OCFS2: anti stale inode for nfs (for 1.4git)
PATCH 1/1 OCFS2: anti stale inode for nfs (for 1.4git)
Back porting from mainline.
For nfs exporting, ocfs2_get_dentry() returns the dentry for fh. ocfs2_get_dentry() may read from disk(when inode not in memory) without any cross cluster lock. this leads to load a stale inode.
this patch fixes above problem.
solution is that in case of inode is not in memory, we get the cluster lock(PR) of alloc inode where the inode in question is allocated from(this causes node on which deletion is done sync the alloc inode) before reading out the inode its self. then we check the bitmap in the group(the inode in question allcated from) to see if the bit is clear. if it's clear then it's stale. if the bit is set, we then check generation as the existing code does.
actually we have to read out the inode in question from disk(not cache) first to know its alloc slot(tells alloc inode) and allot bit(tells alloc group). and if its not stale(by above logic) we read it out using ocfs2_iget(). the second read should from cache.
and also we have to add a per superblock nfs_sync_lock to cover the lock for alloc inode and that for inode in question. this is because ocfs2_get_dentry() and ocfs2_delete_inode() lock on them in reverse order. nfs_sync_lock is locked in EX mode in ocfs2_get_dentry() and in PR mode in ocfs2_delete_inode(). so that mutliple ocfs2_delete_inode() can run concurrently in normal case.
--
dlmglue.c | 45 ++++++++++++++++ dlmglue.h | 2 export.c | 79 +++++++++++++++++++++++++--- inode.c | 25 ++++++++- inode.h | 1 ocfs2.h | 1 ocfs2_lockid.h | 4 + suballoc.c | 157 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++ suballoc.h | 2 9 files changed, 305 insertions(+), 11 deletions(-)
fs/ocfs2/dlmglue.c +45 -0
@@ -246,6 +246,10 @@ static struct ocfs2_lock_res_ops ocfs2_rename_lops = { .flags = 0, }; +static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = { + .flags =0 , +}; +static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { .get_osb = ocfs2_get_dentry_osb, .post_unlock = ocfs2_dentry_post_unlock,
@@ -624,6 +628,17 @@ static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, &ocfs2_rename_lops, osb); } +static void ocfs2_nfs_sync_lock_res_init (struct ocfs2_lock_res *res, +struct ocfs2_super *osb) +{ +/* nfs_sync lockres doesn't come from a slab so we call init +* once on it manually. */ +ocfs2_lock_res_init_once (res); +ocfs2_build_lock_name (OCFS2_LOCK_TYPE_NFS_SYNC ,0 ,0 , res->l_name); +ocfs2_lock_res_init_common (osb, res,OCFS2_LOCK_TYPE_NFS_SYNC , + &ocfs2_nfs_sync_lops, osb); +} +void ocfs2_file_lock_res_init (struct ocfs2_lock_res *lockres,struct ocfs2_file_private *fp) {
@@ -2290,6 +2305,33 @@ void ocfs2_rename_unlock(struct ocfs2_super *osb)ocfs2_cluster_unlock (osb, lockres,LKM_EXMODE ); } +int ocfs2_nfs_sync_lock (struct ocfs2_super *osb,int ex) +{ +int status; +struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; + +if (ocfs2_is_hard_readonly (osb)) +return -EROFS ; + +if (ocfs2_mount_local (osb)) +return 0 ; + + status =ocfs2_cluster_lock (osb, lockres, ex ?LKM_EXMODE :LKM_PRMODE , +0 ,0 ); +if (status <0 ) +mlog (ML_ERROR ,"lock on nfs sync lock failed %d\n" , status); +return status; +} + +void ocfs2_nfs_sync_unlock (struct ocfs2_super *osb,int ex) +{ +struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; + +if (!ocfs2_mount_local (osb)) +ocfs2_cluster_unlock (osb, lockres, + ex ?LKM_EXMODE :LKM_PRMODE ); +} +int ocfs2_dentry_lock (struct dentry *dentry,int ex) {int ret;
@@ -2668,6 +2710,7 @@ int ocfs2_dlm_init(struct ocfs2_super *osb) local:ocfs2_super_lock_res_init (&osb->osb_super_lockres, osb);ocfs2_rename_lock_res_init (&osb->osb_rename_lockres, osb); +ocfs2_nfs_sync_lock_res_init (&osb->osb_nfs_sync_lockres, osb); osb->dlm = dlm;
@@ -2698,6 +2741,7 @@ void ocfs2_dlm_shutdown(struct ocfs2_super *osb)ocfs2_lock_res_free (&osb->osb_super_lockres);ocfs2_lock_res_free (&osb->osb_rename_lockres); +ocfs2_lock_res_free (&osb->osb_nfs_sync_lockres);dlm_unregister_domain (osb->dlm); osb->dlm =NULL ;
@@ -2892,6 +2936,7 @@ static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) {ocfs2_simple_drop_lockres (osb, &osb->osb_super_lockres);ocfs2_simple_drop_lockres (osb, &osb->osb_rename_lockres); +ocfs2_simple_drop_lockres (osb, &osb->osb_nfs_sync_lockres); }int ocfs2_drop_inode_locks (struct inode *inode)
fs/ocfs2/dlmglue.h +2 -0
@@ -99,6 +99,8 @@ void ocfs2_super_unlock(struct ocfs2_super *osb, int ex);int ocfs2_rename_lock (struct ocfs2_super *osb);void ocfs2_rename_unlock (struct ocfs2_super *osb); +int ocfs2_nfs_sync_lock (struct ocfs2_super *osb,int ex); +void ocfs2_nfs_sync_unlock (struct ocfs2_super *osb,int ex);int ocfs2_dentry_lock (struct dentry *dentry,int ex);void ocfs2_dentry_unlock (struct dentry *dentry,int ex);int ocfs2_file_lock (struct file *file,int ex,int trylock);
fs/ocfs2/export.c +69 -10
109 lines not on this page · raw message
fs/ocfs2/inode.c +24 -1
@@ -111,6 +111,18 @@ void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi) oi->ip_attr |=OCFS2_DIRSYNC_FL ; } +struct inode *ocfs2_ilookup (struct super_block *sb,u64 blkno) +{ +struct ocfs2_find_inode_args args; + + args.fi_blkno = blkno; + args.fi_flags =0 ; + args.fi_ino =ino_from_blkno (sb, blkno); + args.fi_sysfile_type =0 ; + +return ilookup5 (sb, blkno, ocfs2_find_actor, &args); +} +struct inode *ocfs2_iget (struct ocfs2_super *osb,u64 blkno,unsigned flags,int sysfile_type) {
@@ -933,6 +945,13 @@ void ocfs2_delete_inode(struct inode *inode) goto bail; } +/* Lock down the nfs_sync lock in PR mode */ + status =ocfs2_nfs_sync_lock (OCFS2_SB (inode->i_sb),0 ); +if (status <0 ) { +mlog (ML_ERROR ,"getting nfs sync lock(PR) failed %d\n" , status); +ocfs2_cleanup_delete_inode (inode,0 ); + goto bail_unblock; + }/* Lock down the inode. This gives us an up to date view of * it's metadata (for verification), and allows us to * serialize delete_inode on multiple nodes.
@@ -946,7 +965,7 @@ void ocfs2_delete_inode(struct inode *inode)if (status != -ENOENT )mlog_errno (status);ocfs2_cleanup_delete_inode (inode,0 ); - goto bail_unblock; + goto bail_unlock_nfs_sync; }/* Query the cluster. This will be the final decision made
@@ -989,6 +1008,10 @@ void ocfs2_delete_inode(struct inode *inode) bail_unlock_inode:ocfs2_inode_unlock (inode,1 );brelse (di_bh); + +bail_unlock_nfs_sync: +ocfs2_nfs_sync_unlock (OCFS2_SB (inode->i_sb),0 ); + bail_unblock: status =sigprocmask (SIG_SETMASK , &oldset,NULL );if (status <0 )
fs/ocfs2/inode.h +1 -0
@@ -126,6 +126,7 @@ void ocfs2_drop_inode(struct inode *inode);/* Flags for ocfs2_iget() */ #define OCFS2_FI_FLAG_SYSFILE 0x1#define OCFS2_FI_FLAG_ORPHAN_RECOVERY 0x2 +struct inode *ocfs2_ilookup (struct super_block *sb,u64 feoff);struct inode *ocfs2_iget (struct ocfs2_super *osb,u64 feoff,unsigned flags,int sysfile_type);int ocfs2_inode_init_private (struct inode *inode);
fs/ocfs2/ocfs2.h +1 -0
@@ -288,6 +288,7 @@ struct ocfs2_superstruct dlm_ctxt *dlm;struct ocfs2_lock_res osb_super_lockres;struct ocfs2_lock_res osb_rename_lockres; +struct ocfs2_lock_res osb_nfs_sync_lockres;struct dlm_eviction_cb osb_eviction_cb;struct ocfs2_dlm_debug *osb_dlm_debug;struct dlm_protocol_version osb_locking_proto;
fs/ocfs2/ocfs2_lockid.h +4 -0
@@ -46,6 +46,7 @@ enum ocfs2_lock_type {OCFS2_LOCK_TYPE_DENTRY ,OCFS2_LOCK_TYPE_OPEN ,OCFS2_LOCK_TYPE_FLOCK , +OCFS2_LOCK_TYPE_NFS_SYNC , OCFS2_NUM_LOCK_TYPES };
@@ -77,6 +78,9 @@ static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)case OCFS2_LOCK_TYPE_FLOCK : c ='F' ;break ; +case OCFS2_LOCK_TYPE_NFS_SYNC : + c ='Y' ; +break ;default : c ='\0' ; }
fs/ocfs2/suballoc.c +157 -0
160 lines not on this page · raw message
fs/ocfs2/suballoc.h +2 -0
@@ -157,4 +157,6 @@ u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster);int ocfs2_check_group_descriptor (struct super_block *sb,struct ocfs2_dinode *di,struct ocfs2_group_desc *gd); + +int ocfs2_test_inode_bit (struct ocfs2_super *osb,u64 blkno,int *res);#endif /* _CHAINALLOC_H_ */
PATCH 1/1 OCFS2: anti stale inode for nfs (for 1.4git)
Quoted 19 lines · wengang wang
Back porting from mainline.
For nfs exporting,
ocfs2_get_dentry()returns the dentry for fh.ocfs2_get_dentry()may read from disk(when inode not in memory) without any cross cluster lock. this leads to load a stale inode.this patch fixes above problem.
solution is that in case of inode is not in memory, we get the cluster lock(PR) of alloc inode where the inode in question is allocated from(this causes node on which deletion is done sync the alloc inode) before reading out the inode its self. then we check the bitmap in the group(the inode in question allcated from) to see if the bit is clear. if it's clear then it's stale. if the bit is set, we then check generation as the existing code does.actually we have to read out the inode in question from disk(not cache) first to know its alloc slot(tells alloc inode) and allot bit(tells alloc group). and if its not stale(by above logic) we read it out using
ocfs2_iget(). the second read should from cache.
Quoted 7 lines
and also we have to add a per superblock
nfs_sync_lockto cover the lock for alloc inode and that for inode in question. this is becauseocfs2_get_dentry()andocfs2_delete_inode()lock on them in reverse order.nfs_sync_lockis locked in EX mode inocfs2_get_dentry()and in PR mode inocfs2_delete_inode(). so that mutlipleocfs2_delete_inode()can run concurrently in normal case.Signed-off-by Wengang Wang <[email protected]>
I think this looks good. Is there any way a customer with this
problem can test the 1.4 backport?
Joel
Signature · 9 lines
Life's Little Instruction Book #173 "Be kinder than necessary." Joel Becker Principal Software Developer Oracle E-mail: joel.becker at oracle.com Phone: (650) 506-8127
PATCH 1/1 OCFS2: anti stale inode for nfs (for 1.4git)
Quoted 3 lines
Is there any way a customer with this
problem can test the 1.4 backport?
no luck to that. the cust is hitting the problem on 1.2 and discussing about upgrading to EL5 + ocfs2-1.4.
hope we can take the test before next release including the patch.
#I have note supports to let me know it if some other customers hit it.
regards,
wengang.