Hi, On Tue 17-04-18 12:39:32, Pavlos Parissis wrote: > In one of our production servers where we run kernel version 4.14.32, I noticed > the following: Thanks for the report! So I believe this is the same underlying problem as for the softlockups you were observing - there I believe the linked list just gets corrupted which causes the list iteration to never finish. But this more points to a place where could be some use-after-free issues (either in fsnotify or somewhere else). I'll look into the code. Honza > > ---- > [138630.417246] cache_from_obj: Wrong slab cache. fsnotify_mark_connector but > object is from kmalloc-2048(361:anycast-healthchecker.service) > [138630.477075] ------------[ cut here ]------------ > [138630.500028] WARNING: CPU: 14 PID: 26002 at mm/slab.h:377 > kmem_cache_free+0x133/0x1c0 > [138630.538130] Modules linked in: netconsole binfmt_misc sctp_diag sctp dccp_diag > dccp tcp_diag udp_diag inet_diag unix_diag cfg80211 rfkill 8021q garp mrp xfs loop > x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm irqbypass > crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcbc aesni_intel crypto_simd > glue_helper cryptd iTCO_wdt intel_cstate hpwdt hpilo iTCO_vendor_support sg > intel_rapl_perf ipmi_si pcspkr ioatdma shpchp lpc_ich i2c_i801 ipmi_devintf dca > mfd_core wmi ipmi_msghandler nfsd auth_rpcgss nfs_acl lockd grace sunrpc ip_tables > ext4 mbcache jbd2 i2c_algo_bit sd_mod drm_kms_helper syscopyarea sysfillrect > sysimgblt fb_sys_fops ttm bnx2x serio_raw mdio libcrc32c crc32c_intel hpsa ptp drm > scsi_transport_sas pps_core dm_mirror dm_region_hash dm_log dm_mod dax > [138630.871763] CPU: 14 PID: 26002 Comm: kworker/u66:3 Not tainted > 4.14.32-1.el7.x86_64 #1 > [138630.910864] Hardware name: HP ProLiant BL460c Gen9, BIOS I36 10/25/2017 > [138630.943498] Workqueue: events_unbound fsnotify_connector_destroy_workfn > [138630.976569] task: ffff88203b62ae80 task.stack: ffffc90008a20000 > [138631.005895] RIP: 0010:kmem_cache_free+0x133/0x1c0 > [138631.029230] RSP: 0018:ffffc90008a23e20 EFLAGS: 00010246 > [138631.055217] RAX: 000000000000007c RBX: ffff882000000000 RCX: 0000000000000000 > [138631.091346] RDX: 0000000000000000 RSI: ffff88203f3969d8 RDI: ffff88203f3969d8 > [138631.127139] RBP: ffffc90008a23e38 R08: 0000000000000001 R09: 000000000000057d > [138631.162575] R10: ffff88102b906cf0 R11: ffff88102b913ea0 R12: ffff88103f579980 > [138631.198018] R13: ffff88203bd57c00 R14: ffff88203bd63f00 R15: ffffffff82128388 > [138631.233371] FS: 0000000000000000(0000) GS:ffff88203f380000(0000) > knlGS:0000000000000000 > [138631.273426] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > [138631.301980] CR2: 0000000003600fd8 CR3: 000000000200a006 CR4: 00000000003606e0 > [138631.338783] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > [138631.374525] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 > [138631.409797] Call Trace: > [138631.422387] fsnotify_connector_destroy_workfn+0x68/0x80 > [138631.448516] process_one_work+0x15c/0x380 > [138631.468352] worker_thread+0x4d/0x3e0 > [138631.486705] kthread+0x10c/0x140 > [138631.503091] ? max_active_store+0x80/0x80 > [138631.523217] ? __kthread_parkme+0x70/0x70 > [138631.543338] ? do_syscall_64+0x79/0x1b0 > [138631.562518] ret_from_fork+0x35/0x40 > [138631.580199] Code: 4c 3b a7 d8 00 00 00 0f 84 11 ff ff ff 48 8b 4f 60 49 8b 54 > 24 60 48 c7 c6 c0 19 c3 81 48 c7 c7 c0 9c e7 81 31 c0 e8 90 a5 eb ff <0f> 0b4c 89 > e7 e9 e9 fe ff ff 65 8b 05 2c 3c de 7e 89 c0 48 0f > [138631.672489] ---[ end trace 6748d86d682915c2 ]--- > [138631.695691] cache_from_obj: Wrong slab cache. fsnotify_mark_connector but > object is from kmalloc-2048(361:anycast-healthchecker.service) > --- > > The kernel was fine and server was responsive. > A coworker mentioned that it may come from > https://github.com/torvalds/linux/blob/master/fs/notify/mark.c#L160 > and commit > 08991e83b728 ("fsnotify: Free fsnotify_mark_connector when there is no mark > attached") arrived with 4.12-rc1 may be the cause of it. > > I am wondering if it is related to the issue I reported > https://lkml.org/lkml/2018/4/16/506. But, that issue causes the whole server to > crash, which is not the case for above kernel dump. > > > Cheers, > Pavlos > -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR