nfsd: add new tracepoint in nfsd_open_break_lease
Subsystems (2)
Jeff Layton
maintainerPATCH 1/3 nfsd: add new tracepoint in nfsd_open_break_lease
--- fs/nfsd/trace.h | 16 ++++++++++++++++ fs/nfsd/vfs.c | 2 ++ 2 files changed, 18 insertions(+)
fs/nfsd/trace.h +16 -0
@@ -531,6 +531,22 @@ DEFINE_STATEID_EVENT(open);DEFINE_STATEID_EVENT (deleg_read);DEFINE_STATEID_EVENT (deleg_recall); +TRACE_EVENT (nfsd_open_break_lease, + TP_PROTO(struct inode *inode, + int access), +TP_ARGS (inode, access), +TP_STRUCT__entry ( + __field(void *, inode) +__field (int, access) + ), +TP_fast_assign ( + __entry->inode = inode; + __entry->access = access; + ), +TP_printk ("inode=%p access=%s" , __entry->inode, +show_nfsd_may_flags (__entry->access)) +) + DECLARE_EVENT_CLASS(nfsd_stateseqid_class, TP_PROTO(u32 seqid, const stateid_t *stp), TP_ARGS(seqid, stp),
fs/nfsd/vfs.c +2 -0
@@ -729,6 +729,8 @@ int nfsd_open_break_lease(struct inode *inode, int access) {unsigned int mode; +trace_nfsd_open_break_lease (inode, access); +if (access &NFSD_MAY_NOT_BREAK_LEASE )return 0 ; mode = (access &NFSD_MAY_WRITE ) ?O_WRONLY :O_RDONLY ;
Signature · 1 line
2.37.1
Jeff Layton
maintainerPATCH 2/3 nfsd: print nf pointer in some nfsd_file tracepoints
--- fs/nfsd/trace.h | 28 ++++++++++++++++++---------- 1 file changed, 18 insertions(+), 10 deletions(-)
fs/nfsd/trace.h +18 -10
@@ -803,19 +803,21 @@ TRACE_EVENT(nfsd_file_alloc, ),TP_ARGS (nf),TP_STRUCT__entry ( + __field(const void *, nf)__field (const void *,nf_inode ) __field(unsigned long, nf_flags) __field(unsigned long, nf_may) __field(unsigned int, nf_ref) ),TP_fast_assign ( + __entry->nf = nf; __entry->nf_inode = nf->nf_inode; __entry->nf_flags = nf->nf_flags; __entry->nf_ref =refcount_read (&nf->nf_ref); __entry->nf_may = nf->nf_may; ), -TP_printk ("inode=%p ref=%u flags=%s may=%s" , - __entry->nf_inode, __entry->nf_ref, +TP_printk ("nf=%p inode=%p ref=%u flags=%s may=%s" , + __entry->nf, __entry->nf_inode, __entry->nf_ref,show_nf_flags (__entry->nf_flags),show_nfsd_may_flags (__entry->nf_may) )
@@ -834,6 +836,7 @@ TRACE_EVENT(nfsd_file_acquire,TP_STRUCT__entry (__field (u32 ,xid ) + __field(const void *, nf) __field(const void *, inode) __field(unsigned long, may_flags) __field(unsigned int, nf_ref)
@@ -845,6 +848,7 @@ TRACE_EVENT(nfsd_file_acquire,TP_fast_assign ( __entry->xid = be32_to_cpu(rqstp->rq_xid); + __entry->nf = nf; __entry->inode = inode; __entry->may_flags = may_flags; __entry->nf_ref = nf ?refcount_read (&nf->nf_ref) :0 ;
@@ -854,8 +858,8 @@ TRACE_EVENT(nfsd_file_acquire, __entry->status =be32_to_cpu (status); ), - TP_printk("xid=0x%x inode=%p may_flags=%s ref=%u nf_flags=%s nf_may=%s nf_file=%p status=%u", - __entry->xid, __entry->inode, + TP_printk("xid=0x%x nf=%p inode=%p may_flags=%s ref=%u nf_flags=%s nf_may=%s nf_file=%p status=%u", + __entry->xid, __entry->nf, __entry->inode, show_nfsd_may_flags(__entry->may_flags), __entry->nf_ref, show_nf_flags(__entry->nf_flags), show_nfsd_may_flags(__entry->nf_may),
@@ -873,6 +877,7 @@ TRACE_EVENT(nfsd_file_create,TP_ARGS (rqstp, may_flags, nf),TP_STRUCT__entry ( + __field(const void *, nf) __field(const void *, nf_inode) __field(const void *, nf_file) __field(unsigned long, may_flags)
@@ -883,6 +888,7 @@ TRACE_EVENT(nfsd_file_create, ), TP_fast_assign( + __entry->nf = nf; __entry->nf_inode = nf->nf_inode; __entry->nf_file = nf->nf_file; __entry->may_flags = may_flags;
@@ -892,8 +898,8 @@ TRACE_EVENT(nfsd_file_create, __entry->xid =be32_to_cpu (rqstp->rq_xid); ), -TP_printk ("xid=0x%x inode=%p may_flags=%s ref=%u nf_flags=%s nf_may=%s nf_file=%p" , - __entry->xid, __entry->nf_inode, +TP_printk ("xid=0x%x nf=%p inode=%p may_flags=%s ref=%u nf_flags=%s nf_may=%s nf_file=%p" , + __entry->xid, __entry->nf, __entry->nf_inode,show_nfsd_may_flags (__entry->may_flags), __entry->nf_ref,show_nf_flags (__entry->nf_flags), show_nfsd_may_flags(__entry->nf_may), __entry->nf_file
@@ -1060,19 +1066,21 @@ DECLARE_EVENT_CLASS(nfsd_file_gc_class, ),TP_ARGS (nf),TP_STRUCT__entry ( - __field(void *, nf_inode) -__field (void *,nf_file ) + __field(const void *, nf) +__field (const void *,nf_inode ) + __field(const void *, nf_file)__field (int, nf_ref) __field(unsigned long, nf_flags) ),TP_fast_assign ( + __entry->nf = nf; __entry->nf_inode = nf->nf_inode; __entry->nf_file = nf->nf_file; __entry->nf_ref =refcount_read (&nf->nf_ref); __entry->nf_flags = nf->nf_flags; ), -TP_printk ("inode=%p ref=%d nf_flags=%s nf_file=%p" , - __entry->nf_inode, __entry->nf_ref, +TP_printk ("nf=%p inode=%p ref=%d nf_flags=%s nf_file=%p" , + __entry->nf, __entry->nf_inode, __entry->nf_ref,show_nf_flags (__entry->nf_flags), __entry->nf_file )
Signature · 1 line
2.37.1
Jeff Layton
maintainerPATCH 3/3 nfsd: eliminate the NFSD_FILE_BREAK_* flags
In reply to Jeff LaytonWe had a report from the spring Bake-a-thon of data corruption in some nfstest_interop tests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.
Currently, we only set NFSD_FILE_BREAK_* flags if NFSD_MAY_NOT_BREAK_LEASE was set when we call nfsd_file_alloc. NFSD_MAY_NOT_BREAK_LEASE was intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.
It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.
This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call nfsd_open_break_lease every time.
--- fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)
fs/nfsd/filecache.c +3 -23
@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct nfsd_file * -nfsd_file_alloc (struct nfsd_file_lookup_key *key,unsigned int may) +nfsd_file_alloc (struct nfsd_file_lookup_key *key) {struct nfsd_file *nf;
@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;
@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;
@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags)); out:if (status == nfs_ok) {if (open)
fs/nfsd/filecache.h +1 -3
@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;
fs/nfsd/trace.h +0 -2
@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"}) DECLARE_EVENT_CLASS(nfsd_file_class,
Signature · 1 line
2.37.1
Re: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
Quoted 29 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.
Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
Quoted 95 lines
---
fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)fs/nfsd/filecache.c +0
-0@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct
nfsd_file*fs/nfsd/filecache.c +1
-1-nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) +nfsd_file_alloc(struct nfsd_file_lookup_key *key){ struct nfsd_file *nf;fs/nfsd/filecache.c +2
-22@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags));out:
if (status == nfs_ok) { if (open)fs/nfsd/filecache.h +1
-3@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;fs/nfsd/trace.h +0
-2@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"}) DECLARE_EVENT_CLASS(nfsd_file_class,Signature · 1 line
2.37.1
--
Chuck Lever
Jeff Layton
maintainerRe: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
Quoted 33 lines · Chuck Lever III
Quoted 28 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>
Quoted 9 lines
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
I based this on top of your for-next branch and I think the filecache is already in there.
It's a pretty nasty bug that we probably will want backported, so it might make sense to respin this on top of mainline and put it in ahead of the filecache overhaul.
Thoughts?
Quoted 101 lines
Quoted 95 lines
---
fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)fs/nfsd/filecache.c +0
-0@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct
nfsd_file*fs/nfsd/filecache.c +1
-1-nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) +nfsd_file_alloc(struct nfsd_file_lookup_key *key){ struct nfsd_file *nf;fs/nfsd/filecache.c +2
-22@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags));out:
if (status == nfs_ok) { if (open)fs/nfsd/filecache.h +1
-3@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;fs/nfsd/trace.h +0
-2@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"}) DECLARE_EVENT_CLASS(nfsd_file_class,Signature · 1 line
2.37.1--
Chuck Lever
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
Quoted 48 lines · Jeff Layton
Quoted 33 lines · Chuck Lever III
Quoted 28 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>Quoted 9 lines
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
Quoted 6 lines
I based this on top of your for-next branch and I think the filecache is already in there.
It's a pretty nasty bug that we probably will want backported, so it might make sense to respin this on top of mainline and put it in ahead of the filecache overhaul.
I am a generally a proponent of enabling fix backports.
I encourage you to test the respin on 5.19 and 5.18 at least because the moment that patch hits upstream, Sasha and Greg will pull it into stable. I don't relish the idea of having to fix the fix, if you catch my drift.
And perhaps when you repost, the fix should be reordered before the patches that add the tracepoints.
Quoted 1 line
Thoughts?
Rebasing all that is mechanically straightforward to do.
The only issue is that the filecache work and the first PR tag is already in my for-next branch. If you don't think it will trigger massive heartburn for Linus and Stephen, I can pull that stuff out now, and postpone my first PR until the second week of the merge window.
Quoted 91 lines
Quoted 91 lines
Quoted 91 lines
---
fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)fs/nfsd/filecache.c +0
-0@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct
nfsd_file*fs/nfsd/filecache.c +1
-1-nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) +nfsd_file_alloc(struct nfsd_file_lookup_key *key){ struct nfsd_file *nf;fs/nfsd/filecache.c +2
-22@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags));out:
if (status == nfs_ok) { if (open)fs/nfsd/filecache.h +1
-3@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;fs/nfsd/trace.h +0
-2@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"})
Quoted 13 lines
Quoted 10 lines
Quoted 4 lines
fs/nfsd/trace.h +0
-0DECLARE_EVENT_CLASS(nfsd_file_class,Signature · 1 line
2.37.1--
Chuck LeverSignature · 1 line
Jeff Layton <[email protected]>
--
Chuck Lever
Jeff Layton
maintainerRe: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
Quoted 56 lines · Chuck Lever III
Quoted 47 lines · Jeff Layton
Quoted 33 lines · Chuck Lever III
Quoted 28 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>Quoted 9 lines
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
Quoted 6 lines
I based this on top of your for-next branch and I think the filecache is already in there.
It's a pretty nasty bug that we probably will want backported, so it might make sense to respin this on top of mainline and put it in ahead of the filecache overhaul.
I am a generally a proponent of enabling fix backports.
Quoted 8 lines
I encourage you to test the respin on 5.19 and 5.18 at least because the moment that patch hits upstream, Sasha and Greg will pull it into stable. I don't relish the idea of having to fix the fix, if you catch my drift.
And perhaps when you repost, the fix should be reordered before the patches that add the tracepoints.
That all sounds good. I'll do that in a bit, and will send a v2.
Quoted 11 lines
Quoted 2 lines
Thoughts?
Rebasing all that is mechanically straightforward to do.
The only issue is that the filecache work and the first PR tag is already in my for-next branch. If you don't think it will trigger massive heartburn for Linus and Stephen, I can pull that stuff out now, and postpone my first PR until the second week of the merge window.
That may be best. I think we want to see this fix go in almost everywhere.
Quoted 111 lines
Quoted 92 lines
Quoted 92 lines
Quoted 92 lines
---
fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)fs/nfsd/filecache.c +0
-0@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct
nfsd_file*fs/nfsd/filecache.c +1
-1-nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) +nfsd_file_alloc(struct nfsd_file_lookup_key *key){ struct nfsd_file *nf;fs/nfsd/filecache.c +2
-22@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags));out:
if (status == nfs_ok) { if (open)fs/nfsd/filecache.h +1
-3@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;fs/nfsd/trace.h +0
-2@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"})Quoted 13 lines
Quoted 10 lines
Quoted 4 lines
fs/nfsd/trace.h +0
-0DECLARE_EVENT_CLASS(nfsd_file_class,Signature · 1 line
2.37.1--
Chuck LeverSignature · 1 line
Jeff Layton <[email protected]>--
Chuck Lever
Signature · 1 line
Jeff Layton <[email protected]>
Re: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
Quoted 84 lines · Jeff Layton
Quoted 56 lines · Chuck Lever III
Quoted 47 lines · Jeff Layton
Quoted 33 lines · Chuck Lever III
Quoted 28 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>Quoted 9 lines
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
Quoted 6 lines
I based this on top of your for-next branch and I think the filecache is already in there.
It's a pretty nasty bug that we probably will want backported, so it might make sense to respin this on top of mainline and put it in ahead of the filecache overhaul.
I am a generally a proponent of enabling fix backports.
Quoted 8 lines
I encourage you to test the respin on 5.19 and 5.18 at least because the moment that patch hits upstream, Sasha and Greg will pull it into stable. I don't relish the idea of having to fix the fix, if you catch my drift.
And perhaps when you repost, the fix should be reordered before the patches that add the tracepoints.
That all sounds good. I'll do that in a bit, and will send a v2.
Quoted 11 lines
Quoted 2 lines
Thoughts?
Rebasing all that is mechanically straightforward to do.
The only issue is that the filecache work and the first PR tag is already in my for-next branch. If you don't think it will trigger massive heartburn for Linus and Stephen, I can pull that stuff out now, and postpone my first PR until the second week of the merge window.
Quoted 2 lines
That may be best. I think we want to see this fix go in almost everywhere.
I will push the removal update to my public tree now. You can send your v2 over the weekend or on Monday.
Quoted 102 lines
Quoted 102 lines
Quoted 102 lines
Quoted 101 lines
Quoted 95 lines
---
fs/nfsd/filecache.c | 26 +++----------------------- fs/nfsd/filecache.h | 4 +--- fs/nfsd/trace.h | 2 -- 3 files changed, 4 insertions(+), 28 deletions(-)fs/nfsd/filecache.c +0
-0@@ -283,7 +283,7 @@ nfsd_file_mark_find_or_create(struct nfsd_file *nf, struct inode *inode) }static struct
nfsd_file*fs/nfsd/filecache.c +1
-1-nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may) +nfsd_file_alloc(struct nfsd_file_lookup_key *key){ struct nfsd_file *nf;fs/nfsd/filecache.c +2
-22@@ -301,12 +301,6 @@ nfsd_file_alloc(struct nfsd_file_lookup_key *key, unsigned int may)/* nf_ref is pre-incremented for hash table */ refcount_set (&nf->nf_ref,2 ); nf->nf_may = key->need; -if (may &NFSD_MAY_NOT_BREAK_LEASE ) { -if (may &NFSD_MAY_WRITE ) -__set_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags); -if (may &NFSD_MAY_READ ) -__set_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); - } nf->nf_mark =NULL ; }return nf;@@ -1090,7 +1084,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,if (nf) goto wait_for_construction; - new =nfsd_file_alloc (&key, may_flags); + new =nfsd_file_alloc (&key);if (!new) { status = nfserr_jukebox; goto out_status;@@ -1130,21 +1124,7 @@ nfsd_file_do_acquire(struct svc_rqst *rqstp, struct svc_fh *fhp,nfsd_file_lru_remove (nf);this_cpu_inc (nfsd_file_cache_hits); -if (!(may_flags &NFSD_MAY_NOT_BREAK_LEASE )) { -bool write = (may_flags &NFSD_MAY_WRITE ); - -if (test_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags) || - (test_bit (NFSD_FILE_BREAK_WRITE , &nf->nf_flags) && write)) { - status =nfserrno (nfsd_open_break_lease ( -file_inode (nf->nf_file), may_flags)); -if (status == nfs_ok) { -clear_bit (NFSD_FILE_BREAK_READ , &nf->nf_flags); -if (write) -clear_bit (NFSD_FILE_BREAK_WRITE , - &nf->nf_flags); - } - } - } + status =nfserrno (nfsd_open_break_lease (file_inode (nf->nf_file), may_flags));out:
if (status == nfs_ok) { if (open)fs/nfsd/filecache.h +1
-3@@ -37,9 +37,7 @@ struct nfsd_file {struct net *nf_net;#define NFSD_FILE_HASHED (0)#define NFSD_FILE_PENDING (1) -#define NFSD_FILE_BREAK_READ (2) -#define NFSD_FILE_BREAK_WRITE (3) -#define NFSD_FILE_REFERENCED (4) +#define NFSD_FILE_REFERENCED (2)unsigned long nf_flags;struct inode *nf_inode;/* don't deref */ refcount_t nf_ref;fs/nfsd/trace.h +0
-2@@ -758,8 +758,6 @@ DEFINE_CLID_EVENT(confirmed_r);__print_flags (val,"|" , \ { 1 << NFSD_FILE_HASHED, "HASHED" }, \ { 1 << NFSD_FILE_PENDING, "PENDING" }, \ - { 1 << NFSD_FILE_BREAK_READ, "BREAK_READ" }, \ - { 1 << NFSD_FILE_BREAK_WRITE, "BREAK_WRITE" }, \ { 1 << NFSD_FILE_REFERENCED, "REFERENCED"}) DECLARE_EVENT_CLASS(nfsd_file_class,Signature · 1 line
2.37.1--
Chuck Lever
Quoted 11 lines
Quoted 8 lines
Quoted 2 lines
Signature · 1 line
Jeff Layton <[email protected]>--
Chuck LeverSignature · 1 line
Jeff Layton <[email protected]>
--
Chuck Lever
Re: [PATCH 3/3] nfsd: eliminate the NFSD_FILE_BREAK_* flags
In reply to Chuck Lever IIIQuoted 61 lines · Chuck Lever III
Quoted 49 lines · Jeff Layton
Quoted 33 lines · Chuck Lever III
Quoted 28 lines · Jeff Layton
We had a report from the spring Bake-a-thon of data corruption in some
nfstest_interoptests. Looking at the traces showed the NFS server allowing a v3 WRITE to proceed while a read delegation was still outstanding.Currently, we only set NFSD_FILE_BREAK_* flags if
NFSD_MAY_NOT_BREAK_LEASEwas set when we callnfsd_file_alloc.NFSD_MAY_NOT_BREAK_LEASEwas intended to be set when finding files for COMMIT ops, where we need a writeable filehandle but don't need to break read leases.It doesn't make any sense to consult that flag when allocating a file since the file may be used on subsequent calls where we do want to break the lease (and the usage of it here seems to be reverse from what it should be anyway).
Also, after calling
nfsd_open_break_lease, we don't want to clear the BREAK_* bits. A lease could end up being set on it later (more than once) and we need to be able to break those leases as well.This means that the NFSD_FILE_BREAK_* flags now just mirror NFSD_MAY_{READ,WRITE} flags, so there's no need for them at all. Just drop those flags and unconditionally call
nfsd_open_break_leaseevery time.Quoted 4 lines
Fixes 65294c1f2c5e (nfsd: add a new struct file caching facility to nfsd)Reported-by Olga Kornieskaia <[email protected]>Signed-off-by Jeff Layton <[email protected]>Quoted 9 lines
I'm going to go out on a limb and predict this will conflict heavily with the filecache overhaul patches I have queued for next. :-)
Do you believe this is something that urgently needs to be backported to stable kernels, or can it be rebased on top of the filecache overhaul work?
Quoted 6 lines
I based this on top of your for-next branch and I think the filecache is already in there.
It's a pretty nasty bug that we probably will want backported, so it might make sense to respin this on top of mainline and put it in ahead of the filecache overhaul.
I am a generally a proponent of enabling fix backports.
Quoted 7 lines
I encourage you to test the respin on 5.19 and 5.18 at least because the moment that patch hits upstream, Sasha and Greg will pull it into stable. I don't relish the idea of having to fix the fix, if you catch my drift.
And perhaps when you repost, the fix should be reordered before the patches that add the tracepoints.
Well... let's rebase the fix, but the tracepoint patches can go in after the filecache overhaul, I think. That should make it a little easier.
--
Chuck Lever
Quoted 15 lines · Jeff Layton
Signed-off-by Jeff Layton <[email protected]>---
fs/nfsd/trace.h| 16 ++++++++++++++++fs/nfsd/vfs.c| 2 ++2 files changed, 18 insertions(+)fs/nfsd/trace.h +0
-0
fs/nfsd/trace.h DEFINE_STATEID_EVENT(open); · line 540
Quoted 10 lines
fs/nfsd/trace.h +7
-0@@ -531,6 +531,22 @@ DEFINE_STATEID_EVENT(open);DEFINE_STATEID_EVENT (deleg_read);DEFINE_STATEID_EVENT (deleg_recall); +TRACE_EVENT(nfsd_open_break_lease, + TP_PROTO(struct inode *inode, + int access), + TP_ARGS(inode, access), + TP_STRUCT__entry( + __field(void *, inode) + __field(int, access)
trace_print_symbols_seq() takes an unsigned long, so I prefer to store values that are passed to it in an unsigned long field to make the type conversion more obvious to readers.
Quoted 17 lines
fs/nfsd/trace.h +9
-0+ ), + TP_fast_assign( + __entry->inode = inode; + __entry->access = access; + ), + TP_printk("inode=%p access=%s", __entry->inode, + show_nfsd_may_flags(__entry->access)) +) + DECLARE_EVENT_CLASS(nfsd_stateseqid_class, TP_PROTO(u32 seqid, const stateid_t *stp), TP_ARGS(seqid, stp),fs/nfsd/vfs.c +0
-0
fs/nfsd/vfs.c int nfsd_open_break_lease(struct inode *inode, int access) · line 733
Quoted 5 lines
fs/nfsd/vfs.c +2
-0@@ -729,6 +729,8 @@ int nfsd_open_break_lease(struct inode *inode, int access) {unsigned int mode; +trace_nfsd_open_break_lease (inode, access); +
IMO this tracepoint should include the incoming XID as well. I checked: each caller of nfsd_open_break_lease() has an svc_rqst pointer to pass in here, and that can be passed along to the tracepoint.
Thus if we're going to change the synopsis of nfsd_open_break_lease() perhaps it would be better to return a __be32 nfserr status, since every one of its call sites appears to convert the returned hosterr value to an nfserr.
Quoted 6 lines
fs/nfsd/vfs.c +0
-0if (access &NFSD_MAY_NOT_BREAK_LEASE )return 0 ; mode = (access &NFSD_MAY_WRITE ) ?O_WRONLY :O_RDONLY ;Signature · 1 line
2.37.1
--
Chuck Lever