Lustre devel - Jan 2007 - [Bug 3244] allow more than 32000 subdirectories

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=3244



Created an attachment (id=9464)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9464&action=view)
Changes to e2fsck

e2fsck uses a very different algorithm to calculate inode counts (& links)
efficiently. A ''single'' bitmap is used to indicate which
inodes have a link
count of 1. If an inode has a link count greater than 1, then it is stored on a

linked list. A ''multiple'' bitmap is used to indicate which
inodes are present
in
the linked list.

ext2fs_icount_increment() is modified such that if nlink count (el->count)
becomes greater than EXT2_LINK_MAX(65000), then it is set to 1. Also when an
inode belongs to ''multiple'' list (has more than 1 links), then
its el->count is

incremented only if it is not equal to 1.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=3244

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Attachment #9464|review?(adilger@clusterfs.co|review+
               Flag|m)                          |


(From update of attachment 9464)
This looks good, but please note that the EXT2_LINK_MAX part is already in the
1.39-WIP patch in e2fsprogs-cfs3, which means that it is also in the upstream
code.  The 1.39-WIP patch also holds the #define for RO_COMPAT_DIR_NLINK.  The
one thing that is missing is the fix for bug 6472, where e2fsck
shouldn''t
report an error for a directory with i_nlink = 1 (this can happen if the
directory is > 65000 subdirs and they are then removed but the parent
directory
is not).

Can you please make a test case for e2fsprogs  (Girish knows how to do this if
you need help).  The test case should contain an empty directory with i_nlink 1,
and a directory with > 65000 subdirectories.

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=3244



Created an attachment (id=9469)
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
 --> (https://bugzilla.lustre.org/attachment.cgi?id=9469&action=view)
Changes to e2fsck for condition when directory has link count of 1, but all
subdirectories are deleted.

This patch uses the logic that a directory can have a nlink = 1, only if the
subdirectory count had ever crossed 65000. So we check that if inode is a
directory and its link_count as read from disk is 1, then we do not report
BAD_REF_COUNT problem.

Even if the link count for a directory has become 1 due to corruption instead
of
subdir count exceeding 65000, it will not cause any further corruption.