Hi, I'm trying both the 2.7.0 release and 2.7.1pre1-2007-11-30-182035 from CVS with an existing PVFS2 filesystem converted from 2.6.3 and am running into two major issues:
Line 1944 (CVS) of /src/io/bmi/bmi_ib/ib.c: bmi_method_addr_forget_callback(ibmap->c->bmi_addr); ...always results in a segfault with any of the admin apps after the command completes, even pvfs2-ping. It doesn't happen when pvfs2-client exits. According to ltrace, it seems to happen on free(), so I guess it's after the callback actually occurs? ibv_close_device(0x5848b0, 33, 0x584a40, 0, 0x5849c8 <unfinished ...> SYS_close(4) = 0 SYS_close(3) = 0 <... ibv_close_device resumed> ) = 0 free(0x584a00) = <void> free(0x5838b0) = <void> pthread_mutex_unlock(0x56fee0, 1, 0x5848a0, 0x5b1590, 0x5849c8) = 0 free(0x583870) = <void> pthread_mutex_unlock(0x56f660, 32, 0x584820, 0, 0x5849c8) = 0 free(0x5753f0) = <void> free(0x583890) = <void> free(0x58afa0) = <void> free(0x58af80) = <void> free(0x58afc0) = <void> --- SIGSEGV (Segmentation fault) --- +++ killed by SIGSEGV +++ The second issue is with the VFS interface - copying anything (e.g. using cp or dd bs=1k) to the fs using a block size of smaller than 8k always results in a file truncated to 4096 bytes and the message "No space left on device." Using dd bs=8k or larger works fine, as does pvfs2-cp (except for the segfault part). I'm currently testing with only openib enabled, but this truncation thing also happened when only TCP was enabled. Current setup: Configured with "./configure --with-openib=/usr --with-kernel=/usr/src/linux --enable-fast --without-bmi-tcp" Kernel: 2.6.23.9/x86_64 Processors: Dual Opteron 280 IB HCA: QLogic QLE7140 (Infinipath 2.1 libs, kernel.org drivers) Data servers: 6 (2.6 TB/server) Metadata servers: 2 OS: Debian 4.0 (glibc 2.3.6) Thanks! Tad _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
