list.
>
> A disk is made by a manufacturer, and has a number of sectors that we
> must regard as given. If a filesystem is created on this disk then
> often the disk size will turn out not to be precisely an integral
> number of filesystem blocks.
>
> Many people first partition the disk in some more or less arbitrary way.
> Partitions may belong to other operating systems. Again we have no control.
>
> In short, absolutely nothing is wrong if a disk, or a partition, has a size
> that is not an integral number of filesystem blocks.
>
> You talk about badblocks, but that is a userspace utility. If something
> is wrong with it, that is not a kernel matter. Moreover, this utility
> allows one to specify blocksize and last block to test.
>
> So - the relevance to the kernel is not clear to me.
>
> Concerning "RedHat 9.0 (2.4.20-8)" - discussion about vendor specific kernels
> is probably best done on vendor lists.
I can see no relevance to the kernel here.
With the advent of zone recording only sector size, total
sector count and head count have any meaning and head count
is often false so that other geometry numbers will fit into
specified field sizes. You cannot fit 255 heads in a half
height 3.5" drive. Therefore cylinder boundaries are
illusory and should be ignored when partitioning if the
partition table allows LBA addressing.
Partition definition is not a kernel function. It is user space.
badblocks is, as Andries reports, user space, not kernel.
If you read the manpage for badblocks you will see
-b block-size
Specify the size of blocks in bytes.
So even badblocks doesn't need changing to work with
partitions having (sector_count % 8 != 0).
This sounds like an install script problem to me.
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: jw@pegasys.ws
Remember Cernan and Schmitt
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2003-08-16 2:15 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-08-15 17:55 possible auto-partition bug? (linux-2.4.20-8) Pete Nishimoto
2003-08-15 20:23 ` Andries Brouwer
2003-08-16 2:15 ` jw schultz
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).