Skip to contentSkip to conversation

Define conflict between ext3 and raid patches against 2.2.19

DISCUSSION Mike Fedyk 5 participants 9 messages Sep 16, 2001 – Oct 1, 2001

lore
Share & tools
lore

Mike Fedyk

author
#0 lore raw

Hi,

I'm trying to setup a 2.2 kernel that I can use for comparison to the latest 2.4 kernels I've been testing, but I came accross a little problem with the patches I've been trying to combine.

I've already applied:
ide.2.2.19.05042001.patch
linux-2.2.19.kdb.diff
linux-2.2.19.ext3.diff

And now I'm trying to apply raid-2.2.19-A1, and I get one reject in include/linux/fs.h.

***************
*** 191,197 ****
#define BH_Req 3 /* 0 if the buffer has been invalidated */
#define BH_Protected 6 /* 1 if the buffer is protected */
#define BH_Wait_IO 7 /* 1 if we should throttle on this buffer */

/*
* Try to keep the most commonly used fields in single cache lines (16
--- 191,196 ----
#define BH_Req 3 /* 0 if the buffer has been invalidated */
#define BH_Protected 6 /* 1 if the buffer is protected */
#define BH_Wait_IO 7 /* 1 if we should throttle on this buffer */
+ #define BH_LowPrio 8 /* 1 if the buffer is lowprio */

/*
* Try to keep the most commonly used fields in single cache lines (16

Now I have two defines from different patches for the same bit. The top line is from RAID, and bottom is of ext3 origin.

#define BH_LowPrio 8 /* 1 if the buffer is lowprio */
#define BH_Temp 8 /* 1 if the buffer is temporary (unlinked)

Is this a fatal conflict? How can I resolve this?

Mike

#1 lore raw
Quoted 2 lines · Mike Fedyk

#define BH_LowPrio 8 /* 1 if the buffer is lowprio */
#define BH_Temp 8 /* 1 if the buffer is temporary (unlinked)

Change BH_Temp to:

#define BH_Temp 9 /* 1 if the buffer is temporary (unlinked)

and it should work.

Signature · 7 lines
      -----==-                                             |
      ----==-- _                                           |
      ---==---(_)__  __ ____  __       Marc Lehmann      +--
      --==---/ / _ \/ // /\ \/ /       [email protected]      |e|
      -=====/_/_//_/\_,_/ /_/\_\       XX11-RIPE         --+
    The choice of a GNU generation                       |
                                                         |
author
#2 lore raw
In reply to Mike Fedyk
Quoted 10 lines · Marc A. Lehmann
Quoted 2 lines · Mike Fedyk

#define BH_LowPrio 8 /* 1 if the buffer is lowprio */
#define BH_Temp 8 /* 1 if the buffer is temporary (unlinked)

Change BH_Temp to:

#define BH_Temp 9 /* 1 if the buffer is temporary (unlinked)

and it should work.

/* bh state bits */

#define BH_Uptodate     0       /* 1 if the buffer contains valid data */
#define BH_Dirty        1       /* 1 if the buffer is dirty */
#define BH_Lock         2       /* 1 if the buffer is locked */
#define BH_Req          3       /* 0 if the buffer has been invalidated */
#define BH_Protected    6       /* 1 if the buffer is protected */
#define BH_Wait_IO      7       /* 1 if we should throttle on this buffer */
#define BH_Temp         8       /* 1 if the buffer is temporary (unlinked) */
#define BH_JWrite       9       /* 1 if being written to log (@@@ DEBUGGING)*/
#define BH_QuickFree    10      /* 1 if alloced and freed quickly (see below)*/
#define BH_Alloced      11      /* 1 if buffer has been allocated */
#define BH_Freed        12      /* 1 if buffer has been freed (truncated)*/
#define BH_Revoked      13      /* 1 if buffer has been revoked from the log*/
#define BH_RevokeValid  14      /* 1 if buffer revoked flag is valid */
#define BH_JDirty       15      /* 1 if buffer is dirty but journaled */

As you can see, that is already taken from ext3. Is this ok?

#define BH_LowPrio 16 /* 1 if the buffer is temporary (unlinked)

Or do I only have 16 bits to work with?

TIA,

Mike

#3 lore raw
In reply to Mike Fedyk
Quoted 23 lines · Mike Fedyk
Quoted 6 lines

Change BH_Temp to:

#define BH_Temp 9 /* 1 if the buffer is temporary (unlinked)

and it should work.

/* bh state bits */

#define BH_Uptodate     0       /* 1 if the buffer contains valid data */
#define BH_Dirty        1       /* 1 if the buffer is dirty */
#define BH_Lock         2       /* 1 if the buffer is locked */
#define BH_Req          3       /* 0 if the buffer has been invalidated */
#define BH_Protected    6       /* 1 if the buffer is protected */
#define BH_Wait_IO      7       /* 1 if we should throttle on this buffer */
#define BH_Temp         8       /* 1 if the buffer is temporary (unlinked) */
#define BH_JWrite       9       /* 1 if being written to log (@@@ DEBUGGING)*/
#define BH_QuickFree    10      /* 1 if alloced and freed quickly (see below)*/
#define BH_Alloced      11      /* 1 if buffer has been allocated */
#define BH_Freed        12      /* 1 if buffer has been freed (truncated)*/
#define BH_Revoked      13      /* 1 if buffer has been revoked from the log*/
#define BH_RevokeValid  14      /* 1 if buffer revoked flag is valid */
#define BH_JDirty       15      /* 1 if buffer is dirty but journaled */
Quoted 5 lines

As you can see, that is already taken from ext3. Is this ok?

#define BH_LowPrio 16 /* 1 if the buffer is temporary (unlinked)

Or do I only have 16 bits to work with?

No, you have 32 bits to work with. In my ext3-2.2 patches I just increased all of the ext3 numbers by one. I don't think it is critical what number is used in the end, as long as you don't get another patch defining these flags (reiserfs uses these flags in 2.2 as well).

Cheers, Andreas

Signature · 3 lines
Andreas Dilger  \ "If a man ate a pound of pasta and a pound of antipasto,
                 \  would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/               -- Dogbert
author
#4 lore raw
In reply to Mike Fedyk
Quoted 15 lines · Neil Brown
Quoted 6 lines · On Sunday September 16, [email protected] wrote:

Hi,

I'm trying to setup a 2.2 kernel that I can use for comparison to the latest 2.4 kernels I've been testing, but I came accross a little problem with the patches I've been trying to combine.

Quoted 7 lines

I've already applied:
ide.2.2.19.05042001.patch
linux-2.2.19.kdb.diff
linux-2.2.19.ext3.diff

And now I'm trying to apply raid-2.2.19-A1, and I get one reject in include/linux/fs.h.

Quoted 7 lines

You should be aware that ext3 (and other journalling filesystems) do not work reliably over RAID1 or RAID5 in 2.2. Inparticular, you can get problems when the array is rebuilding/resyncing.

But if you only plan to use ext3 with raid0 or linear, you should be fine.

Crap.

I was just about to test ext3 on a raid1 with 2.2.20pre. Everything else I've tried has worked great.

As it looks, this probably won't be fixed in 2.2, especially since the raid patch won't be merged into 2.2 (as stated by Alan).

I'll just have to test 2.4 harder now. 2.4.x-ac has been pretty good concerning swap usage, and performance on my workstations. I hope 2.4.10 will merge those improvements.

Mike

author
#5 lore raw
In reply to Mike Fedyk
Quoted 15 lines · Neil Brown
Quoted 6 lines · On Sunday September 16, [email protected] wrote:

Hi,

I'm trying to setup a 2.2 kernel that I can use for comparison to the latest 2.4 kernels I've been testing, but I came accross a little problem with the patches I've been trying to combine.

Quoted 7 lines

I've already applied:
ide.2.2.19.05042001.patch
linux-2.2.19.kdb.diff
linux-2.2.19.ext3.diff

And now I'm trying to apply raid-2.2.19-A1, and I get one reject in include/linux/fs.h.

Quoted 7 lines

You should be aware that ext3 (and other journalling filesystems) do not work reliably over RAID1 or RAID5 in 2.2. Inparticular, you can get problems when the array is rebuilding/resyncing.

But if you only plan to use ext3 with raid0 or linear, you should be fine.

Can you point me to an archive that describes how to trigger this bug?

Was it in linux-raid or ext3-users or ...?

Mike

#6 lore raw
In reply to Mike Fedyk

Hi,

Quoted 8 lines · Neil Brown

However when a RAID rebuild happens, every block on the array is read into the buffer cache (if it isn't already there) and then written back out again. This defeats the control that ext3 tries to maintain on the buffer cache.

I don't know exactly what large-scale effects this might have. It could be simply that a crash at the wrong time could leave the filesystem corrupted.

Immediately after a crash, the fs will be OK. But during the subsequent background raid reconstruction, it can get out of sync again. This can result in silent data loss in some cases, but it is also likely to trigger some internal ext3 debugging which detects out-of-order data writes, resulting in a kernel panic.

Cheers,
Stephen

author
#7 lore raw
In reply to Mike Fedyk
Quoted 13 lines · Stephen C. Tweedie

Hi,

Quoted 8 lines · Neil Brown

However when a RAID rebuild happens, every block on the array is read into the buffer cache (if it isn't already there) and then written back out again. This defeats the control that ext3 tries to maintain on the buffer cache.

I don't know exactly what large-scale effects this might have. It could be simply that a crash at the wrong time could leave the filesystem corrupted.

Quoted 6 lines

Immediately after a crash, the fs will be OK. But during the subsequent background raid reconstruction, it can get out of sync again. This can result in silent data loss in some cases, but it is also likely to trigger some internal ext3 debugging which detects out-of-order data writes, resulting in a kernel panic.

Ok, thanks guys, on ext3-users and lkml.

Now I know that if I want RAID1 or RAID5 with ext3, I'm going to need 2.4...

Test test test...

Mike

#8 lore raw
In reply to Mike Fedyk

More precisely, as long as you understand that resync requires doing the resync offline, and
accept that, and remember it, and your colleagues at work remember it:-), you can use it. Personally, I would just go to 2.4 myself.

Hans

Quoted 44 lines · Neil Brown
Quoted 10 lines · On Wednesday September 19, [email protected] wrote:
Quoted 8 lines · Neil Brown

You should be aware that ext3 (and other journalling filesystems) do not work reliably over RAID1 or RAID5 in 2.2. Inparticular, you can get problems when the array is rebuilding/resyncing.

But if you only plan to use ext3 with raid0 or linear, you should be fine.

Quoted 5 lines

Can you point me to an archive that describes how to trigger this bug?

Was it in linux-raid or ext3-users or ...?

Mike

I don't remember exactly where or when I read it - either linux-raid or linux-kernel. It was asserted by Stephen Tweedie.

The problem is that ext3 is very careful about when it writes buffer to disk : it won't release a buffer until the relevant journal entry is committed.

However when a RAID rebuild happens, every block on the array is read into the buffer cache (if it isn't already there) and then written back out again. This defeats the control that ext3 tries to maintain on the buffer cache.

I don't know exactly what large-scale effects this might have. It could be simply that a crash at the wrong time could leave the filesystem corrupted. But I heard of one person who claimed to get filesystem corruption after using reiserfs over RAID1 in 2.2, so maybe it's worse than that.

If you really need to know, I suggest you ask on ext3-users.

NeilBrown
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/