Malahal and I think that my fix will resolve the issue. Please test V2.7.0.3. If the double free issue is really resolved, I will immediately tag V2.7.1 (unless we have some other worthy fixes in flight next week). Thanks Frank From: LUCAS Patrice [mailto:patrice.lucas@cea.fr] Sent: Friday, October 5, 2018 12:49 AM To: Malahal Naineni <malahal@gmail.com>; mbenjami@redhat.com Cc: ffilzlnx@mindspring.com; devel@lists.nfs-ganesha.org Subject: Re: [NFS-Ganesha-Devel] Re: double-free bug On 10/05/18 09:20, Malahal Naineni wrote: Initially, I thought that the nfs4_Compound_Free() called from nfs_dupreq_rele (Thread10) and nfs4_Compound_Free() called from nfs4_op_sequence while freeing the slot cache (Thread7) take the same nfs_res_t POINTER. If they both run at the same time, then nfs4_op_sequence could set res_cached to False, then they both find it false and release resources leading to a double free. Well, my assumption was wrong! The nfs4_op_sequence would be using memory from the SLOT cache and the nfs_dupreq_rele() will be using memory allocated with alloc_nfs_res(). For this double free to happen, the above 2 structures should be out of SYNC as far as "res_cached" field. This can happen if data.sa_cachethis is not true. I posted a patch here with WIP: https://review.gerrithub.io/c/ffilz/nfs-ganesha/+/428128 "sa_cachethis" is operation specific as well as NFS client implementation specific looks like. What NFS client is used here? This doesn't seem to be a timing issue. It's a Fedora28 nfs client, using a NFSv4.1 mount. Regards, Patrice Regards, Malahal. On Fri, Oct 5, 2018 at 1:26 AM Matt Benjamin <mbenjami@redhat.com <mailto:mbenjami@redhat.com> > wrote: The drc path does no work but frees the request; I don't see how that's the problem On Thu, Oct 4, 2018 at 3:49 PM, Malahal Naineni <malahal@gmail.com <mailto:malahal@gmail.com> > wrote:
nfs4_Compound_Free() looks at res_cached to free the stuff or not. The code in nfs4_op_sequence() sets it to False and then calls nfs4_Compound_Free(). I don't see any lock that prevents Thread10 (nfs_dupreq_rele path) running at the same time. I am new to this code, so I might be wrong in my analysis though!
One option is to bypass DRC for NFS4.1 and above.
Regards, Malahal.
On Fri, Oct 5, 2018 at 12:01 AM Frank Filz <ffilzlnx@mindspring.com <mailto:ffilzlnx@mindspring.com> > wrote:
Yea, pretty much 4.1 are not cacheable (4.1 uses the slot cache and so has no need of the dupreq cache).
With my patch, test_stateid isn’t doing anything different than any of the other 4.1 ops that actually free memory in their Free routines.
Maybe they all can lead to double free? In which case somewhere along the line we are doing something wrong with the dup req cache for 4.1///
Frank
From: Malahal Naineni [mailto:malahal@gmail.com <mailto:malahal@gmail.com> ] Sent: Thursday, October 4, 2018 11:16 AM To: ffilzlnx@mindspring.com <mailto:ffilzlnx@mindspring.com> Cc: patrice.lucas@cea.fr <mailto:patrice.lucas@cea.fr> ; devel@lists.nfs-ganesha.org <mailto:devel@lists.nfs-ganesha.org> Subject: [NFS-Ganesha-Devel] Re: double-free bug
Thread10 thinks that op_test stateid_is not cachable, so it actually frees the response and other goodies allocated. But thread7 finds in the slot cache and tries to free leading to a double free. The code path has to be for minor version 1 or 2 (not zero) based on line numbers. I don't know much about 4.1 slot cache.
Regards, Malahal.
On Thu, Oct 4, 2018 at 8:52 PM Frank Filz <ffilzlnx@mindspring.com <mailto:ffilzlnx@mindspring.com> > wrote:
The only thing I can think of is thata TEST_STATEID was issued with minor version = 0 which is the only way it can fail.
I’m going to submit a fix that checks for return status before freeing.
A couple Free routines NULL out the values they free, but almost all check for NFS4_OK. There are a couple others that also don’t check. I’ll fix those too.
Frank
From: patrice.lucas@cea.fr <mailto:patrice.lucas@cea.fr> [mailto:patrice.lucas@cea.fr <mailto:patrice.lucas@cea.fr> ] Sent: Thursday, October 4, 2018 6:43 AM To: devel@lists.nfs-ganesha.org <mailto:devel@lists.nfs-ganesha.org> Subject: [NFS-Ganesha-Devel] double-free bug
Hello everyone,
Frequent memory crashs have been occurring for few weeks in the nfs-ganesha CEA FSAL-PROXY continuous integration test. I finally make time for looking at these problems today by running the nfs-ganesha server under Address Sanitizer.
I got the following stack wih a double-free error. Could anyone explain this error ? Someone who well understand the dup-req cache ? Or someone who already works with the code of the nfs4_op_test_stateid operation ?
The nfs4_op_test_stateid was introduce this summer by gerrit patch 418826 from fatih-acar, 07/22/2018.
The dup-req cache stack seems to be involved in this error.
Regards,
Patrice
==7037==ERROR: AddressSanitizer: attempting double-free on 0x60200001ced0 in thread T7: #0 0x480c09 in __interceptor_free (/usr/bin/ganesha.nfsd+0x480c09) #1 0x897125 in gsh_free /opt/nfs-ganesha/src/include/abstract_mem.h:299 #2 0x896f88 in nfs4_op_test_stateid_Free /opt/nfs-ganesha/src/Protocols/NFS/nfs4_op_test_stateid.c:121 #3 0x703702 in nfs4_Compound_FreeOne /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:1081 #4 0x7042c4 in nfs4_Compound_Free /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:1119 #5 0x865c4a in nfs4_op_sequence /opt/nfs-ganesha/src/Protocols/NFS/nfs4_op_sequence.c:185 #6 0x6fd80f in nfs4_Compound /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:903 #7 0x67167c in nfs_rpc_process_request /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1329 #8 0x663040 in nfs_rpc_valid_NFS /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1539 #9 0x7ffff7bb94a1 in svc_vc_decode /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:824 #10 0x6542ce in nfs_rpc_decode_request /opt/nfs-ganesha/src/MainNFSD/nfs_rpc_dispatcher_thread.c:1341 #11 0x7ffff7bb934c in svc_vc_recv /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:797 #12 0x7ffff7bb47be in svc_rqst_xprt_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:767 #13 0x7ffff7bb51af in svc_rqst_epoll_events /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:939 #14 0x7ffff7bb4e94 in svc_rqst_epoll_loop /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1012:8 #15 0x7ffff7bb38bf in svc_rqst_run_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1048:14 #16 0x7ffff7bc077c in work_pool_thread /opt/nfs-ganesha/src/libntirpc/src/work_pool.c:181 #17 0x7ffff6367e24 in start_thread (/lib64/libpthread.so.0+0x7e24) #18 0x7ffff575c34c in __clone (/lib64/libc.so.6+0xf834c)
0x60200001ced0 is located 0 bytes inside of 4-byte region [0x60200001ced0,0x60200001ced4) freed by thread T10 here: #0 0x480c09 in __interceptor_free (/usr/bin/ganesha.nfsd+0x480c09) #1 0x897125 in gsh_free /opt/nfs-ganesha/src/include/abstract_mem.h:299 #2 0x896f88 in nfs4_op_test_stateid_Free /opt/nfs-ganesha/src/Protocols/NFS/nfs4_op_test_stateid.c:121 #3 0x703702 in nfs4_Compound_FreeOne /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:1081 #4 0x7042c4 in nfs4_Compound_Free /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:1119 #5 0xcec2a4 in nfs_dupreq_rele /opt/nfs-ganesha/src/RPCAL/nfs_dupreq.c:1315 #6 0x673196 in nfs_rpc_process_request /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1442 #7 0x663040 in nfs_rpc_valid_NFS /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1539 #8 0x7ffff7bb94a1 in svc_vc_decode /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:824 #9 0x6542ce in nfs_rpc_decode_request /opt/nfs-ganesha/src/MainNFSD/nfs_rpc_dispatcher_thread.c:1341 #10 0x7ffff7bb934c in svc_vc_recv /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:797 #11 0x7ffff7bb47be in svc_rqst_xprt_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:767 #12 0x7ffff7bb51af in svc_rqst_epoll_events /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:939 #13 0x7ffff7bb4e94 in svc_rqst_epoll_loop /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1012:8 #14 0x7ffff7bb38bf in svc_rqst_run_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1048:14 #15 0x7ffff7bc077c in work_pool_thread /opt/nfs-ganesha/src/libntirpc/src/work_pool.c:181 #16 0x7ffff6367e24 in start_thread (/lib64/libpthread.so.0+0x7e24)
previously allocated by thread T10 here: #0 0x480e59 in calloc (/usr/bin/ganesha.nfsd+0x480e59) #1 0x89689a in gsh_calloc__ /opt/nfs-ganesha/src/include/abstract_mem.h:145 #2 0x895c4e in nfs4_op_test_stateid /opt/nfs-ganesha/src/Protocols/NFS/nfs4_op_test_stateid.c:88:3 #3 0x6fd80f in nfs4_Compound /opt/nfs-ganesha/src/Protocols/NFS/nfs4_Compound.c:903 #4 0x67167c in nfs_rpc_process_request /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1329 #5 0x663040 in nfs_rpc_valid_NFS /opt/nfs-ganesha/src/MainNFSD/nfs_worker_thread.c:1539 #6 0x7ffff7bb94a1 in svc_vc_decode /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:824 #7 0x6542ce in nfs_rpc_decode_request /opt/nfs-ganesha/src/MainNFSD/nfs_rpc_dispatcher_thread.c:1341 #8 0x7ffff7bb934c in svc_vc_recv /opt/nfs-ganesha/src/libntirpc/src/svc_vc.c:797 #9 0x7ffff7bb47be in svc_rqst_xprt_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:767 #10 0x7ffff7bb51af in svc_rqst_epoll_events /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:939 #11 0x7ffff7bb4e94 in svc_rqst_epoll_loop /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1012:8 #12 0x7ffff7bb38bf in svc_rqst_run_task /opt/nfs-ganesha/src/libntirpc/src/svc_rqst.c:1048:14 #13 0x7ffff7bc077c in work_pool_thread /opt/nfs-ganesha/src/libntirpc/src/work_pool.c:181 #14 0x7ffff6367e24 in start_thread (/lib64/libpthread.so.0+0x7e24)
_______________________________________________ Devel mailing list -- devel@lists.nfs-ganesha.org <mailto:devel@lists.nfs-ganesha.org> To unsubscribe send an email to devel-leave@lists.nfs-ganesha.org <mailto:devel-leave@lists.nfs-ganesha.org>
_______________________________________________ Devel mailing list -- devel@lists.nfs-ganesha.org <mailto:devel@lists.nfs-ganesha.org> To unsubscribe send an email to devel-leave@lists.nfs-ganesha.org <mailto:devel-leave@lists.nfs-ganesha.org>
-- Matt Benjamin Red Hat, Inc. 315 West Huron Street, Suite 140A Ann Arbor, Michigan 48103 http://www.redhat.com/en/technologies/storage tel. 734-821-5101 fax. 734-769-8938 cel. 734-216-5309 -- Patrice LUCAS Ingenieur-Chercheur, CEA-DAM/DSSI/SISR/LA2S tel : +33 (0)1 69 26 47 86 e-mail : patrice.lucas@cea.fr <mailto:patrice.lucas@cea.fr>