Re: [PATCH][CFT] (updated) ext2 directories in pagecache

Re: [PATCH][CFT] (updated) ext2 directories in pagecache

[Andreas Dilger]

I previously wrote:
> I was looking at the new patch, and I saw something that puzzles me.
> Why do you set the EXT2_INDEX_FL on a new (empty) directory, rather
> than only setting it when the dx_root index is created?
> 
> Setting the flag earlier than that makes it mostly useless, since it
> will be set on basically every directory.  Not setting it would also
> make your is_dx() check simply a check for the EXT2_INDEX_FL bit (no
> need to also check size).
> 
> Also no need to set EXT2_COMPAT_DIR_INDEX until such a time that we have
> a (real) directory with an index, to avoid gratuitous incompatibility
> with e2fsck.

I have changed the code to do the following:
- If the COMPAT_DIR_INDEX flag is set at mount/remount time, set the
  INDEX mount option (the same as "mount -o index").  This removes
  the need to specify the "-o index" option each time for filesystems
  which already have indexed directories.
- New directories NEVER have the INDEX flag set on them.
- If the INDEX mount option is set, then when directories grow past 1
  block (and have the index added) they will get the directory INDEX
  flag set and turn on the superblock COMPAT_DIR_INDEX flag (if off).

This means that you can have common code for indexed and non-indexed ext2
filesystems, and the admin either needs to explicitly set COMPAT_DIR_INDEX
in the superblock or mount with "-o index" (and create a directory > 1 block).

I have also added some tricks to ext2_inc_count() and ext2_dec_count() so
that indexed directories are not subject to the EXT2_LINK_MAX.  I've done
the same as reiserfs, and set i_nlink = 1 if we overflow EXT2_LINK_MAX
(which has been increased to 65500 for indexed directories). Apparently
i_nlink = 1 is the right think to do w.r.t. find and other user tools.

Patches need some light testing before being posted.

Cheers, Andreas
-- 
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

Re: [PATCH][CFT] (updated) ext2 directories in pagecache

[Daniel Phillips]

On Thursday 10 May 2001 09:21, Andreas Dilger wrote:
> I previously wrote:
> > I was looking at the new patch, and I saw something that puzzles
> > me. Why do you set the EXT2_INDEX_FL on a new (empty) directory,
> > rather than only setting it when the dx_root index is created?
> >
> > Setting the flag earlier than that makes it mostly useless, since
> > it will be set on basically every directory.  Not setting it would
> > also make your is_dx() check simply a check for the EXT2_INDEX_FL
> > bit (no need to also check size).
> >
> > Also no need to set EXT2_COMPAT_DIR_INDEX until such a time that we
> > have a (real) directory with an index, to avoid gratuitous
> > incompatibility with e2fsck.

This is fixed in today's patch.

> I have changed the code to do the following:
> - If the COMPAT_DIR_INDEX flag is set at mount/remount time, set the
>   INDEX mount option (the same as "mount -o index").  This removes
>   the need to specify the "-o index" option each time for filesystems
>   which already have indexed directories.

This applied fine.  The semantics are a little loose though: after 
mounting -o index, the first directory that overflows its first block 
will turn on 'sticky indexing' for that partition.

I was originally thinking we should give the admin the ability to 
create a nonindexed directory if desired, and that's how it used to be 
before we changed the setting of INDEX_FL from directory creation time 
to later, when the first directory block overflows.

You are probably right on both counts, but it's something to think 
about.

> - New directories NEVER have the INDEX flag set on them.
> - If the INDEX mount option is set, then when directories grow past 1
>   block (and have the index added) they will get the directory INDEX
>   flag set and turn on the superblock COMPAT_DIR_INDEX flag (if off).

It works this way now.

> This means that you can have common code for indexed and non-indexed
> ext2 filesystems, and the admin either needs to explicitly set
> COMPAT_DIR_INDEX in the superblock or mount with "-o index" (and
> create a directory > 1 block).

Actually, I've had common code for indexed and non-indexed right from 
the first prototype, I hope there isn't any misunderstanding there.

> I have also added some tricks to ext2_inc_count() and
> ext2_dec_count() so that indexed directories are not subject to the
> EXT2_LINK_MAX.  I've done the same as reiserfs, and set i_nlink = 1
> if we overflow EXT2_LINK_MAX (which has been increased to 65500 for
> indexed directories). Apparently i_nlink = 1 is the right think to do
> w.r.t. find and other user tools.
>
> Patches need some light testing before being posted.

Today's Patch
-----------

The main change in today's patch is the simplified handling of the index
creation.  Besides that there are quite a few incremental improvements
and most, but not all of your recent patch set is in.

I broke out 'ext2_add_compat_feature' as a trial balloon.  It's not
really very satisfying - it's annoying to have a great piece of
machinery that's used in exactly one place.  I'd prefer to have
'ext2_add_feature' that takes the kind of compatibility as a parameter. 
We can do this by treating the three ext2_super_block compatibility
fields as an array, what do you think?  Then we'd get to use it in two
places :-]

I wrote ext2_append as a wrapper on ext2_bread, mainly to bury the
i_size handling.  Its function is to append a single block to a file.  
I pass back the block number of the appended block via a pointer, which 
is clumsy but the only other choice is to calculate the block number 
every time outside the function, which just leaves more things to get 
out of sync.

I found a real bug in the continued-hash handling - instead of advancing
to the next index block the same block would be reread.  Nobody hit this
because even with a million file directory the error condition would
exist for less than 1 in 250 million entries.  Still, correct is
correct.  Having to deal with such rare conditions is the price we have
to pay for the benefits of hashed keys.

The max link count logic in your previous patch didn't seem to be
complete so I didn't put it in.  I'll wait for more from you on that.  I
followed the ReseirFS thread on this but I'm not sure exactly what form
you want this to take in ext2.

I followed your recommendation and simplified the index creation and
INDEX_FL flag interpretation.  Now the index is created for any
directory that grows over a block, whereas before the condition was 'any
new directory created with indexing enabled that later grows over one
block'.

The index create code has been moved out of the non-indexed code path
and into the scope of the indexed path.  This is with a view to cutting
ext2_add_entry into pieces at some point, for aesthetic reasons.

The one place where I'm relying on a newly-created block to be contain
zeros got a comment.

I was unable to apply your patch all at once, and of course, I'd hacked
on the code some more in the meantime, but I went through by hand and
mined out the gold.  Please check, I'm sure I missed some.

I haven't done anything about the block checking code yet.

The patch is at:

    http://nl.linux.org/~phillips/htree/dx.pcache-2.4.4-5

As before, this is applied on top of Al's patch, available at:

   ftp://ftp.math.psu.edu/pub/viro/ext2-dir-patch-S4.gz

To apply:

    cd source/tree
    zcat ext2-dir-patch-S4.gz | patch -p1
    cat dx.pcache-2.4.4-4 | patch -p0

To create an indexed directory:

    mount /dev/hdxxx /test -o index
    mkdir /test/foo

--
Daniel

CMD64x - full DMA support?

[Paul Dorman]

Hi all. First time writing to the kernel list, so let me know if I break
any conventions. This is long, so I thank you all for your help now :o)

Paul Dorman.

I have two CMD648 PCI ATA66 controllers - one generic and one by
Leadtek. At the moment the Leadtek is installed. I have a dual PII 450
machine on a Gigabyte BXD motherboard. My current kernel is 2.4.4-ac4. I
understand that there may be hardware flaws with the CMD640 chipset. If
this is the case, is there any chance of a work-around? If not, how can
I get the best performance out of this card?

If I use hdparm to set DMA=1 on a drive attached to the controller I get
all kinds of errors when I attempt to access its filesystem:

...
...
ide_dmaproc: chipset supported ide_dma_lostirq func only: 13
hde: lost interrupt

*or* 
...
...
/dev/ide/host2/bus0/target0/lun0: p1
/dev/ide/host2/bus0/target0/lun0: p1
reiserfs: checking transaction log (device 21:01) ...
Using r5 hash to sort names
ReiserFS version 3.6.25
hde: timeout waiting for DMA
ide_dmaproc: chipset supported ide_dma_timeout func only: 14
hde: irq timeout: status=0x58 { DriveReady SeekComplete DataRequest }
hde: timeout waiting for DMA
ide_dmaproc: chipset supported ide_dma_timeout func only: 14
hde: irq timeout: status=0x58 { DriveReady SeekComplete DataRequest }
hde: timeout waiting for DMA
ide_dmaproc: chipset supported ide_dma_timeout func only: 14
hde: irq timeout: status=0x58 { DriveReady SeekComplete DataRequest }
hde: timeout waiting for DMA
ide_dmaproc: chipset supported ide_dma_timeout func only: 14
hde: irq timeout: status=0x58 { DriveReady SeekComplete DataRequest }
hde: DMA disabled
ide2: reset: success
<computer hangs>

*or*
...
...
hde: dma_intr: status=0xff { Busy }
hde: DMA disabled
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
hde: drive not ready for command
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
end_request: I/O error, dev 21:01 (hde), sector 95948856
hde: drive not ready for command
hde: status timeout: status=0xff { Busy }
hde: drive not ready for command
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
hde: drive not ready for command
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
end_request: I/O error, dev 21:01 (hde), sector 95949360
hde: drive not ready for command
hde: status timeout: status=0xff { Busy }
hde: drive not ready for command
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
hde: drive not ready for command
ide2: reset timed-out, status=0xff
hde: status timeout: status=0xff { Busy }
end_request: I/O error, dev 21:01 (hde), sector 95949368
...

I have:-

* Shifted cards around,
* Tried different interrupts,
* Tried THREE different 80 wire cables.

I hope this is the pertinent information...

=============================[DMESG output]============================
CMD648: IDE controller on PCI bus 00 dev 50
CMD648: chipset revision 1
CMD648: not 100% native mode: will probe irqs later
CMD648: ROM enabled at 0xee000000
    ide2: BM-DMA at 0xe800-0xe807, BIOS settings: hde:pio, hdf:pio
    ide3: BM-DMA at 0xe808-0xe80f, BIOS settings: hdg:pio, hdh:pio

A little further down is:

ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
ide1 at 0x170-0x177,0x376 on irq 15

I have tried to force DMA using boot parameters, but without success.

===========================[/proc/interrupts]==========================
           CPU0       CPU1
  0:    1834858    1823909    IO-APIC-edge  timer
  1:       5073       5179    IO-APIC-edge  keyboard
  2:          0          0          XT-PIC  cascade
  7:          0          0    IO-APIC-edge  parport0
  8:          1          0    IO-APIC-edge  rtc
  9:          0          0    IO-APIC-edge  acpi
 10:          4          4    IO-APIC-edge  advansys
 12:      97363     103087    IO-APIC-edge  PS/2 Mouse
 14:     181732     211754    IO-APIC-edge  ide0
 15:          2         10    IO-APIC-edge  ide1
 16:    1431214    1528254   IO-APIC-level  bttv, Ensoniq AudioPCI,
nvidia
 17:      55827      56939   IO-APIC-level  eth0
 19:          0          0   IO-APIC-level  usb-uhci
NMI:          0          0
LOC:    3667630    3667650
ERR:          0
MIS:         21

===========================[/proc/ide/cmd64x]==========================

                                CMD648 Chipset.
--------------- Primary Channel ---------------- Secondary Channel
-------------
                 enabled                          enabled
--------------- drive0 --------- drive1 -------- drive0 ----------
drive1 ------
DMA enabled:    yes              no              no                no 
DMA Mode:       UDMA(4)           PIO(?)          PIO(?)           
PIO(?)
PIO Mode:       ?                ?               ?                 ?
                polling                          polling
                clear                            clear
                enabled                          enabled
CFR       = 0x40, HI = 0x04, LOW = 0x00
ARTTIM23  = 0x0c, HI = 0x00, LOW = 0x0c
MRDMODE   = 0x00, HI = 0x00, LOW = 0x00

======================[/proc/ide/hde/settings]========================

name                  value        min          max          mode
----                  -----        ---          ---          ----
bios_cyl              119150        0           65535         rw
bios_head             16            0           255           rw
bios_sect             63            0           63            rw
breada_readahead      4             0           127           rw
bswap                 0             0           1             r
current_speed         68            0           69            rw
file_readahead        0             0           2097151       rw
ide_scsi              0             0           1             rw
init_speed            68            0           69            rw
io_32bit              0             0           3             rw
keepsettings          0             0           1             rw
lun                   0             0           7             rw
max_kb_per_request    127           1           127           rw
multcount             0             0           8             rw
nice1                 1             0           1             rw
nowerr                0             0           1             rw
number                0             0           3             rw
pio_mode              write-only    0           255           w
slow                  0             0           1             rw
unmaskirq             0             0           1             rw
using_dma             1             0           1             rw

=============================[/proc/pci]==============================

PCI devices found:
  Bus  0, device   0, function  0:
    Host bridge: Intel Corporation 440BX/ZX - 82443BX/ZX Host bridge
(rev 2).
      Master Capable.  Latency=64.  
      Prefetchable 32 bit memory at 0xe8000000 [0xebffffff].
  Bus  0, device   1, function  0:
    PCI bridge: Intel Corporation 440BX/ZX - 82443BX/ZX AGP bridge (rev
2).
      Master Capable.  Latency=64.  Min Gnt=136.
  Bus  0, device   7, function  0:
    ISA bridge: Intel Corporation 82371AB PIIX4 ISA (rev 2).
  Bus  0, device   7, function  1:
    IDE interface: Intel Corporation 82371AB PIIX4 IDE (rev 1).
      Master Capable.  Latency=64.  
      I/O at 0xf000 [0xf00f].
  Bus  0, device   7, function  2:
    USB Controller: Intel Corporation 82371AB PIIX4 USB (rev 1).
      IRQ 19.
      Master Capable.  Latency=64.  
      I/O at 0xd000 [0xd01f].
  Bus  0, device   7, function  3:
    Bridge: Intel Corporation 82371AB PIIX4 ACPI (rev 2).
      IRQ 9.
  Bus  0, device   8, function  0:
    RAID bus controller: CMD Technology Inc PCI0648 (rev 1).
      IRQ 16.
      Master Capable.  Latency=64.  Min Gnt=2.Max Lat=4.
      I/O at 0xd400 [0xd407].
      I/O at 0xd800 [0xd803].
      I/O at 0xdc00 [0xdc07].
      I/O at 0xe000 [0xe003].
      I/O at 0xe400 [0xe40f].
  Bus  0, device   9, function  0:
    Multimedia video controller: Brooktree Corporation Bt848 TV with DMA
push (rev 17).
      IRQ 17.
      Master Capable.  Latency=64.  Min Gnt=16.Max Lat=40.
      Prefetchable 32 bit memory at 0xef000000 [0xef000fff].
  Bus  0, device  10, function  0:
    Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8029(AS)
(rev 0).
      IRQ 18.
      I/O at 0xe800 [0xe81f].
  Bus  0, device  11, function  0:
    Multimedia audio controller: Ensoniq ES1371 [AudioPCI-97] (rev 7).
      IRQ 19.
      Master Capable.  Latency=64.  Min Gnt=12.Max Lat=128.
      I/O at 0xec00 [0xec3f].
  Bus  1, device   0, function  0:
    VGA compatible controller: nVidia Corporation NV15 (Geforce2 GTS)
(rev 163).
      IRQ 16.
      Master Capable.  Latency=248.  Min Gnt=5.Max Lat=1.
      Non-prefetchable 32 bit memory at 0xec000000 [0xecffffff].
      Prefetchable 32 bit memory at 0xe0000000 [0xe7ffffff].

Re: [PATCH][CFT] (updated) ext2 directories in pagecache

[Andreas Dilger]

Daniel writes:
> On Thursday 10 May 2001 09:21, Andreas Dilger wrote:
> > I have changed the code to do the following:
> > - If the COMPAT_DIR_INDEX flag is set at mount/remount time, set the
> >   INDEX mount option (the same as "mount -o index").  This removes
> >   the need to specify the "-o index" option each time for filesystems
> >   which already have indexed directories.
> 
> This applied fine.  The semantics are a little loose though: after 
> mounting -o index, the first directory that overflows its first block 
> will turn on 'sticky indexing' for that partition.

Maybe we can have a "noindex" mount option for this?  In any case, if
you have a kernel which supports indexed directories, no harm is done,
and it will be ignored by other kernels.  There is always the option
of doing "tune2fs -O ^dir_index" to turn off that bit.

> I was originally thinking we should give the admin the ability to 
> create a nonindexed directory if desired, and that's how it used to be 
> before we changed the setting of INDEX_FL from directory creation time 
> to later, when the first directory block overflows.

But you still got INDEX_FL on each new (unindexed) directory, which
didn't help much, IMHO.  You can always do "chattr -h <dir>" (or whatever
the flag will be called) for a directory which you don't want indexed,
and it will immediately cease to be an indexed directory (and the index
block will revert to a normal directory block).

The 'i' flag is already used for the immutable (unchangeable) property.
We could use 'I' for indexed directories, but sometimes you will make a
mistake and use 'i' and in _theory_ you cannot remove the immutable flag
from a file while in multi-user mode for security reasons, so it would
be a hassle to fix this.  Note I haven't sent a patch to Ted for the
hash_indexed flag for chattr, lsattr yet.

The only point of contention is what e2fsck will do in the case where
COMPAT_DIR_INDEX is set on a filesystem.  Does it index all multi-block
directories (and set INDEX_FL for all of them)?  That would be easiest,
but it breaks the ability to selectively turn on/off directory indexing.

If e2fsck doesn't index all multi-block directories, then how do you add
indexing to an existing directory?  Something (kludgy) like:

mkdir newdir
mv olddir/* newdir
rmdir olddir

or the following (the "<one filesystem block>" is the tricky part...)(*)

find <filesystem> -xdev -type d -size +<one filesystem block> | xargs chattr +h
umount
e2fsck <device>

In general, I can't see a good reason why you would NOT want indexing on
some directories and not others once you have chosen to use indexing on
a filesystem.  If people don't want to use indexing, they just shouldn't
turn it on.

Since indexing is a "zero information" feature, it is easy to turn the
flag on and off and have e2fsck fix it (assuming you have a version of
e2fsck that understands indexing).  Even so, you can turn off indexing
for old e2fsck (by using a backup superblock as it will suggest) and
your filesystem is totally OK.  When you later get a new e2fsck you can
re-enable indexing for the filesystem with no loss of information.

(*) This brings up a problem with the "is_dx(dir)" simply checking if the
    EXT2_INDEX_FL is set.  We need to be sure that the code will bail out
    to linear directory searching properly if there is no index present,
    so it is possible to set this flag from user space for later e2fsck.

> I broke out 'ext2_add_compat_feature' as a trial balloon.  It's not
> really very satisfying - it's annoying to have a great piece of
> machinery that's used in exactly one place.  I'd prefer to have
> 'ext2_add_feature' that takes the kind of compatibility as a parameter. 
> We can do this by treating the three ext2_super_block compatibility
> fields as an array, what do you think?  Then we'd get to use it in two
> places :-]

Actually, this is used a lot more in ext3, because we set the HAS_JOURNAL
flag, and set/unset the RECOVERY flag in several places, so it will be
useful there.

> The max link count logic in your previous patch didn't seem to be
> complete so I didn't put it in.  I'll wait for more from you on that.  I
> followed the ReseirFS thread on this but I'm not sure exactly what form
> you want this to take in ext2.

I'm not sure what you mean by incomplete (although I did have the
aforementioned testing problems, so I never did try > 64k subdirectories
in a single directory).  Basically, it works the same as reiserfs, with
the extra caveat that we limit ourselves to EXT2_LINK_MAX for non-indexed
directories (for performance/sanity reasons).  We keep link counts for
indexed directories until we overflow 64k subdirectories (as opposed
to the arbitrary 32000 limit of EXT2_LINK_MAX), after which we use "1"
as the directory link count.  I also want to make sure the code degrades
to the original (slightly simpler) code if CONFIG_EXT2_INDEX is not set.

> The index create code has been moved out of the non-indexed code path
> and into the scope of the indexed path.  This is with a view to cutting
> ext2_add_entry into pieces at some point, for aesthetic reasons.

Yes, I was hoping for something like this.  That function is just getting
way too long...

I'll hopefully get a chance to look at the new code soon.

Cheers, Andreas
-- 
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

Re: [PATCH][CFT] (updated) ext2 directories in pagecache

[Daniel Phillips]

On Monday 14 May 2001 22:04, Andreas Dilger wrote:
> Daniel writes:
> > I was originally thinking we should give the admin the ability to
> > create a nonindexed directory if desired, and that's how it used to
> > be before we changed the setting of INDEX_FL from directory
> > creation time to later, when the first directory block overflows.
>
> But you still got INDEX_FL on each new (unindexed) directory, which
> didn't help much, IMHO.  You can always do "chattr -h <dir>" (or
> whatever the flag will be called) for a directory which you don't
> want indexed, and it will immediately cease to be an indexed
> directory (and the index block will revert to a normal directory
> block).

This is my point.  The can't do that nicely because they have to wait 
for the directory to grow past one block.  I don't really want to go 
back to the original way of doing it - the new code is simpler.  But... 
as long as this sits fine with you, it's done.

> The 'i' flag is already used for the immutable (unchangeable)
> property. We could use 'I' for indexed directories, but sometimes you
> will make a mistake and use 'i' and in _theory_ you cannot remove the
> immutable flag from a file while in multi-user mode for security
> reasons, so it would be a hassle to fix this.  Note I haven't sent a
> patch to Ted for the hash_indexed flag for chattr, lsattr yet.

We still have lots of time.  I see "x" isn't taken.

> The only point of contention is what e2fsck will do in the case where
> COMPAT_DIR_INDEX is set on a filesystem.  Does it index all
> multi-block directories (and set INDEX_FL for all of them)?  That
> would be easiest, but it breaks the ability to selectively turn
> on/off directory indexing.
>
> If e2fsck doesn't index all multi-block directories, then how do you
> add indexing to an existing directory?  Something (kludgy) like:
>
> mkdir newdir
> mv olddir/* newdir
> rmdir olddir
>
> or the following (the "<one filesystem block>" is the tricky
> part...)(*)
>
> find <filesystem> -xdev -type d -size +<one filesystem block> | xargs
> chattr +h umount
> e2fsck <device>
>
> In general, I can't see a good reason why you would NOT want indexing
> on some directories and not others once you have chosen to use
> indexing on a filesystem.  If people don't want to use indexing, they
> just shouldn't turn it on.

"Always on" is so much simpler, given that they asked for it.

> Since indexing is a "zero information" feature, it is easy to turn
> the flag on and off and have e2fsck fix it (assuming you have a
> version of e2fsck that understands indexing).  Even so, you can turn
> off indexing for old e2fsck (by using a backup superblock as it will
> suggest) and your filesystem is totally OK.  When you later get a new
> e2fsck you can re-enable indexing for the filesystem with no loss of
> information.
>
> (*) This brings up a problem with the "is_dx(dir)" simply checking if
> the EXT2_INDEX_FL is set.  We need to be sure that the code will bail
> out to linear directory searching properly if there is no index
> present, so it is possible to set this flag from user space for later
> e2fsck.

You mean in the case that EXT2_INDEX_FL is set, but there isn't 
actually an index there, or it's corrupt?  It's on my list of things to 
do.

> > The max link count logic in your previous patch didn't seem to be
> > complete so I didn't put it in.  I'll wait for more from you on
> > that.  I followed the ReseirFS thread on this but I'm not sure
> > exactly what form you want this to take in ext2.
>
> I'm not sure what you mean by incomplete

I didn't mean that, I found the rest of the code (in the other patch) 
shortly after I wrote it and forgot to edit the mail.

> (although I did have the
> aforementioned testing problems, so I never did try > 64k
> subdirectories in a single directory).  Basically, it works the same
> as reiserfs, with the extra caveat that we limit ourselves to
> EXT2_LINK_MAX for non-indexed directories (for performance/sanity
> reasons).  We keep link counts for indexed directories until we
> overflow 64k subdirectories (as opposed to the arbitrary 32000 limit
> of EXT2_LINK_MAX), after which we use "1" as the directory link
> count.  I also want to make sure the code degrades to the original
> (slightly simpler) code if CONFIG_EXT2_INDEX is not set.
>
> > The index create code has been moved out of the non-indexed code
> > path and into the scope of the indexed path.  This is with a view
> > to cutting ext2_add_entry into pieces at some point, for aesthetic
> > reasons.
>
> Yes, I was hoping for something like this.  That function is just
> getting way too long...
>
> I'll hopefully get a chance to look at the new code soon.

There's more new code today, you can skip yesterday's :-)

--
Daniel

Re: [PATCH][CFT] (updated) ext2 directories in pagecache

[Daniel Phillips]

On Monday 14 May 2001 22:04, Andreas Dilger wrote:
> Maybe we can have a "noindex" mount option for this?

We need that regardless, I just keep forgetting to put it in.  I assume 
the semantics are obvious: no new indexes are created but existing ones 
are maintained.  I.e., -o noindex does not mean 'seek out and destroy 
all indexes'.

--
Daniel