Bug 151255

All depends on whether the lock scoping is desired to be global
(as implemented) or per-chroot. The path has been made configurable,
and changed to be per-root in rpm-4.4.1:
#
# Default path to the file used for transaction fcmtl lock.
#
# The previous, FHS clompliant, name was /var/lock/run/transaction,
# but the transaction lock needs to be per-database, not global, for
# some rpmdb operations.
#
# XXX Note: the file name is chosesn as __db.000 to expedite
# support issues, many users are doing "rm -f /var/lib/rpm/__db*"
# these days.
%_rpmlock_path  %{_dbpath}/__db.000

perhaps it needs to be per-database 'If and only' if the EUID can write locks,
and simply ignored if it is not possible for the EUID to open the DB +w ?

For a non-+w user, a process local copy of the rpmdb would permit simultaneous
build processes (for example, in some work I an doing), living somewhere in
probably the %_builddir tree seems to be a comprehensive if perhaps overly
cautious solution.

The path has been made configurable in rpm-4.4.1, however it still does not work
as expected.

The path of the lock does not take into account the --root.

To fix this issue in lib/rpmlock.c

function: rpmlock_new
-           char * t = rpmExpand(rpmlock_path_default, NULL);
+           char * t = rpmGenPath(rootdir, rpmlock_path_default, NULL);

function: rpmtsAcquireLock
-       if (!rootDir)
+       if (!rootDir || rpmtsChrootDone(ts))

(The above changes are made in the MontaVista version of rpm to address this issue.)

Paul, if these changes look good to you I will apply to head.

Cheers...james

Just an FYI.  These changes are still required for RPM 4.4.2.

Added to rpm cvs, should be in rpm-4.4.4-0.6 when built.

Reopening to get fixed in RHEL4 proper.

Created attachment 123780 [details]
rpm-transaction-lock-in-root.patch

Example patch for RHEL4.
This would use a transaction lock in the root specified.

This issue is on Red Hat Engineering's list of planned work items 
for the upcoming Red Hat Enterprise Linux 4.4 release.  Engineering 
resources have been assigned and barring unforeseen circumstances, Red 
Hat intends to include this item in the 4.4 release.

I believe I correctly applied this patch and --initdb still looks in the wrong 
spot for the transaction lock.

$ rpm --root /export/test --initdb
error: can't create transaction lock

strace snip:
open("/var/lock/rpm/transaction", O_RDONLY) = 3
write(2, "error: ", 7error: )                  = 7

my patch line from my spec
%patch9 -p1

as defined by:
Patch9: rpm-transaction-lock-in-root.patch

If my patch is suspect and you can provide me with a build "rpm" command 
binary or binary package, I'll give it a try as well... I must be missing 
something.

An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2006-0440.html

Not sure where the problem lies, but I can confirm Jerry's problem (comment #35)
with fix/patch as-is doesn't work.  Just upgraded a box to centos-4.4, and rpm
--root no longer works (which makes my mock builds hosted here fail miserably)

$ rpm -q rpm
rpm-4.3.3-18_nonptl

$ rpm --root /tmp/foo --initdb
error: can't create transaction lock on /tmp/foo/var/lock/rpm/transaction

the errata https://rhn.redhat.com/errata/RHBA-2006-0440.html says:

Transaction lock issues using RPM with --root and --dbpath have been fixed.
Note, to use --dbpath as non root, customers will need to define the
%_rpmlock_path macro to a location that can be written to; future versions
of RPM use '%{_dbpath}/__db.000'.

Appears to work for us, even if cumbersome.

Rex, mock will need to create ROOT/var/lock/rpm or you'd need to mkdir -p
/tmp/foo/var/lock/rpm as before it was locking outside the chroot which is
incorrect, now it's locking in the chroot and probably that path does not exist.
 Else defining _rpmlock_path in ~/.rpmmacros to somewhere within %_dbpath which
dir structure is created by rpmdb operations.

Thanks Paul, looks like mock needs some patching to ensure creation of
ROOT/var/lock/rpm too.