ocfs2: fix crash caused by stale lvb with fsdlm plugin
PATCH 1/1 ocfs2: fix crash caused by stale lvb with fsdlm plugin
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"
[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN) ocfs2_nodemanager ocfs2_stackglue(OEN) quota_tree dlm(OEN) configfs fuse sd_mod iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi af_packet iscsi_ibft iscsi_boot_sysfs softdog xfs libcrc32c ppdev parport_pc pcspkr parport joydev virtio_balloon virtio_net i2c_piix4 acpi_cpufreq button processor ext4 crc16 jbd2 mbcache ata_generic cirrus virtio_blk ata_piix drm_kms_helper ahci syscopyarea libahci sysfillrect sysimgblt fb_sys_fops ttm floppy libata drm virtio_pci virtio_ring uhci_hcd virtio ehci_hcd usbcore serio_raw usb_common sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod autofs4 [ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:
[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>
"
It's because ocfs2_inode_lock() get us stale LVB in which the i_size is not equal to the disk i_size. We mistakenly trust the LVB because the underlaying fsdlm dlm_lock() doesn't set lkb_sbflags with DLM_SBF_VALNOTVALID properly for us. But, why?
The current code tries to downconvert lock without DLM_LKF_VALBLK flag to tell o2cb don't update RSB's LVB if it's a PR->NULL conversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.
The fsdlm plugin behaves different on DLM_LKF_VALBLK, it depends on DLM_LKF_VALBLK to decide if we care about the LVB in the LKB. If DLM_LKF_VALBLK is not set, fsdlm will skip recovering RSB's LVB from this lkb and set the right DLM_SBF_VALNOTVALID appropriately when node failure happens.
The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with LOCK_TYPE_USES_LVB;
The 1st round:
Node1 Node2
RSB1: PR
RSB1(master): NULL->EX
ocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwrite lkb->lkb_exflags
with no DLM_LKF_VALBLK)
RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()
/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should set RSB_VALNOTVALID for RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK)) /* This means we miss the chance to
return; * to invalid the LVB here.
*/
The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)
/* dlm_lock() return the stale lvb without setting DLM_SBF_VALNOTVALID */ ocfs2_meta_lvb_is_trustable() return 1 /* so we don't refresh inode from disk */
ocfs2_truncate_file()
mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */
The fix is quite straightforward. We keep to set DLM_LKF_VALBLK flag for dlm_lock() if the lock resource type needs LVB and the fsdlm plugin is uesed.
--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)
fs/ocfs2/dlmglue.c +10 -0
@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); +/* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always +* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get +* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID??being set. +*/ +if (!ocfs2_is_o2cb_active () && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb =1 ; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ;
fs/ocfs2/stackglue.c +6 -0
@@ -48,6 +48,12 @@ static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";*/ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active (void ) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {struct ocfs2_stack_plugin *p;
fs/ocfs2/stackglue.h +3 -0
@@ -298,6 +298,9 @@ void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_pint ocfs2_stack_glue_register (struct ocfs2_stack_plugin *plugin);void ocfs2_stack_glue_unregister (struct ocfs2_stack_plugin *plugin); +/* In ocfs2_downconvert_lock(), we need to know which stack we are using */ +int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */
Signature · 1 line
2.6.6
Hi Eric,
Looks good for me.
Just one suggestion,
please monitor if the LVB sharing mechanism in the cluster still works well in the normal scenario, to avoid any performance decrease regression problem.
Thanks
Gang
Quoted 215 lines
Quoted 1 line
Quoted 1 line
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN)
ocfs2_nodemanagerocfs2_stackglue(OEN)quota_treedlm(OEN) configfs fusesd_modiscsi_tcplibiscsi_tcplibiscsiscsi_transport_iscsiaf_packetiscsi_ibftiscsi_boot_sysfssoftdog xfs libcrc32c ppdevparport_pcpcspkr
parport joydevvirtio_balloonvirtio_neti2c_piix4acpi_cpufreqbutton
processor ext4 crc16 jbd2 mbcacheata_genericcirrusvirtio_blkata_piix
drm_kms_helperahci syscopyarea libahci sysfillrect sysimgbltfb_sys_fopsttm floppy libata drmvirtio_pcivirtio_ringuhci_hcdvirtioehci_hcdusbcoreserio_rawusb_commonsgdm_multipathdm_modscsi_dh_rdacscsi_dh_emcscsi_dh_aluascsi_modautofs4[ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>"
It's because
ocfs2_inode_lock()get us stale LVB in which thei_sizeis not equal to the diski_size. We mistakenly trust the LVB because the underlaying
fsdlmdlm_lock()doesn't setlkb_sbflagswithDLM_SBF_VALNOTVALIDproperly for
us. But, why?The current code tries to downconvert lock without
DLM_LKF_VALBLKflag to tell o2cb don't update RSB's LVB if it's aPR->NULLconversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.The fsdlm plugin behaves different on
DLM_LKF_VALBLK, it depends onDLM_LKF_VALBLKto decide if we care about the LVB in the LKB. IfDLM_LKF_VALBLK
is not set, fsdlm will skip recovering RSB's LVB from this lkb and set the rightDLM_SBF_VALNOTVALIDappropriately when node failure happens.The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with
LOCK_TYPE_USES_LVB;The 1st round:
Node1 Node2
RSB1: PRRSB1(master): NULL->EX
ocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwritelkb->lkb_exflags
with noDLM_LKF_VALBLK)RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should setRSB_VALNOTVALIDfor RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK))/* This means we miss the chance to
return; * to invalid the LVB here.
*/The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)/* dlm_lock() return the stale lvb without setting DLM_SBF_VALNOTVALID*/
ocfs2_meta_lvb_is_trustable()return 1 /* so we don't refresh inode from
disk */
ocfs2_truncate_file()mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */The fix is quite straightforward. We keep to set
DLM_LKF_VALBLKflag fordlm_lock()
if the lock resource type needs LVB and the fsdlm plugin is uesed.Signed-off-by Eric Ren <[email protected]>--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)fs/ocfs2/dlmglue.c +10
-0@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); +/* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always +* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get +* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set. +*/ +if (!ocfs2_is_o2cb_active () && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb =1 ; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ;fs/ocfs2/stackglue.c +6
-0@@ -48,6 +48,12 @@ static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] ="/sbin/ocfs2_hb_ctl"; */ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active (void ) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {fs/ocfs2/stackglue.c +0
-0struct ocfs2_stack_plugin *p;fs/ocfs2/stackglue.h +0
-0@@ -298,6 +298,9 @@ void ocfs2_stack_glue_set_max_proto_version(struct
ocfs2_protocol_version*max_p
intocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin); voidocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin);fs/ocfs2/stackglue.h +3
-0+/* In ocfs2_downconvert_lock(), we need to know which stack we are using */ +int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */ Signature · 1 line
2.6.6
Hi Gang,
Quoted 6 lines · Gang He
Hi Eric,
Looks good for me.
Just one suggestion,
please monitor if the LVB sharing mechanism in the cluster still works well in the normal scenario, to avoid any performance decrease regression problem.
Thanks for your review. I have done the testing as you suggested, and it works as expected.
Thanks,
Eric
Quoted 226 lines
Reviewed-by Gang He <[email protected]>Thanks
GangQuoted 214 lines
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN)
ocfs2_nodemanagerocfs2_stackglue(OEN)quota_treedlm(OEN) configfs fusesd_modiscsi_tcplibiscsi_tcplibiscsiscsi_transport_iscsiaf_packetiscsi_ibftiscsi_boot_sysfssoftdog xfs libcrc32c ppdevparport_pcpcspkr
parport joydevvirtio_balloonvirtio_neti2c_piix4acpi_cpufreqbutton
processor ext4 crc16 jbd2 mbcacheata_genericcirrusvirtio_blkata_piix
drm_kms_helperahci syscopyarea libahci sysfillrect sysimgbltfb_sys_fopsttm floppy libata drmvirtio_pcivirtio_ringuhci_hcdvirtioehci_hcdusbcoreserio_rawusb_commonsgdm_multipathdm_modscsi_dh_rdacscsi_dh_emcscsi_dh_aluascsi_modautofs4[ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>"
It's because
ocfs2_inode_lock()get us stale LVB in which thei_sizeis not equal to the diski_size. We mistakenly trust the LVB because the underlaying
fsdlmdlm_lock()doesn't setlkb_sbflagswithDLM_SBF_VALNOTVALIDproperly for
us. But, why?The current code tries to downconvert lock without
DLM_LKF_VALBLKflag to tell o2cb don't update RSB's LVB if it's aPR->NULLconversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.The fsdlm plugin behaves different on
DLM_LKF_VALBLK, it depends onDLM_LKF_VALBLKto decide if we care about the LVB in the LKB. IfDLM_LKF_VALBLK
is not set, fsdlm will skip recovering RSB's LVB from this lkb and set the rightDLM_SBF_VALNOTVALIDappropriately when node failure happens.The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with
LOCK_TYPE_USES_LVB;The 1st round:
Node1 Node2
RSB1: PRRSB1(master): NULL->EX
ocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwritelkb->lkb_exflags
with noDLM_LKF_VALBLK)RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should setRSB_VALNOTVALIDfor RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK))/* This means we miss the chance to
return; * to invalid the LVB here.
*/The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)/* dlm_lock() return the stale lvb without setting DLM_SBF_VALNOTVALID*/
ocfs2_meta_lvb_is_trustable()return 1 /* so we don't refresh inode from
disk */
ocfs2_truncate_file()mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */The fix is quite straightforward. We keep to set
DLM_LKF_VALBLKflag fordlm_lock()
if the lock resource type needs LVB and the fsdlm plugin is uesed.Signed-off-by Eric Ren <[email protected]>--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)fs/ocfs2/dlmglue.c +10
-0@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); +/* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always +* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get +* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set. +*/ +if (!ocfs2_is_o2cb_active () && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb =1 ; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ;fs/ocfs2/stackglue.c +6
-0@@ -48,6 +48,12 @@ static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] ="/sbin/ocfs2_hb_ctl"; */ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active (void ) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {struct
ocfs2_stack_plugin*p;fs/ocfs2/stackglue.h +0
-0@@ -298,6 +298,9 @@ void ocfs2_stack_glue_set_max_proto_version(struct
ocfs2_protocol_version*max_p
intocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin); voidocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin);fs/ocfs2/stackglue.h +3
-0+/* In ocfs2_downconvert_lock(), we need to know which stack we are using */ +int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */ Signature · 1 line
2.6.6_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel at oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel
In you description, this issue can only happen in case of stack user +
fsdlm.
So I feel we'd better to make stack user and o2cb behaves the same,
other than treat it as a special case.
Thanks,
Joseph
Quoted 169 lines · Eric Ren
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN) ocfs2_nodemanager ocfs2_stackglue(OEN) quota_tree dlm(OEN) configfs fuse sd_mod iscsi_tcp libiscsi_tcp libiscsi scsi_transport_iscsi af_packet iscsi_ibft iscsi_boot_sysfs softdog xfs libcrc32c ppdev parport_pc pcspkr parport joydev virtio_balloon virtio_net i2c_piix4 acpi_cpufreq button processor ext4 crc16 jbd2 mbcache ata_generic cirrus virtio_blk ata_piix drm_kms_helper ahci syscopyarea libahci sysfillrect sysimgblt fb_sys_fops ttm floppy libata drm virtio_pci virtio_ring uhci_hcd virtio ehci_hcd usbcore serio_raw usb_common sg dm_multipath dm_mod scsi_dh_rdac scsi_dh_emc scsi_dh_alua scsi_mod autofs4 [ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>"
It's because
ocfs2_inode_lock()get us stale LVB in which thei_sizeis not equal to the diski_size. We mistakenly trust the LVB because the underlaying fsdlmdlm_lock()doesn't setlkb_sbflagswithDLM_SBF_VALNOTVALIDproperly for us. But, why?The current code tries to downconvert lock without
DLM_LKF_VALBLKflag to tell o2cb don't update RSB's LVB if it's aPR->NULLconversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.The fsdlm plugin behaves different on
DLM_LKF_VALBLK, it depends onDLM_LKF_VALBLKto decide if we care about the LVB in the LKB. IfDLM_LKF_VALBLKis not set, fsdlm will skip recovering RSB's LVB from this lkb and set the rightDLM_SBF_VALNOTVALIDappropriately when node failure happens.The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with
LOCK_TYPE_USES_LVB;The 1st round:
Node1 Node2
RSB1: PRRSB1(master): NULL->EX
ocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwritelkb->lkb_exflags
with noDLM_LKF_VALBLK)RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should setRSB_VALNOTVALIDfor RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK))/* This means we miss the chance to
return; * to invalid the LVB here.
*/The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)
/*dlm_lock()return the stale lvb without settingDLM_SBF_VALNOTVALID*/ocfs2_meta_lvb_is_trustable()return 1 /* so we don't refresh inode from disk */
ocfs2_truncate_file()mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */The fix is quite straightforward. We keep to set
DLM_LKF_VALBLKflag fordlm_lock()if the lock resource type needs LVB and the fsdlm plugin is uesed.Signed-off-by Eric Ren <[email protected]>--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)fs/ocfs2/dlmglue.c +10
-0@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); +/* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always +* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get +* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set. +*/ +if (!ocfs2_is_o2cb_active () && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb =1 ; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ;fs/ocfs2/stackglue.c +6
-0@@ -48,6 +48,12 @@ static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";*/ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active (void ) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {struct ocfs2_stack_plugin *p;fs/ocfs2/stackglue.h +3
-0@@ -298,6 +298,9 @@ void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_pint ocfs2_stack_glue_register (struct ocfs2_stack_plugin *plugin);void ocfs2_stack_glue_unregister (struct ocfs2_stack_plugin *plugin); +/* In ocfs2_downconvert_lock(), we need to know which stack we are using */ +int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */
Hi,
Quoted 3 lines · Joseph Qi
In you description, this issue can only happen in case of stack user +
fsdlm.
Yes.
Quoted 4 lines
So I feel we'd better to make stack user and o2cb behaves the same,
other than treat it as a special case.
Yes, I agree. But, actually, there is nothing wrong with fsdlm. I think o2cb does some tricks with DLM_LKF_VALBLK flag in such a special case where down conversion is PR->NULL.
I'd like to see this quick and small fix to be merged at this moment, because this issue is little emergency for us.
Anyway, we can supersede this one easily if someone familiar with o2cb works out a patch for o2cb in the future.
Does this sounds good to you?
Thanks,
Eric
Quoted 191 lines
Thanks,
Joseph
Quoted 182 lines · Eric Ren
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN) ocfs2_nodemanager
ocfs2_stackglue(OEN)quota_treedlm(OEN) configfs fusesd_modiscsi_tcplibiscsi_tcp
libiscsiscsi_transport_iscsiaf_packetiscsi_ibftiscsi_boot_sysfssoftdog xfs libcrc32c ppdevparport_pcpcspkr parport joydevvirtio_balloonvirtio_neti2c_piix4acpi_cpufreqbutton processor ext4 crc16 jbd2 mbcacheata_genericcirrusvirtio_blkata_piixdrm_kms_helperahci syscopyarea libahci sysfillrect sysimgbltfb_sys_fopsttm floppy libata drmvirtio_pcivirtio_ringuhci_hcdvirtioehci_hcdusbcoreserio_rawusb_commonsgdm_multipathdm_modscsi_dh_rdacscsi_dh_emcscsi_dh_aluascsi_modautofs4[ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>"
It's because
ocfs2_inode_lock()get us stale LVB in which thei_sizeis not equal to the diski_size. We mistakenly trust the LVB because the underlaying fsdlmdlm_lock()doesn't setlkb_sbflagswithDLM_SBF_VALNOTVALIDproperly for us. But, why?The current code tries to downconvert lock without
DLM_LKF_VALBLKflag to tell o2cb don't update RSB's LVB if it's aPR->NULLconversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.The fsdlm plugin behaves different on
DLM_LKF_VALBLK, it depends onDLM_LKF_VALBLKto decide if we care about the LVB in the LKB. IfDLM_LKF_VALBLKis not set, fsdlm will skip recovering RSB's LVB from this lkb and set the rightDLM_SBF_VALNOTVALIDappropriately when node failure happens.The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with
LOCK_TYPE_USES_LVB;The 1st round:
Node1 Node2
RSB1: PRRSB1(master): NULL->EX
ocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwritelkb->lkb_exflags
with noDLM_LKF_VALBLK)RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should setRSB_VALNOTVALIDfor RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK))/* This means we miss the chance to
return; * to invalid the LVB here.
*/The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)
/*dlm_lock()return the stale lvb without settingDLM_SBF_VALNOTVALID*/ocfs2_meta_lvb_is_trustable()return 1 /* so we don't refresh inode from disk */
ocfs2_truncate_file()mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */The fix is quite straightforward. We keep to set
DLM_LKF_VALBLKflag fordlm_lock()if the lock resource type needs LVB and the fsdlm plugin is uesed.Signed-off-by Eric Ren <[email protected]>--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)fs/ocfs2/dlmglue.c +9
-0@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); +/* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always +* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get +* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set. +*/ + if (!ocfs2_is_o2cb_active() && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb = 1; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ; diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.cindex 52c07346b..8203590 100644
fs/ocfs2/stackglue.c +5
-0@@ -48,6 +48,12 @@ static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] ="/sbin/ocfs2_hb_ctl"; */ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active(void) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {struct ocfs2_stack_plugin *p;fs/ocfs2/stackglue.h +0
-0@@ -298,6 +298,9 @@ void ocfs2_stack_glue_set_max_proto_version(struct
ocfs2_protocol_version*max_p
intocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin); voidocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin); +/* Inocfs2_downconvert_lock(), we need to know which stack we are using */fs/ocfs2/stackglue.h +2
-0+int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */
Joseph Qi
Reviewed-byQuoted 16 lines · Eric Ren
Hi,
Quoted 3 lines · Joseph Qi
In you description, this issue can only happen in case of stack user +
fsdlm.
Yes.
Quoted 4 lines
So I feel we'd better to make stack user and o2cb behaves the same,
other than treat it as a special case.
Yes, I agree. But, actually, there is nothing wrong with fsdlm. I think o2cb does some tricks
withDLM_LKF_VALBLKflag in such a special case where down conversion isPR->NULL.
Quoted 6 lines
I'd like to see this quick and small fix to be merged at this moment, because this issue is little emergency for us.
Anyway, we can supersede this one easily if someone familiar with o2cb works out a patch for o2cb in the future.Does this sounds good to you?
Fine.
Thanks,
Joseph
Quoted 234 lines
Thanks,
EricQuoted 230 lines
Thanks,
Joseph
Quoted 221 lines · Eric Ren
The crash happens rather often when we reset some cluster nodes while nodes contend fiercely to do truncate and append.
The crash backtrace is below:
"[ 245.197849] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover_grant 1 locks on 971 resources [ 245.197859] dlm: C21CBDA5E0774F4BA5A9D4F317717495: dlm_recover 9 generation 5 done: 4 ms [ 245.198379] ocfs2: Begin replay journal (node 318952601, slot 2) on device (253,18) [ 247.272338] ocfs2: End replay journal (node 318952601, slot 2) on device (253,18) [ 247.547084] ocfs2: Beginning quota recovery on device (253,18) for slot 2 [ 247.683263] ocfs2: Finishing quota recovery on device (253,18) for slot 2 [ 247.833022] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: bug expression: le64_to_cpu(fe->i_size) != i_size_read(inode) [ 247.833029] (truncate,30154,1):ocfs2_truncate_file:470 ERROR: Inode 290321, inode i_size = 732 != di i_size = 937, i_flags = 0x1 [ 247.833074] ------------[ cut here ]------------ [ 247.833077] kernel BUG at /usr/src/linux/fs/ocfs2/file.c:470! [ 247.833079] invalid opcode: 0000 [#1] SMP [ 247.833081] Modules linked in: ocfs2_stack_user(OEN) ocfs2(OEN)
ocfs2_nodemanagerocfs2_stackglue(OEN)quota_treedlm(OEN) configfs
fusesd_modiscsi_tcplibiscsi_tcplibiscsiscsi_transport_iscsiaf_packetiscsi_ibftiscsi_boot_sysfssoftdog xfs libcrc32c ppdevparport_pcpcspkr parport joydevvirtio_balloonvirtio_neti2c_piix4acpi_cpufreqbutton processor ext4 crc16 jbd2 mbcacheata_genericcirrusvirtio_blkata_piixdrm_kms_helperahci syscopyarea libahci sysfillrect sysimgbltfb_sys_fopsttm floppy libata drmvirtio_pcivirtio_ringuhci_hcdvirtioehci_hcdusbcoreserio_rawusb_commonsgdm_multipathdm_modscsi_dh_rdacscsi_dh_emcscsi_dh_aluascsi_modautofs4[ 247.833107] Supported: No, Unsupported modules are loaded [ 247.833110] CPU: 1 PID: 30154 Comm: truncate Tainted: G OE N 4.4.21-69-default #1 [ 247.833111] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.8.1-0-g4adadbd-20151112_172657-sheep25 04/01/2014 [ 247.833112] task: ffff88004ff6d240 ti: ffff880074e68000 task.ti: ffff880074e68000 [ 247.833113] RIP: 0010:[<ffffffffa05c8c30>] [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.833151] RSP: 0018:ffff880074e6bd50 EFLAGS: 00010282 [ 247.833152] RAX: 0000000000000074 RBX: 000000000000029e RCX: 0000000000000000 [ 247.833153] RDX: 0000000000000001 RSI: 0000000000000246 RDI: 0000000000000246 [ 247.833154] RBP: ffff880074e6bda8 R08: 000000003675dc7a R09: ffffffff82013414 [ 247.833155] R10: 0000000000034c50 R11: 0000000000000000 R12: ffff88003aab3448 [ 247.833156] R13: 00000000000002dc R14: 0000000000046e11 R15: 0000000000000020 [ 247.833157] FS: 00007f839f965700(0000) GS:ffff88007fc80000(0000) knlGS:0000000000000000 [ 247.833158] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [ 247.833159] CR2: 00007f839f97e000 CR3: 0000000036723000 CR4: 00000000000006e0 [ 247.833164] Stack: [ 247.833165] 00000000000003a9 0000000000000001 ffff880060554000 ffff88004fcaf000 [ 247.833167] ffff88003aa7b090 1000000000000000 ffff88003aab3448 ffff880074e6beb0 [ 247.833169] 0000000000000001 0000000000002068 0000000000000020 0000000000000000 [ 247.833171] Call Trace: [ 247.833208] [<ffffffffa05cb1a8>] ocfs2_setattr+0x698/0xa90 [ocfs2] [ 247.833225] [<ffffffff8121587e>] notify_change+0x1ae/0x380 [ 247.833242] [<ffffffff811f8bfe>] do_truncate+0x5e/0x90 [ 247.833246] [<ffffffff811f8f58>] do_sys_ftruncate.constprop.11+0x108/0x160 [ 247.833257] [<ffffffff815e142e>] entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834724] DWARF2 unwinder stuck at entry_SYSCALL_64_fastpath+0x12/0x6d [ 247.834725] [ 247.834726] Leftover inexact backtrace:[ 247.834728] Code: 24 28 ba d6 01 00 00 48 c7 c6 30 43 62 a0 8b 41 2c 89 44 24 08 48 8b 41 20 48 c7 c1 78 a3 62 a0 48 89 04 24 31 c0 e8 a0 97 f9 ff <0f> 0b 3d 00 fe ff ff 0f 84 ab fd ff ff 83 f8 fc 0f 84 a2 fd ff [ 247.834748] RIP [<ffffffffa05c8c30>] ocfs2_truncate_file+0x640/0x6c0 [ocfs2] [ 247.834774] RSP <ffff880074e6bd50>"
It's because
ocfs2_inode_lock()get us stale LVB in which thei_sizeis not
equal to the diski_size. We mistakenly trust the LVB because the underlaying
fsdlmdlm_lock()doesn't setlkb_sbflagswithDLM_SBF_VALNOTVALIDproperly for
us. But, why?The current code tries to downconvert lock without
DLM_LKF_VALBLKflag to tell o2cb don't update RSB's LVB if it's aPR->NULLconversion, even if the lock resource type needs LVB. This is not the right way for fsdlm.The fsdlm plugin behaves different on
DLM_LKF_VALBLK, it depends onDLM_LKF_VALBLKto decide if we care about the LVB in the LKB. IfDLM_LKF_VALBLK
is not set, fsdlm will skip recovering RSB's LVB from this lkb and set the rightDLM_SBF_VALNOTVALIDappropriately when node failure happens.The following diagram briefly illustrates how this crash happens:
RSB1 is inode metadata lock resource with
LOCK_TYPE_USES_LVB;The 1st round:
Node1 Node2
RSB1: PR
RSB1(master):NULL->EXocfs2_downconvert_lock(PR->NULL, set_lvb==0)
ocfs2_dlm_lock(no DLM_LKF_VALBLK)- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
dlm_lock(no DLM_LKF_VALBLK)
convert_lock(overwritelkb->lkb_exflags
with noDLM_LKF_VALBLK)RSB1: NULL RSB1: EX
reset Node2dlm_recover_rsbs()
recover_lvb()/* The LVB is not trustable if the node with EX fails and
* no lock >= PR is left. We should setRSB_VALNOTVALIDfor RSB1. */
if(!(kb_exflags & DLM_LKF_VALBLK))/* This means we miss the
chance to
return; * to invalid the LVB here.
*/The 2nd round:
Node 1 Node2
RSB1(become master from recovery)
ocfs2_setattr()
ocfs2_inode_lock(NULL->EX)/* dlm_lock() return the stale lvb without setting
DLM_SBF_VALNOTVALID*/
ocfs2_meta_lvb_is_trustable()return 1 /* so we don't refresh
inode from disk */
ocfs2_truncate_file()mlog_bug_on_msg(disk isize != i_size_read(inode)) /* crash! */The fix is quite straightforward. We keep to set
DLM_LKF_VALBLKflag fordlm_lock()
if the lock resource type needs LVB and the fsdlm plugin is uesed.Signed-off-by Eric Ren <[email protected]>--- fs/ocfs2/dlmglue.c | 10 ++++++++++ fs/ocfs2/stackglue.c | 6 ++++++ fs/ocfs2/stackglue.h | 3 +++ 3 files changed, 19 insertions(+)fs/ocfs2/dlmglue.c +1
-0@@ -3303,6 +3303,16 @@ static int ocfs2_downconvert_lock(struct ocfs2_super *osb,mlog (ML_BASTS ,"lockres %s, level %d => %d\n" , lockres->l_name, lockres->l_level, new_level); + /* +* On DLM_LKF_VALBLK, fsdlm behaves differently with o2cb. It always
fs/ocfs2/dlmglue.c +2
-0+* expects DLM_LKF_VALBLK being set if the LKB has LVB, so that +* we can recover correctly from node failure. Otherwise, we may get
fs/ocfs2/dlmglue.c +6
-0+* invalid LVB in LKB, but without DLM_SBF_VALNOTVALID being set. +*/ +if (!ocfs2_is_o2cb_active () && + lockres->l_ops->flags &LOCK_TYPE_USES_LVB ) + lvb =1 ; +if (lvb) dlm_flags |=DLM_LKF_VALBLK ; diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.cindex 52c07346b..8203590 100644
fs/ocfs2/stackglue.c +5
-0@@ -48,6 +48,12 @@ static charocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl"; */ static struct ocfs2_stack_plugin *active_stack; +inline int ocfs2_is_o2cb_active(void) +{ +return !strcmp (active_stack->sp_name,OCFS2_STACK_PLUGIN_O2CB ); +} +EXPORT_SYMBOL_GPL (ocfs2_is_o2cb_active); +static struct ocfs2_stack_plugin *ocfs2_stack_lookup (const char *name) {struct ocfs2_stack_plugin *p;fs/ocfs2/stackglue.h +0
-0@@ -298,6 +298,9 @@ voidocfs2_stack_glue_set_max_proto_version(struct
ocfs2_protocol_version*max_p
intocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin); voidocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin); +/* Inocfs2_downconvert_lock(), we need to know which stack we
are using */fs/ocfs2/stackglue.h +2
-0+int ocfs2_is_o2cb_active (void ); +extern struct kset *ocfs2_kset;#endif /* STACKGLUE_H */
Quoted 9 lines · On Mon, 19 Dec 2016 17:55:06 +0800 Joseph Qi <[email protected]> wrote:
Quoted 6 lines
I'd like to see this quick and small fix to be merged at this moment, because this issue is little emergency for us.
Anyway, we can supersede this one easily if someone familiar with o2cb works out a patch for o2cb in the future.Does this sounds good to you?
Fine.
Reviewed-by Joseph Qi <[email protected]>
I'll add cc:stable to this one so the fix gets backported into older kernels.