Weird partititon recocnising problem in 2.6.0-testX

Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

Hi!

I have a PC here with an intel i875p chipset which seems not to recognize
logical partititions. 2.4.18 (kernel bf24 image) boots fine, but i do
not get a 2.6.0-test5 or 2.6.0-test9 to boot.

I compiled all necessary IDE drivers in, IDE, ide disk and piix and
such. All attempts yield in

Kernel Panic: VFS: Unable to mount root fs on hda5

I tried to submit root=0305 and such to pass to it. On my later
investigations I realized that the partition recocnizion seems to fail.

When booting Kernels report

hda: hda1 hda2 <hda5 hda6>

and boot, this one reports only:

hda: hda hda2

So, where are the logical ones?

In "Advanced partition selection" I switched on 
"PC BIOS (MSDOS partition tables) support" but that made no
difference...

hm. what next?

Regards, Konsti

-- 
2.6.0-test6-mm4
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 4:24, 1 user

Re: Weird partititon recocnising problem in 2.6.0-testX

[Andries Brouwer]

On Sun, Nov 09, 2003 at 02:12:05AM +0100, Konstantin Kletschke wrote:

> I have a PC here with an intel i875p chipset which seems not to recognize
> logical partititions. 2.4.18 (kernel bf24 image) boots fine, but i do
> not get a 2.6.0-test5 or 2.6.0-test9 to boot.
> 
> When booting Kernels report
> 
> hda: hda1 hda2 <hda5 hda6>
> 
> and boot, this one reports only:
> 
> hda: hda hda2

I suppose that second hda is a typo for hda1?

What partition table? (fdisk -l /dev/hda or sfdisk -l -x -uS /dev/hda)

Re: Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

* Andries Brouwer <aebr@win.tue.nl> [Sun, Nov 09, 2003 at 03:36:25AM +0100]:

> > hda: hda hda2
> 
> I suppose that second hda is a typo for hda1?

Yes ;)

> What partition table? (fdisk -l /dev/hda or sfdisk -l -x -uS /dev/hda)

	 Disk /dev/hda: 255 heads, 63 sectors, 1245 cylinders
Units = cylinders of 16065 * 512 bytes

   Device Boot    Start       End    Blocks   Id  System
/dev/hda1   *         1       365   2931831   83  Linux
/dev/hda2           366      1245   7068600    5  Extended
/dev/hda5           366       487    979933+  83  Linux
/dev/hda6          1185      1245    489951   82  Linux swap

Disk /dev/hdb: 255 heads, 63 sectors, 16709 cylinders
Units = cylinders of 16065 * 512 bytes

   Device Boot    Start       End    Blocks   Id  System
/dev/hdb1             1     16709 134215011   83  Linux


Thats it :) 

.config is at
http://ludenkalle.de/.config

Konsti

Re: Weird partititon recocnising problem in 2.6.0-testX

[Andries Brouwer]

On Sun, Nov 09, 2003 at 04:49:40AM +0100, Konstantin Kletschke wrote:
> * Andries Brouwer <aebr@win.tue.nl> [Sun, Nov 09, 2003 at 03:36:25AM +0100]:
> 
> > > hda: hda hda2
> > 
> > I suppose that second hda is a typo for hda1?
> 
> Yes ;)
> 
> > What partition table? (fdisk -l /dev/hda or sfdisk -l -x -uS /dev/hda)
> 
> 	 Disk /dev/hda: 255 heads, 63 sectors, 1245 cylinders
> Units = cylinders of 16065 * 512 bytes
> 
>    Device Boot    Start       End    Blocks   Id  System
> /dev/hda1   *         1       365   2931831   83  Linux
> /dev/hda2           366      1245   7068600    5  Extended
> /dev/hda5           366       487    979933+  83  Linux
> /dev/hda6          1185      1245    489951   82  Linux swap

Hmm. msdos.c has

                        put_partition(state, slot, start, size == 1 ? 1 : 2);
                        printk(" <");
                        parse_extended(state, bdev, start, size);
                        printk(" >");

The "hda2" is printed by put_partition(). But no " <" is printed.
An impossible error. Recompile your kernel.

Andries


[If a clean compile still fails, add
	printk("partition start %u size %u type %u",
		start, size, SYS_IND(p))
after the
        for (slot = 1 ; slot <= 4 ; slot++, p++) {
                u32 start = START_SECT(p)*sector_size;
                u32 size = NR_SECTS(p)*sector_size;
around line 413 in fs/partitions/msdos.c.]

Re: Weird partititon recocnising problem in 2.6.0-testX

[Stefan Smietanowski]

Hi Andries.

> Hmm. msdos.c has
> 
>                         put_partition(state, slot, start, size == 1 ? 1 : 2);
>                         printk(" <");
>                         parse_extended(state, bdev, start, size);
>                         printk(" >");
> 
> The "hda2" is printed by put_partition(). But no " <" is printed.
> An impossible error. Recompile your kernel.

I wouldn't say it's impossible. If the partition isn't recognized
as extended for some reason that code won't be called, will it?
And it'll only print "hda2" for him then.

// Stefan

Re: Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

* Stefan Smietanowski <stesmi@stesmi.com> [Sun, Nov 09, 2003 at 01:10:41PM +0100]:
> >An impossible error. Recompile your kernel.

We recompiled it from fresh unpacked tar.bz2 same error. Instead of
gcc-2.95.4 we tried gcc-3.3.2 same errot. I got serial console running,
so a detailed log is at

http://www.ludenkalle.de/jan.log

My friend is trying to add this printk stuff (his PC) lets see...

Konsti

-- 
2.6.0-test6-mm4
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 4:58, 1 user

Re: Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

Hi again!

The suggested printk code is added no but I see no difference:

http://www.ludenkalle.de/printk.log


Regards, Konsti


-- 
2.6.0-test6-mm4
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 5:54, 1 user

Re: Weird partititon recocnising problem in 2.6.0-testX

[Andries Brouwer]

On Mon, Nov 10, 2003 at 12:09:40AM +0100, Konstantin Kletschke wrote:

> The suggested printk code is added no but I see no difference:

  hda: 9766MB, CHS=1245/255/63
  hdb: 130551MB, CHS=16643/255/63
  devfs_mk_dir: invalid argument.
   hda: hda1 hda2
  devfs_mk_dir: invalid argument.
  devfs_mk_bdev: could not append to parent for /disc
   hdb: hdb1
  devfs_mk_bdev: could not append to parent for /part1
  
Aha, this "hda: 9766MB, CHS=1245/255/63" looks ancient.
You are using the old hd.c instead of the new ide stuff.

Nobody else does so, so there might easily be something wrong.
So, try to set CONFIG_BLK_DEV_IDEDISK=y but unset CONFIG_BLK_DEV_HD_IDE.

Andries

(2.4 question) open_namei hangs

[Shan Sinha]

Hi everyone-

I know everyone is probably focused on 2.6 right now, but I hope someone
will have the time to answer this question about 2.4.20! I'm extremely
confused, and am not sure what work-around I could employ.

I am trying to read the contents of a file from a kernel thread, spawned
when an LKM is insmod'ed. open_namei appears to hang.  The LKM is
something that I created.  When I call filp_open("/mydirectory",
O_RDONLY, 0), and then subsequently call vfs_readdir on the returned
file *, it traverses the files in the directory correctly.  

However, in the call back passed to vfs_readdir, if I try to do a
filp_open("/mydirectory/file1"), where "file1" is the string passed to
the call back, filp_open appears to hang somewhere inside open_namei (I
verified this by modifying filp_open).

But!  if I do "touch /mydirectory/*" in user space before I insert the
module and spawn my thread, the open and subsequent read for any files
in /mydirectory work fine.  

I suspect this is something to do with the files being loaded into the
buffer cache and my doing something incorrectly with respect to being in
a kernel thread.  Does anyone have any ideas of what may be happening?

I have included snips of the relevant code below.  I guessing my thread
is getting deadlocked somehow when the kernel has to go to disk to fetch
the file.  However, I created the code using khttpd as a sample (not
sure if this was a mistake).

Cheers-
Shan Sinha
Network and Mobile Systems
MIT CSAIL
ssinha@nms.mit.edu

------------------------------------------------------------
int my_callback(void * buf, const char * name, int namlen, loff_t
offset,
	     ino_t ino, unsigned int d_type)
{

       // for each file
  if (d_type == DT_REG) {
    struct file * f;
    read_descriptor_t read_desc;

    char * file_name = make_file_name("/mydirectory/", name, namlen);

    printk(KERN_INFO "loading file: %s\n", file_name);

    // WE HANG IN HERE (IF FILES HAVE NOT BEEN LOADED INTO BUFFER
CACHE??)!!!
    f = filp_open(file_name, O_RDONLY, 0);

    printk(KERN_INFO "opened file: %s\n", file_name);

    // Code snipped - set up code for call to do_generic_file_read
    // Code snipped - call to do_generic_file_read

    kfree(file_name);

  }

  return 0;
}

// Main daemon thread
int pm_daemon(void * data)
{

  DECLARE_WAIT_QUEUE_HEAD(WQ);

  sigset_t tmpsig;
  daemon_running = 1;
  leave_daemon_running = 1;
  sprintf(current->comm, "mythread");
  daemonize();

  /* Code snipped that blocks all signals except SIGKILL and SIGTERM */

  /* main loop */
  while (leave_daemon_running) {
    int signal_counter = 0;


    // count the number of files
    int num_files = 0;

    struct file * f = filp_open("/mydirectory/", O_RDONLY, 0);
    vfs_readdir(f, file_counter, &num_files);
 
    // loop through all files
    fput(f);
    printk(KERN_INFO "Counted files: %d\n", num_files );

    f = filp_open("/mydirectory/", O_RDONLY, 0);
    vfs_readdir(f, my_callback, NULL);

    // code snipped sleep and respond to signals
  }
  leave_daemon_running = 0;
  daemon_running = 0;

  return 0;
}



// called on module init
void pm_init(void)
{
   
  if (!daemon_running)
    kernel_thread(pm_daemon, NULL, CLONE_FS | CLONE_FILES |
CLONE_SIGHAND);

}

Re: Weird partititon recocnising problem in 2.6.0-testX

[Rogier Wolff]

On Mon, Nov 10, 2003 at 05:02:36PM +0100, Andries.Brouwer@cwi.nl wrote:
> 
> Boot a kernel that remaps, say vanilla 2.4.

[...]

> Now boot a kernel that does not remap. Since you won't see

[...]

All this rebooting into kernels that DO or DON'T remap would have been
unneccesary if the remapping would not have cloned block 1 to appear
in block 0, but would just have swapped them. 

In remapped mode, you now read block 1 wether you seek to block 0 or 1. 
If the remapping would have swapped them, just copying over the block
from block 1 to block 0 would have allowed the removal of the stupid
remapping. 

ide.c:1381
        /* Yecch - this will shift the entire interval,
           possibly killing some innocent following sector */
        if (block == 0 && drive->remap_0_to_1 == 1)
             block = 1;  /* redirect MBR access to EZ-Drive partn table */

should have been:

	if ((block <= 1)  && drive->remap_0_to_1) 
		block ^= 1;

(IMHO, it's not worth going in and fixing this for say 2.4 or 2.6
kernels(*), but it IS worth noting and not making the same mistake again in
the future.....)

	Roger. 

(*) For example because that would change established behaviour which is
bad and can lead to surprises. If we'd change it, Andries' "how to fix"
emails would become even longer: He'd have to start: IF you have a
modern kernel, you can just ... but otherwise you'll have to .... and
then comes the whole mess I didn't fully quote.....

-- 
** R.E.Wolff@BitWizard.nl ** http://www.BitWizard.nl/ ** +31-15-2600998 **
*-- BitWizard writes Linux device drivers for any device you may have! --*
**** "Linux is like a wigwam -  no windows, no gates, apache inside!" ****

Re: Weird partititon recocnising problem in 2.6.0-testX

[Andries.Brouwer]

>> I suppose that booting with boot parameter "hda=remap" should work.

> Seems so...

Good. So there were two problems. One was devfs+hd.c, caused by
the fact that hd.c fails to set devfs_name. And the second was
EZDrive, solved by booting with "hda=remap".

Probably it is a good idea to warn when we see EZD or DM.
A minimal patch would be

--- 2.6.0test9/linux/fs/partitions/msdos.c	Fri May 30 18:12:57 2003
+++ fs/partitions/msdos.c	Mon Nov 10 16:20:11 2003
@@ -425,6 +425,10 @@
 		put_partition(state, slot, start, size);
 		if (SYS_IND(p) == LINUX_RAID_PARTITION)
 			state->parts[slot].flags = 1;
+		if (SYS_IND(p) == DM6_PARTITION)
+			printk("[DM]");
+		if (SYS_IND(p) == EZD_PARTITION)
+			printk("[EZD]");
 	}
 
 	printk("\n");

The next project for you is getting rid of EZD (if you want).
Backup valuable stuff.

Boot a kernel that remaps, say vanilla 2.4.
Print partition table (in sector units) on paper.
  sfdisk -d /dev/hda > hda.pt
gives the full table in a format that sfdisk can restore.
  dd if=/dev/hda of=hdapt bs=1 count=64 skip=446
gives the primary part of the table in binary.
  dd if=/dev/hda of=hdambr bs=512 count=1
gives the entire fake sector 0 (that is, sector 1).
Save the output files somewhere not on hda.

Now boot a kernel that does not remap. Since you won't see
your root filesystem on hda, this should probably be from
a rescue floppy or CD. Recent kernels understand the "hda=noremap"
boot parameter. Make sure no remap is done (e.g., *fdisk
mentions partitions of type 55). Now write the desired partition
table (e.g.
  dd if=hdapt of=/dev/hda bs=1 count=64 seek=446
or
  dd if=hdambr of=/dev/hda
or
  sfdisk /dev/hda < hda.pt
).

That should do it, but be careful that you can boot again afterwards.
Sector 0 has the partition table in bytes 446-509,
a signature in bytes 510-511, and a boot loader in bytes 0-445.
You may have to reinstall LILO or grub or so. In some cases
copying the entire sector 1 to sector 0 will suffice.

Good luck!
Andries

Re: Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

On Mon, Nov 10, 2003 at 11:56:50AM +0100, Andries Brouwer wrote:
> 
> I suppose that booting with boot parameter "hda=remap" should work.

Seems so...
http://ludenkalle.de/2.6.0-test6-mm4-remap.txt

Konsti

-- 
2.6.0-test8-love3
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 2:35, 19 users

Re: Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

On Mon, Nov 10, 2003 at 11:56:50AM +0100, Andries Brouwer wrote:
> > 
> > With CONFIG_BLK_DEV_IDEDISK=y but unset CONFIG_BLK_DEV_HD_IDE and the
> > inserted (and no used) printk stuff.
> 
> At some later time we must come back and find out what is
> wrong with hd.c. You call the printk stuff unused, but
> it was used and printed

It was unused earlier, now, with CONFIG_BLK_DEV_IDEDISK=y but _unset_
CONFIG_BLK_DEV_HD_IDE it gets printed (to clear that).

> Clearly, you have EZDrive installed, the table below is what
> is found in sector 1, the data printed above is what is in
> sector 0. The tables differ - fdisk was used after installation
> of EZDrive.

Hm, interesting... I don't know what was on this disk in earlier days...
So it may be a help dd'ing Sektor 0 and 1 and create new table?

But of course after debugging that if you are interesting in that (which
seems to be the case here :)).

Here is what 2.6.0-test6-mm4 looks like when booting (wo printk mod):

http://ludenkalle.de/2.6.0-test6-mm4.txt

> I suppose that booting with boot parameter "hda=remap" should work.
> At some later time we must worry about how to get rid of EZDrive.

OK, lets see...

Konsti

-- 
2.6.0-test8-love3
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 2:00, 18 users

Re: Weird partititon recocnising problem in 2.6.0-testX

[Andries Brouwer]

On Mon, Nov 10, 2003 at 11:24:44AM +0100, Konstantin Kletschke wrote:
> We have a new boot log at
> 
> http://ludenkalle.de/ide_change_kernel.txt
> 
> With CONFIG_BLK_DEV_IDEDISK=y but unset CONFIG_BLK_DEV_HD_IDE and the
> inserted (and no used) printk stuff.

Good. Now it really does partition reading.
At some later time we must come back and find out what is
wrong with hd.c. You call the printk stuff unused, but
it was used and printed

partition start 63 size 4096512 type 85 p1
partition start 4096575 size 15904350 type 85 p2
partition start 0 size 0 type 0
partition start 0 size 0 type 0                          

Let us compare with your data from earlier:

   Device Boot    Start       End    Blocks   Id  System
/dev/hda1   *         1       365   2931831   83  Linux
/dev/hda2           366      1245   7068600    5  Extended
/dev/hda5           366       487    979933+  83  Linux
/dev/hda6          1185      1245    489951   82  Linux swap

Clearly, you have EZDrive installed, the table below is what
is found in sector 1, the data printed above is what is in
sector 0. The tables differ - fdisk was used after installation
of EZDrive.

I suppose that booting with boot parameter "hda=remap" should work.
At some later time we must worry about how to get rid of EZDrive.

Andries

Weird partititon recocnising problem in 2.6.0-testX

[Konstantin Kletschke]

We have a new boot log at

http://ludenkalle.de/ide_change_kernel.txt

With CONFIG_BLK_DEV_IDEDISK=y but unset CONFIG_BLK_DEV_HD_IDE and the
inserted (and no used) printk stuff.

Regards, Konsti

-- 
2.6.0-test8-love3
Konstantin Kletschke <konsti@ludenkalle.de>, <konsti@ku-gbr.de>
GPG KeyID EF62FCEF
Fingerprint: 13C9 B16B 9844 EC15 CC2E  A080 1E69 3FDA EF62 FCEF
keulator.homelinux.org up 1:01, 13 users