November 2018 - Devel - NFS Ganesha List Archives

Crash in mdcache_readdir_chunked
by Malahal Naineni 12 Nov '18

12 Nov '18

We are getting this crash from a couple of customers (using ganesha2.5 based code with some backports!). Looks like commit 3aa8575880190d7a20946a38ae3bc70b278d1099 (see the last paragraph of the commit message) may fix this. Did anyone see this backtrace before, if so what fixed it? (gdb) bt #0 0x00007fd511af64ab in raise () from /lib64/libpthread.so.0 #1 0x0000000000455447 in crash_handler (signo=11, info=0x7fd47bef47f0, ctx=0x7fd47bef46c0) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/MainNFSD/nfs_init.c:225 #2 <signal handler called> #3 0x000000000054fecf in mdcache_readdir_chunked (directory=0x7fd1697a6c80, whence=0, dir_state=0x7fd47bef57c0, cb=0x4330f3 <populate_dirent>, attrmask=122830, eod_met=0x7fd47bef5e9b) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/FSAL/Stackable_FSALs/FSAL_MDCACHE/mdcache_helpers.c:2971 #4 0x000000000053f44c in mdcache_readdir (dir_hdl=0x7fd1697a6cb8, whence=0x7fd47bef57a0, dir_state=0x7fd47bef57c0, cb=0x4330f3 <populate_dirent>, attrmask=122830, eod_met=0x7fd47bef5e9b) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/FSAL/Stackable_FSALs/FSAL_MDCACHE/mdcache_handle.c:639 #5 0x00000000004339aa in fsal_readdir (directory=0x7fd1697a6cb8, cookie=0, nbfound=0x7fd47bef5e9c, eod_met=0x7fd47bef5e9b, attrmask=122830, cb=0x496a1a <nfs3_readdirplus_callback>, opaque=0x7fd47bef5e50) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/FSAL/fsal_helper.c:1497 #6 0x000000000049680d in nfs3_readdirplus (arg=0x7fd1ed2071d0, req=0x7fd1ed2069c8, res=0x7fd1589bd3d0) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/Protocols/NFS/nfs3_readdirplus.c:309 #7 0x000000000044cc86 in nfs_rpc_execute (reqdata=0x7fd1ed2069a0) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/MainNFSD/nfs_worker_thread.c:1290 #8 0x000000000044d490 in worker_run (ctx=0x4029b20) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/MainNFSD/nfs_worker_thread.c:1562 #9 0x000000000050c4b0 in fridgethr_start_routine (arg=0x4029b20) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/support/fridgethr.c:550 #10 0x00007fd511aeee25 in start_thread () from /lib64/libpthread.so.0 #11 0x00007fd5111af34d in clone () from /lib64/libc.so.6 (gdb) frame 3 #3 0x000000000054fecf in mdcache_readdir_chunked (directory=0x7fd1697a6c80, whence=0, dir_state=0x7fd47bef57c0, cb=0x4330f3 <populate_dirent>, attrmask=122830, eod_met=0x7fd47bef5e9b) at /usr/src/debug/nfs-ganesha-2.5.3-ibm028.00-0.1.1-Source/FSAL/Stackable_FSALs/FSAL_MDCACHE/mdcache_helpers.c:2971 2971 if (dirent->ck == whence) { (gdb) p dirent $1 = (mdcache_dir_entry_t *) 0x1e58408900000000 (gdb) p dirent->ck Cannot access memory at address 0x1e58408900000060 (gdb) l 2966 fsal_status_t status; 2967 enum fsal_dir_result cb_result; 2968 mdcache_entry_t *entry = NULL; 2969 struct attrlist attrs; 2970 2971 if (dirent->ck == whence) { 2972 /* When called with whence, the caller always wants the 2973 * next entry, skip this entry. 2974 */ 2975 continue;

2 4

Announce Push of V2.8-dev.5
by Frank Filz 09 Nov '18

09 Nov '18

Branch next Tag:V2.8-dev.5 Release Highlights * Fix incorrect parent file handle update on rename. * Restore op_ctx->ctx_export and fsal_export in nfs4_op_readdir * 2 delegations fixes Signed-off-by: Frank S. Filz <ffilzlnx(a)mindspring.com> Contents: 4939231 Frank S. Filz V2.8-dev.5 30a5d14 Jiffin Tony Thottan MDCACHE : check conflicts in delegations for rename if destination exists 79a714e Jiffin Tony Thottan SAL : check conflicts for delegation 63fcaf8 Malahal Naineni Restore op_ctx->ctx_export and fsal_export in nfs4_op_readdir cf509a7 Malahal Naineni Fix incorrect parent file handle update on rename.

1 0

Change in ffilz/nfs-ganesha[next]: MDCACHE: fix invalid assert in mdcache_avl_lookup_ck()
by Kinglong Mee (GerritHub) 08 Nov '18

08 Nov '18

Kinglong Mee has uploaded this change for review. ( https://review.gerrithub.io/432517 Change subject: MDCACHE: fix invalid assert in mdcache_avl_lookup_ck() ...................................................................... MDCACHE: fix invalid assert in mdcache_avl_lookup_ck() Change-Id: If6e3cae6737e233c089e66315624e17bd229afa6 Signed-off-by: Kinglong Mee <kinglongmee(a)gmail.com> --- M src/FSAL/Stackable_FSALs/FSAL_MDCACHE/mdcache_avl.c 1 file changed, 6 insertions(+), 9 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/17/432517/1 -- To view, visit https://review.gerrithub.io/432517 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: If6e3cae6737e233c089e66315624e17bd229afa6 Gerrit-Change-Number: 432517 Gerrit-PatchSet: 1 Gerrit-Owner: Kinglong Mee <kinglongmee(a)gmail.com>

1 0

Change in ffilz/nfs-ganesha[next]: Delegations : Handle OPEN4_SHARE_DENY_WRITE for read delegation
by Jiffin Tony Thottan (GerritHub) 08 Nov '18

08 Nov '18

Jiffin Tony Thottan has uploaded this change for review. ( https://review.gerrithub.io/432426 Change subject: Delegations : Handle OPEN4_SHARE_DENY_WRITE for read delegation ...................................................................... Delegations : Handle OPEN4_SHARE_DENY_WRITE for read delegation If read delegation granted for open call with OPEN4_SHARE_DENY_WRITE, then do not send recall incase of writes. Change-Id: I78eafad574940549ff83b85d44e6af2a0e3a8ecd Signed-off-by: Jiffin Tony Thottan <jthottan(a)redhat.com> --- M src/Protocols/NFS/nfs4_op_open.c M src/SAL/state_deleg.c M src/include/sal_data.h 3 files changed, 7 insertions(+), 1 deletion(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/26/432426/1 -- To view, visit https://review.gerrithub.io/432426 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: I78eafad574940549ff83b85d44e6af2a0e3a8ecd Gerrit-Change-Number: 432426 Gerrit-PatchSet: 1 Gerrit-Owner: Jiffin Tony Thottan <jthottan(a)redhat.com>

1 0

Change in ffilz/nfs-ganesha[next]: rados_recov: convert rados_recov_old_oid to be rcu-managed refstr
by Jeff Layton (GerritHub) 07 Nov '18

07 Nov '18

Jeff Layton has uploaded this change for review. ( https://review.gerrithub.io/432327 Change subject: rados_recov: convert rados_recov_old_oid to be rcu-managed refstr ...................................................................... rados_recov: convert rados_recov_old_oid to be rcu-managed refstr Change-Id: I5ede0b89175879fb24cb145946ce6f4cb5e9dc6b Signed-off-by: Jeff Layton <jlayton(a)redhat.com> --- M src/SAL/recovery/recovery_rados.h M src/SAL/recovery/recovery_rados_cluster.c M src/SAL/recovery/recovery_rados_kv.c 3 files changed, 54 insertions(+), 27 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/27/432327/1 -- To view, visit https://review.gerrithub.io/432327 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: I5ede0b89175879fb24cb145946ce6f4cb5e9dc6b Gerrit-Change-Number: 432327 Gerrit-PatchSet: 1 Gerrit-Owner: Jeff Layton <jlayton(a)redhat.com>

1 0

Change in ffilz/nfs-ganesha[next]: rados_recov: convert rados_recov_oid to be RCU-managed gsh_refstr
by Jeff Layton (GerritHub) 07 Nov '18

07 Nov '18

Jeff Layton has uploaded this change for review. ( https://review.gerrithub.io/432326 Change subject: rados_recov: convert rados_recov_oid to be RCU-managed gsh_refstr ...................................................................... rados_recov: convert rados_recov_oid to be RCU-managed gsh_refstr In the big scheme of things, the rados_recov_oid changes very infrequently, and only in response to a different server requesting a grace period. Change the rados_recov_oid to be managed with an atomic refcount and RCU. The general pattern is to take the rcu_read_lock and atomically bump the refcount and then release it. Once we are finished with the string we drop the refcount. When we need to change the string, we swap the old object with the new with an xchg operation, and then call synchronize_rcu to wait for any users of the old pointer to complete. This allows us to fetch and update the thing locklessly but safely. Change-Id: I6df9ce3f0f562b8d760d4c02311528bbe635d17b Signed-off-by: Jeff Layton <jlayton(a)redhat.com> --- M src/SAL/recovery/recovery_rados.h M src/SAL/recovery/recovery_rados_cluster.c M src/SAL/recovery/recovery_rados_kv.c M src/SAL/recovery/recovery_rados_ng.c 4 files changed, 126 insertions(+), 31 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/26/432326/1 -- To view, visit https://review.gerrithub.io/432326 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: I6df9ce3f0f562b8d760d4c02311528bbe635d17b Gerrit-Change-Number: 432326 Gerrit-PatchSet: 1 Gerrit-Owner: Jeff Layton <jlayton(a)redhat.com>

1 0

Change in ffilz/nfs-ganesha[next]: support: add a gsh_refstr type and functions
by Jeff Layton (GerritHub) 07 Nov '18

07 Nov '18

Jeff Layton has uploaded this change for review. ( https://review.gerrithub.io/432325 Change subject: support: add a gsh_refstr type and functions ...................................................................... support: add a gsh_refstr type and functions Create a new structure that consists of an atomic refcount value and a flexarray that is intended to hold a string. When the refcount reaches zero, then the object will be freed. Unfortunately, the liburcu version currently in epel7 doesn't have urcu_ref_get_unless_zero, so we check for that and provide a copypasta'ed version if it isn't present. Change-Id: I742e1fff3e47646e64fec235fad461b206924458 Signed-off-by: Jeff Layton <jlayton(a)redhat.com> --- M src/CMakeLists.txt M src/include/config-h.in.cmake A src/include/gsh_refstr.h M src/support/CMakeLists.txt A src/support/refstr.c 5 files changed, 88 insertions(+), 0 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/25/432325/1 -- To view, visit https://review.gerrithub.io/432325 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: I742e1fff3e47646e64fec235fad461b206924458 Gerrit-Change-Number: 432325 Gerrit-PatchSet: 1 Gerrit-Owner: Jeff Layton <jlayton(a)redhat.com>

1 0

Change in ffilz/nfs-ganesha[next]: Have threads register and unregister with RCU
by Jeff Layton (GerritHub) 07 Nov '18

07 Nov '18

Jeff Layton has uploaded this change for review. ( https://review.gerrithub.io/432324 Change subject: Have threads register and unregister with RCU ...................................................................... Have threads register and unregister with RCU If you want to use rcu-critical sections, then you must register your threads with RCU on creation. Since it's often difficult to predict which threads may end up running a particular callback, ensure that we register all threads that ganesha creates with RCU. Change-Id: Ie9b027cfe12e5ec6e95b13c1b3090ca6be5047e2 Signed-off-by: Jeff Layton <jlayton(a)redhat.com> --- M src/FSAL/FSAL_GLUSTER/fsal_up.c M src/FSAL/FSAL_GPFS/fsal_up.c M src/FSAL/FSAL_PROXY/handle.c M src/FSAL/FSAL_VFS/panfs/mds.c M src/MainNFSD/9p_dispatcher.c M src/MainNFSD/9p_rdma_dispatcher.c M src/MainNFSD/nfs_admin_thread.c M src/MainNFSD/nfs_init.c M src/dbus/dbus_server.c M src/support/delayed_exec.c M src/support/fridgethr.c 11 files changed, 61 insertions(+), 20 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/24/432324/1 -- To view, visit https://review.gerrithub.io/432324 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: Ie9b027cfe12e5ec6e95b13c1b3090ca6be5047e2 Gerrit-Change-Number: 432324 Gerrit-PatchSet: 1 Gerrit-Owner: Jeff Layton <jlayton(a)redhat.com>

1 0

Change in ffilz/nfs-ganesha[next]: cmake: always build in liburcu
by Jeff Layton (GerritHub) 07 Nov '18

07 Nov '18

Jeff Layton has uploaded this change for review. ( https://review.gerrithub.io/432323 Change subject: cmake: always build in liburcu ...................................................................... cmake: always build in liburcu There is some infrastructure in liburcu for handling things locklessly that could be of great use to ganesha. That library is available on both Linux and FreeBSD, so there should be no reason we can't use it with ganesha. Change cmake to always search for liburcu and fail to configure if it isn't present. Change-Id: I7404115a762dc4d4bb8547f0fe7dcde2d9f3720e Signed-off-by: Jeff Layton <jlayton(a)redhat.com> --- M src/CMakeLists.txt M src/cmake/modules/FindLTTng.cmake M src/nfs-ganesha.spec-in.cmake 3 files changed, 5 insertions(+), 3 deletions(-) git pull ssh://review.gerrithub.io:29418/ffilz/nfs-ganesha refs/changes/23/432323/1 -- To view, visit https://review.gerrithub.io/432323 To unsubscribe, or for help writing mail filters, visit https://review.gerrithub.io/settings Gerrit-Project: ffilz/nfs-ganesha Gerrit-Branch: next Gerrit-MessageType: newchange Gerrit-Change-Id: I7404115a762dc4d4bb8547f0fe7dcde2d9f3720e Gerrit-Change-Number: 432323 Gerrit-PatchSet: 1 Gerrit-Owner: Jeff Layton <jlayton(a)redhat.com>

1 0

Blocked Lock Not found when calling up_async_lock_grant
by ntvietvn＠gmail.com 07 Nov '18

07 Nov '18

Hello, When reading the code, I wonder if the blocking lock can work with GPFS. When the FSAL upcalls the SAL with up_async_lock_grant, it passes handle info in gsh_buffdesc so that in lock_grant we can rebuild the handle later (using create_handle) to look for the existing blocked lock in the queue. The issue is create_handle of GPFS always creates a new pointer to obj_hdl and in the lookup fct find_blocked_lock_upcall, ganesha compares the pointer of obj_hdl and not the value of this handle, thus it will never find the existing blocked lock. Should we use a fct like handle_compare instead or did I misunderstand something? Thank you Viet

4 4