Bug 919435

+++ This bug was initially created as a clone of Bug #706935 +++

Description of problem:
RPM sigfaults when installing a package with huge (80k) number of files.

Version-Release number of selected component (if applicable):


How reproducible:
Try to install package with huge number of files. Can be done using clean local DB under non-root user.

Steps to Reproduce:
1. rpm --define "_dbpath /tmp" --initdb
2. wget http://atlas-computing.web.cern.ch/atlas-computing/links/reposDirectory/offline/slc5/noarch/AtlasSimulation_16_6_6_noarch-1-1.noarch.rpm 
(NB: 90Mb package)
3. rpm --define "_dbpath /tmp" -i --prefix=/tmp --nodeps AtlasSimulation_16_6_6_noarch-1-1.noarch.rpm
Segmentation fault (core dumped)

Actual results:

#0  headerCheck (ts=0x9e146c0, uh=0x0, uc=21704528, msg=0x7fff1aa33e68) at package.c:331
331	    int_32 il = ntohl(ei[0]);
(gdb) bt
#0  headerCheck (ts=0x9e146c0, uh=0x0, uc=21704528, msg=0x7fff1aa33e68) at package.c:331
#1  0x00000037a2634bec in rpmdbAdd (db=0x9e16080, iid=<value optimized out>, h=0xc6580c0, ts=0x9e146c0, hdrchk=0x37a2a20210 <headerCheck>)
    at rpmdb.c:3140
#2  0x00000037a2a2595a in rpmpsmStage (psm=0xc6579d0, stage=<value optimized out>) at psm.c:2111
#3  0x00000037a2a266c5 in rpmpsmStage (psm=0xc6579d0, stage=PSM_POST) at psm.c:1871
#4  0x00000037a2a25f8d in rpmpsmStage (psm=0xc6579d0, stage=PSM_PKGINSTALL) at psm.c:1994
#5  0x00000037a2a48845 in rpmtsRun (ts=0x9e146c0, okProbs=0x0, ignoreSet=<value optimized out>) at transaction.c:2185
#6  0x00000037a2a37aa3 in rpmInstall (ts=0x9e146c0, ia=0x37a2c6bf80, fileArgv=<value optimized out>) at rpminstall.c:701
#7  0x00000000004043a0 in main (argc=7, argv=<value optimized out>) at ./rpmqv.c:798


Expected results:
installed package.

Additional info:
"uh" pointer returned earlier by doHeaderUnload not checked but may be zero.
This is not native RHEL5, but the same code base.
The same behaviour seen on RHEL6.

--- Additional comment from Andrei on 2011-05-24 03:57:23 EDT ---

we realized that the package installs without --prefix option.

--- Additional comment from Panu Matilainen on 2011-05-24 06:21:41 EDT ---

Interesting case. There are actually several different (but related) issues here:
The original header is big, but below the artificially limited 16MB max header size. However relocation nearly doubles the size because rpm saves both the original and relocated paths in the header, and that pushes the total header size above the limit, causing headerUnload() to return NULL which then blows up in rpmdbAdd().

rpm >= 4.9.0 doesn't crash (it checks for the NULL) but adding the header fails there too, causing the install to be considered failed despite all the files having been laid on disk already. Failing this because of an artificial size limit is silly, especially since it's rpm itself which causes the limit to be exceeded here when the original header from the package is perfectly legal even with the current (arguably too low) limit on the header size.

--- Additional comment from Panu Matilainen on 2013-03-08 08:02:13 EST ---

Sorry, this has gone all but forgotten. We wont be able to address the fundamental limitation in rhel-5 (or -6 for that matter) but segfaulting is not acceptable, that needs to be fixed.