NFSD add COPY_NOTIFY operation
authorOlga Kornievskaia <olga.kornievskaia@gmail.com>
Fri, 4 Oct 2019 20:34:26 +0000 (16:34 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Mon, 9 Dec 2019 16:42:14 +0000 (11:42 -0500)
Introducing the COPY_NOTIFY operation.

Create a new unique stateid that will keep track of the copy
state and the upcoming READs that will use that stateid.
Each associated parent stateid has a list of copy
notify stateids. A copy notify structure makes a copy of
the parent stateid and a clientid and will use it to look
up the parent stateid during the READ request (suggested
by Trond Myklebust <trond.myklebust@hammerspace.com>).

At nfs4_put_stid() time, we walk the list of the associated
copy notify stateids and delete them.

Laundromat thread will traverse globally stored copy notify
stateid in idr and notice if any haven't been referenced in the
lease period, if so, it'll remove them.

Return single netaddr to advertise to the copy.

Suggested-by: Trond Myklebust <trond.myklebust@hammerspace.com>
Signed-off-by: Olga Kornievskaia <kolga@netapp.com>
Signed-off-by: Andy Adamson <andros@netapp.com>
fs/nfsd/nfs4proc.c
fs/nfsd/nfs4state.c
fs/nfsd/state.h
fs/nfsd/xdr4.h

index d3f6056..dee5aec 100644 (file)
@@ -37,6 +37,7 @@
 #include <linux/falloc.h>
 #include <linux/slab.h>
 #include <linux/kthread.h>
+#include <linux/sunrpc/addr.h>
 
 #include "idmap.h"
 #include "cache.h"
@@ -776,7 +777,7 @@ nfsd4_read(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        /* check stateid */
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
                                        &read->rd_stateid, RD_STATE,
-                                       &read->rd_nf);
+                                       &read->rd_nf, NULL);
        if (status) {
                dprintk("NFSD: nfsd4_read: couldn't process stateid!\n");
                goto out;
@@ -948,7 +949,7 @@ nfsd4_setattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        if (setattr->sa_iattr.ia_valid & ATTR_SIZE) {
                status = nfs4_preprocess_stateid_op(rqstp, cstate,
                                &cstate->current_fh, &setattr->sa_stateid,
-                               WR_STATE, NULL);
+                               WR_STATE, NULL, NULL);
                if (status) {
                        dprintk("NFSD: nfsd4_setattr: couldn't process stateid!\n");
                        return status;
@@ -999,7 +1000,7 @@ nfsd4_write(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        trace_nfsd_write_start(rqstp, &cstate->current_fh,
                               write->wr_offset, cnt);
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
-                                               stateid, WR_STATE, &nf);
+                                               stateid, WR_STATE, &nf, NULL);
        if (status) {
                dprintk("NFSD: nfsd4_write: couldn't process stateid!\n");
                return status;
@@ -1034,14 +1035,14 @@ nfsd4_verify_copy(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                return nfserr_nofilehandle;
 
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->save_fh,
-                                           src_stateid, RD_STATE, src);
+                                           src_stateid, RD_STATE, src, NULL);
        if (status) {
                dprintk("NFSD: %s: couldn't process src stateid!\n", __func__);
                goto out;
        }
 
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
-                                           dst_stateid, WR_STATE, dst);
+                                           dst_stateid, WR_STATE, dst, NULL);
        if (status) {
                dprintk("NFSD: %s: couldn't process dst stateid!\n", __func__);
                goto out_put_src;
@@ -1221,7 +1222,7 @@ static void dup_copy_fields(struct nfsd4_copy *src, struct nfsd4_copy *dst)
 
 static void cleanup_async_copy(struct nfsd4_copy *copy)
 {
-       nfs4_free_cp_state(copy);
+       nfs4_free_copy_state(copy);
        nfsd_file_put(copy->nf_dst);
        nfsd_file_put(copy->nf_src);
        spin_lock(&copy->cp_clp->async_lock);
@@ -1275,7 +1276,7 @@ nfsd4_copy(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                async_copy = kzalloc(sizeof(struct nfsd4_copy), GFP_KERNEL);
                if (!async_copy)
                        goto out;
-               if (!nfs4_init_cp_state(nn, copy)) {
+               if (!nfs4_init_copy_state(nn, copy)) {
                        kfree(async_copy);
                        goto out;
                }
@@ -1343,7 +1344,44 @@ static __be32
 nfsd4_copy_notify(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                  union nfsd4_op_u *u)
 {
-       return nfserr_notsupp;
+       struct nfsd4_copy_notify *cn = &u->copy_notify;
+       __be32 status;
+       struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
+       struct nfs4_stid *stid;
+       struct nfs4_cpntf_state *cps;
+       struct nfs4_client *clp = cstate->clp;
+
+       status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
+                                       &cn->cpn_src_stateid, RD_STATE, NULL,
+                                       &stid);
+       if (status)
+               return status;
+
+       cn->cpn_sec = nn->nfsd4_lease;
+       cn->cpn_nsec = 0;
+
+       status = nfserrno(-ENOMEM);
+       cps = nfs4_alloc_init_cpntf_state(nn, stid);
+       if (!cps)
+               goto out;
+       memcpy(&cn->cpn_cnr_stateid, &cps->cp_stateid.stid, sizeof(stateid_t));
+       memcpy(&cps->cp_p_stateid, &stid->sc_stateid, sizeof(stateid_t));
+       memcpy(&cps->cp_p_clid, &clp->cl_clientid, sizeof(clientid_t));
+
+       /* For now, only return one server address in cpn_src, the
+        * address used by the client to connect to this server.
+        */
+       cn->cpn_src.nl4_type = NL4_NETADDR;
+       status = nfsd4_set_netaddr((struct sockaddr *)&rqstp->rq_daddr,
+                                &cn->cpn_src.u.nl4_addr);
+       WARN_ON_ONCE(status);
+       if (status) {
+               nfs4_put_cpntf_state(nn, cps);
+               goto out;
+       }
+out:
+       nfs4_put_stid(stid);
+       return status;
 }
 
 static __be32
@@ -1355,7 +1393,7 @@ nfsd4_fallocate(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
                                            &fallocate->falloc_stateid,
-                                           WR_STATE, &nf);
+                                           WR_STATE, &nf, NULL);
        if (status != nfs_ok) {
                dprintk("NFSD: nfsd4_fallocate: couldn't process stateid!\n");
                return status;
@@ -1414,7 +1452,7 @@ nfsd4_seek(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 
        status = nfs4_preprocess_stateid_op(rqstp, cstate, &cstate->current_fh,
                                            &seek->seek_stateid,
-                                           RD_STATE, &nf);
+                                           RD_STATE, &nf, NULL);
        if (status) {
                dprintk("NFSD: nfsd4_seek: couldn't process stateid!\n");
                return status;
index 369e574..ae3dff2 100644 (file)
@@ -80,6 +80,7 @@ static u64 current_sessionid = 1;
 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
 void nfsd4_end_grace(struct nfsd_net *nn);
+static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
 
 /* Locking: */
 
@@ -722,6 +723,7 @@ struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *sla
        /* Will be incremented before return to client: */
        refcount_set(&stid->sc_count, 1);
        spin_lock_init(&stid->sc_lock);
+       INIT_LIST_HEAD(&stid->sc_cp_list);
 
        /*
         * It shouldn't be a problem to reuse an opaque stateid value.
@@ -741,30 +743,76 @@ out_free:
 /*
  * Create a unique stateid_t to represent each COPY.
  */
-int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
+static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
+                             unsigned char sc_type)
 {
        int new_id;
 
+       stid->stid.si_opaque.so_clid.cl_boot = nn->boot_time;
+       stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
+       stid->sc_type = sc_type;
+
        idr_preload(GFP_KERNEL);
        spin_lock(&nn->s2s_cp_lock);
-       new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, copy, 0, 0, GFP_NOWAIT);
+       new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
+       stid->stid.si_opaque.so_id = new_id;
        spin_unlock(&nn->s2s_cp_lock);
        idr_preload_end();
        if (new_id < 0)
                return 0;
-       copy->cp_stateid.si_opaque.so_id = new_id;
-       copy->cp_stateid.si_opaque.so_clid.cl_boot = nn->boot_time;
-       copy->cp_stateid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
        return 1;
 }
 
-void nfs4_free_cp_state(struct nfsd4_copy *copy)
+int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
+{
+       return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
+}
+
+struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
+                                                    struct nfs4_stid *p_stid)
+{
+       struct nfs4_cpntf_state *cps;
+
+       cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
+       if (!cps)
+               return NULL;
+       cps->cpntf_time = get_seconds();
+       refcount_set(&cps->cp_stateid.sc_count, 1);
+       if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
+               goto out_free;
+       spin_lock(&nn->s2s_cp_lock);
+       list_add(&cps->cp_list, &p_stid->sc_cp_list);
+       spin_unlock(&nn->s2s_cp_lock);
+       return cps;
+out_free:
+       kfree(cps);
+       return NULL;
+}
+
+void nfs4_free_copy_state(struct nfsd4_copy *copy)
 {
        struct nfsd_net *nn;
 
+       WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
        nn = net_generic(copy->cp_clp->net, nfsd_net_id);
        spin_lock(&nn->s2s_cp_lock);
-       idr_remove(&nn->s2s_cp_stateids, copy->cp_stateid.si_opaque.so_id);
+       idr_remove(&nn->s2s_cp_stateids,
+                  copy->cp_stateid.stid.si_opaque.so_id);
+       spin_unlock(&nn->s2s_cp_lock);
+}
+
+static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
+{
+       struct nfs4_cpntf_state *cps;
+       struct nfsd_net *nn;
+
+       nn = net_generic(net, nfsd_net_id);
+       spin_lock(&nn->s2s_cp_lock);
+       while (!list_empty(&stid->sc_cp_list)) {
+               cps = list_first_entry(&stid->sc_cp_list,
+                                      struct nfs4_cpntf_state, cp_list);
+               _free_cpntf_state_locked(nn, cps);
+       }
        spin_unlock(&nn->s2s_cp_lock);
 }
 
@@ -915,6 +963,7 @@ nfs4_put_stid(struct nfs4_stid *s)
                return;
        }
        idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
+       nfs4_free_cpntf_statelist(clp->net, s);
        spin_unlock(&clp->cl_lock);
        s->sc_free(s);
        if (fp)
@@ -5215,6 +5264,9 @@ nfs4_laundromat(struct nfsd_net *nn)
        struct list_head *pos, *next, reaplist;
        time_t cutoff = get_seconds() - nn->nfsd4_lease;
        time_t t, new_timeo = nn->nfsd4_lease;
+       struct nfs4_cpntf_state *cps;
+       copy_stateid_t *cps_t;
+       int i;
 
        dprintk("NFSD: laundromat service - starting\n");
 
@@ -5225,6 +5277,17 @@ nfs4_laundromat(struct nfsd_net *nn)
        dprintk("NFSD: end of grace period\n");
        nfsd4_end_grace(nn);
        INIT_LIST_HEAD(&reaplist);
+
+       spin_lock(&nn->s2s_cp_lock);
+       idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
+               cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
+               if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
+                               !time_after((unsigned long)cps->cpntf_time,
+                               (unsigned long)cutoff))
+                       _free_cpntf_state_locked(nn, cps);
+       }
+       spin_unlock(&nn->s2s_cp_lock);
+
        spin_lock(&nn->client_lock);
        list_for_each_safe(pos, next, &nn->client_lru) {
                clp = list_entry(pos, struct nfs4_client, cl_lru);
@@ -5600,6 +5663,24 @@ nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
 out:
        return status;
 }
+static void
+_free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
+{
+       WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
+       if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
+               return;
+       list_del(&cps->cp_list);
+       idr_remove(&nn->s2s_cp_stateids,
+                  cps->cp_stateid.stid.si_opaque.so_id);
+       kfree(cps);
+}
+
+void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
+{
+       spin_lock(&nn->s2s_cp_lock);
+       _free_cpntf_state_locked(nn, cps);
+       spin_unlock(&nn->s2s_cp_lock);
+}
 
 /*
  * Checks for stateid operations
@@ -5607,7 +5688,8 @@ out:
 __be32
 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
                struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
-               stateid_t *stateid, int flags, struct nfsd_file **nfp)
+               stateid_t *stateid, int flags, struct nfsd_file **nfp,
+               struct nfs4_stid **cstid)
 {
        struct inode *ino = d_inode(fhp->fh_dentry);
        struct net *net = SVC_NET(rqstp);
@@ -5656,8 +5738,12 @@ done:
        if (status == nfs_ok && nfp)
                status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
 out:
-       if (s)
-               nfs4_put_stid(s);
+       if (s) {
+               if (!status && cstid)
+                       *cstid = s;
+               else
+                       nfs4_put_stid(s);
+       }
        return status;
 }
 
index d61b83b..35eb717 100644 (file)
@@ -56,6 +56,14 @@ typedef struct {
        stateid_opaque_t        si_opaque;
 } stateid_t;
 
+typedef struct {
+       stateid_t               stid;
+#define NFS4_COPY_STID 1
+#define NFS4_COPYNOTIFY_STID 2
+       unsigned char           sc_type;
+       refcount_t              sc_count;
+} copy_stateid_t;
+
 #define STATEID_FMT    "(%08x/%08x/%08x/%08x)"
 #define STATEID_VAL(s) \
        (s)->si_opaque.so_clid.cl_boot, \
@@ -96,6 +104,7 @@ struct nfs4_stid {
 #define NFS4_REVOKED_DELEG_STID 16
 #define NFS4_CLOSED_DELEG_STID 32
 #define NFS4_LAYOUT_STID 64
+       struct list_head        sc_cp_list;
        unsigned char           sc_type;
        stateid_t               sc_stateid;
        spinlock_t              sc_lock;
@@ -104,6 +113,17 @@ struct nfs4_stid {
        void                    (*sc_free)(struct nfs4_stid *);
 };
 
+/* Keep a list of stateids issued by the COPY_NOTIFY, associate it with the
+ * parent OPEN/LOCK/DELEG stateid.
+ */
+struct nfs4_cpntf_state {
+       copy_stateid_t          cp_stateid;
+       struct list_head        cp_list;        /* per parent nfs4_stid */
+       stateid_t               cp_p_stateid;   /* copy of parent's stateid */
+       clientid_t              cp_p_clid;      /* copy of parent's clid */
+       time_t                  cpntf_time;     /* last time stateid used */
+};
+
 /*
  * Represents a delegation stateid. The nfs4_client holds references to these
  * and they are put when it is being destroyed or when the delegation is
@@ -618,14 +638,17 @@ struct nfsd4_copy;
 
 extern __be32 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
                struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
-               stateid_t *stateid, int flags, struct nfsd_file **filp);
+               stateid_t *stateid, int flags, struct nfsd_file **filp,
+               struct nfs4_stid **cstid);
 __be32 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
                     stateid_t *stateid, unsigned char typemask,
                     struct nfs4_stid **s, struct nfsd_net *nn);
 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
                                  void (*sc_free)(struct nfs4_stid *));
-int nfs4_init_cp_state(struct nfsd_net *nn, struct nfsd4_copy *copy);
-void nfs4_free_cp_state(struct nfsd4_copy *copy);
+int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy);
+void nfs4_free_copy_state(struct nfsd4_copy *copy);
+struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
+                       struct nfs4_stid *p_stid);
 void nfs4_unhash_stid(struct nfs4_stid *s);
 void nfs4_put_stid(struct nfs4_stid *s);
 void nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid);
@@ -655,6 +678,8 @@ void put_nfs4_file(struct nfs4_file *fi);
 extern void nfs4_put_copy(struct nfsd4_copy *copy);
 extern struct nfsd4_copy *
 find_async_copy(struct nfs4_client *clp, stateid_t *staetid);
+extern void nfs4_put_cpntf_state(struct nfsd_net *nn,
+                                struct nfs4_cpntf_state *cps);
 static inline void get_nfs4_file(struct nfs4_file *fi)
 {
        refcount_inc(&fi->fi_ref);
index 8231fe0..2937e06 100644 (file)
@@ -542,7 +542,7 @@ struct nfsd4_copy {
        struct nfsd_file        *nf_src;
        struct nfsd_file        *nf_dst;
 
-       stateid_t               cp_stateid;
+       copy_stateid_t          cp_stateid;
 
        struct list_head        copies;
        struct task_struct      *copy_task;