nfsd4: don't sleep in lease-break callback
[safe/jmp/linux-2.6] / fs / nfsd / nfs4state.c
index c65a27b..adc51d1 100644 (file)
@@ -1,6 +1,4 @@
 /*
-*  linux/fs/nfsd/nfs4state.c
-*
 *  Copyright (c) 2001 The Regents of the University of Michigan.
 *  All rights reserved.
 *
 *
 */
 
-#include <linux/param.h>
-#include <linux/major.h>
-#include <linux/slab.h>
-
-#include <linux/sunrpc/svc.h>
-#include <linux/nfsd/nfsd.h>
-#include <linux/nfsd/cache.h>
 #include <linux/file.h>
-#include <linux/mount.h>
-#include <linux/workqueue.h>
 #include <linux/smp_lock.h>
-#include <linux/kthread.h>
-#include <linux/nfs4.h>
-#include <linux/nfsd/state.h>
-#include <linux/nfsd/xdr4.h>
 #include <linux/namei.h>
 #include <linux/swap.h>
-#include <linux/mutex.h>
-#include <linux/lockd/bind.h>
-#include <linux/module.h>
 #include <linux/sunrpc/svcauth_gss.h>
+#include <linux/sunrpc/clnt.h>
+#include "xdr4.h"
+#include "vfs.h"
 
 #define NFSDDBG_FACILITY                NFSDDBG_PROC
 
 /* Globals */
-static time_t lease_time = 90;     /* default lease time */
-static time_t user_lease_time = 90;
+time_t nfsd4_lease = 90;     /* default lease time */
+time_t nfsd4_grace = 90;
 static time_t boot_time;
 static u32 current_ownerid = 1;
 static u32 current_fileid = 1;
@@ -182,7 +167,7 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
 {
        struct nfs4_delegation *dp;
        struct nfs4_file *fp = stp->st_file;
-       struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
+       struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
 
        dprintk("NFSD alloc_init_deleg\n");
        if (fp->fi_had_conflict)
@@ -203,10 +188,8 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
        get_file(stp->st_vfs_file);
        dp->dl_vfs_file = stp->st_vfs_file;
        dp->dl_type = type;
-       dp->dl_recall.cbr_dp = NULL;
-       dp->dl_recall.cbr_ident = cb->cb_ident;
-       dp->dl_recall.cbr_trunc = 0;
-       dp->dl_stateid.si_boot = boot_time;
+       dp->dl_ident = cb->cb_ident;
+       dp->dl_stateid.si_boot = get_seconds();
        dp->dl_stateid.si_stateownerid = current_delegid++;
        dp->dl_stateid.si_fileid = 0;
        dp->dl_stateid.si_generation = 0;
@@ -215,6 +198,7 @@ alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_f
        atomic_set(&dp->dl_count, 1);
        list_add(&dp->dl_perfile, &fp->fi_delegations);
        list_add(&dp->dl_perclnt, &clp->cl_delegations);
+       INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
        return dp;
 }
 
@@ -415,100 +399,164 @@ gen_sessionid(struct nfsd4_session *ses)
 }
 
 /*
- * Give the client the number of slots it requests bound by
- * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
+ * The protocol defines ca_maxresponssize_cached to include the size of
+ * the rpc header, but all we need to cache is the data starting after
+ * the end of the initial SEQUENCE operation--the rest we regenerate
+ * each time.  Therefore we can advertise a ca_maxresponssize_cached
+ * value that is the number of bytes in our cache plus a few additional
+ * bytes.  In order to stay on the safe side, and not promise more than
+ * we can cache, those additional bytes must be the minimum possible: 24
+ * bytes of rpc header (xid through accept state, with AUTH_NULL
+ * verifier), 12 for the compound header (with zero-length tag), and 44
+ * for the SEQUENCE op response:
+ */
+#define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
+
+/*
+ * Give the client the number of ca_maxresponsesize_cached slots it
+ * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
+ * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
+ * than NFSD_MAX_SLOTS_PER_SESSION.
  *
- * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
- * should (up to a point) re-negotiate active sessions and reduce their
- * slot usage to make rooom for new connections. For now we just fail the
- * create session.
+ * If we run out of reserved DRC memory we should (up to a point)
+ * re-negotiate active sessions and reduce their slot usage to make
+ * rooom for new connections. For now we just fail the create session.
  */
-static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
+static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
 {
-       int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
+       int mem, size = fchan->maxresp_cached;
 
-       spin_lock(&nfsd_serv->sv_lock);
-       if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
-               np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
-       nfsd_serv->sv_drc_pages_used += np;
-       spin_unlock(&nfsd_serv->sv_lock);
+       if (fchan->maxreqs < 1)
+               return nfserr_inval;
 
-       if (np <= 0) {
-               status = nfserr_resource;
-               fchan->maxreqs = 0;
-       } else
-               fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
+       if (size < NFSD_MIN_HDR_SEQ_SZ)
+               size = NFSD_MIN_HDR_SEQ_SZ;
+       size -= NFSD_MIN_HDR_SEQ_SZ;
+       if (size > NFSD_SLOT_CACHE_SIZE)
+               size = NFSD_SLOT_CACHE_SIZE;
+
+       /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
+       mem = fchan->maxreqs * size;
+       if (mem > NFSD_MAX_MEM_PER_SESSION) {
+               fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
+               if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
+                       fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
+               mem = fchan->maxreqs * size;
+       }
 
-       return status;
+       spin_lock(&nfsd_drc_lock);
+       /* bound the total session drc memory ussage */
+       if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
+               fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
+               mem = fchan->maxreqs * size;
+       }
+       nfsd_drc_mem_used += mem;
+       spin_unlock(&nfsd_drc_lock);
+
+       if (fchan->maxreqs == 0)
+               return nfserr_serverfault;
+
+       fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
+       return 0;
 }
 
 /*
  * fchan holds the client values on input, and the server values on output
+ * sv_max_mesg is the maximum payload plus one page for overhead.
  */
 static int init_forechannel_attrs(struct svc_rqst *rqstp,
-                                   struct nfsd4_session *session,
-                                   struct nfsd4_channel_attrs *fchan)
+                                 struct nfsd4_channel_attrs *session_fchan,
+                                 struct nfsd4_channel_attrs *fchan)
 {
        int status = 0;
-       __u32   maxcount = svc_max_payload(rqstp);
+       __u32   maxcount = nfsd_serv->sv_max_mesg;
 
        /* headerpadsz set to zero in encode routine */
 
        /* Use the client's max request and max response size if possible */
        if (fchan->maxreq_sz > maxcount)
                fchan->maxreq_sz = maxcount;
-       session->se_fmaxreq_sz = fchan->maxreq_sz;
+       session_fchan->maxreq_sz = fchan->maxreq_sz;
 
        if (fchan->maxresp_sz > maxcount)
                fchan->maxresp_sz = maxcount;
-       session->se_fmaxresp_sz = fchan->maxresp_sz;
-
-       /* Set the max response cached size our default which is
-        * a multiple of PAGE_SIZE and small */
-       session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
-       fchan->maxresp_cached = session->se_fmaxresp_cached;
+       session_fchan->maxresp_sz = fchan->maxresp_sz;
 
        /* Use the client's maxops if possible */
        if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
                fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
-       session->se_fmaxops = fchan->maxops;
-
-       /* try to use the client requested number of slots */
-       if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
-               fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
+       session_fchan->maxops = fchan->maxops;
 
        /* FIXME: Error means no more DRC pages so the server should
         * recover pages from existing sessions. For now fail session
         * creation.
         */
-       status = set_forechannel_maxreqs(fchan);
+       status = set_forechannel_drc_size(fchan);
 
-       session->se_fnumslots = fchan->maxreqs;
+       session_fchan->maxresp_cached = fchan->maxresp_cached;
+       session_fchan->maxreqs = fchan->maxreqs;
+
+       dprintk("%s status %d\n", __func__, status);
        return status;
 }
 
+static void
+free_session_slots(struct nfsd4_session *ses)
+{
+       int i;
+
+       for (i = 0; i < ses->se_fchannel.maxreqs; i++)
+               kfree(ses->se_slots[i]);
+}
+
+/*
+ * We don't actually need to cache the rpc and session headers, so we
+ * can allocate a little less for each slot:
+ */
+static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
+{
+       return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
+}
+
 static int
 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
                   struct nfsd4_create_session *cses)
 {
        struct nfsd4_session *new, tmp;
-       int idx, status = nfserr_resource, slotsize;
+       struct nfsd4_slot *sp;
+       int idx, slotsize, cachesize, i;
+       int status;
 
        memset(&tmp, 0, sizeof(tmp));
 
        /* FIXME: For now, we just accept the client back channel attributes. */
-       status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
+       tmp.se_bchannel = cses->back_channel;
+       status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
+                                       &cses->fore_channel);
        if (status)
                goto out;
 
-       /* allocate struct nfsd4_session and slot table in one piece */
-       slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
+       BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
+                    + sizeof(struct nfsd4_session) > PAGE_SIZE);
+
+       status = nfserr_serverfault;
+       /* allocate struct nfsd4_session and slot table pointers in one piece */
+       slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
        new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
        if (!new)
                goto out;
 
        memcpy(new, &tmp, sizeof(*new));
 
+       /* allocate each struct nfsd4_slot and data cache in one piece */
+       cachesize = slot_bytes(&new->se_fchannel);
+       for (i = 0; i < new->se_fchannel.maxreqs; i++) {
+               sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
+               if (!sp)
+                       goto out_free;
+               new->se_slots[i] = sp;
+       }
+
        new->se_client = clp;
        gen_sessionid(new);
        idx = hash_sessionid(&new->se_sessionid);
@@ -525,6 +573,10 @@ alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
        status = nfs_ok;
 out:
        return status;
+out_free:
+       free_session_slots(new);
+       kfree(new);
+       goto out;
 }
 
 /* caller must hold sessionid_lock */
@@ -567,20 +619,18 @@ release_session(struct nfsd4_session *ses)
        nfsd4_put_session(ses);
 }
 
-static void nfsd4_release_respages(struct page **respages, short resused);
-
 void
 free_session(struct kref *kref)
 {
        struct nfsd4_session *ses;
-       int i;
+       int mem;
 
        ses = container_of(kref, struct nfsd4_session, se_ref);
-       for (i = 0; i < ses->se_fnumslots; i++) {
-               struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
-               nfsd4_release_respages(e->ce_respages, e->ce_resused);
-       }
-       kfree(ses->se_slots);
+       spin_lock(&nfsd_drc_lock);
+       mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
+       nfsd_drc_mem_used -= mem;
+       spin_unlock(&nfsd_drc_lock);
+       free_session_slots(ses);
        kfree(ses);
 }
 
@@ -630,27 +680,9 @@ static struct nfs4_client *alloc_client(struct xdr_netobj name)
        return clp;
 }
 
-static void
-shutdown_callback_client(struct nfs4_client *clp)
-{
-       struct rpc_clnt *clnt = clp->cl_callback.cb_client;
-
-       if (clnt) {
-               /*
-                * Callback threads take a reference on the client, so there
-                * should be no outstanding callbacks at this point.
-                */
-               clp->cl_callback.cb_client = NULL;
-               rpc_shutdown_client(clnt);
-       }
-}
-
 static inline void
 free_client(struct nfs4_client *clp)
 {
-       shutdown_callback_client(clp);
-       nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
-                            clp->cl_slot.sl_cache_entry.ce_resused);
        if (clp->cl_cred.cr_group_info)
                put_group_info(clp->cl_cred.cr_group_info);
        kfree(clp->cl_principal);
@@ -672,9 +704,6 @@ expire_client(struct nfs4_client *clp)
        struct nfs4_delegation *dp;
        struct list_head reaplist;
 
-       dprintk("NFSD: expire_client cl_count %d\n",
-                           atomic_read(&clp->cl_count));
-
        INIT_LIST_HEAD(&reaplist);
        spin_lock(&recall_lock);
        while (!list_empty(&clp->cl_delegations)) {
@@ -703,28 +732,12 @@ expire_client(struct nfs4_client *clp)
                                 se_perclnt);
                release_session(ses);
        }
+       nfsd4_set_callback_client(clp, NULL);
+       if (clp->cl_cb_xprt)
+               svc_xprt_put(clp->cl_cb_xprt);
        put_nfs4_client(clp);
 }
 
-static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
-{
-       struct nfs4_client *clp;
-
-       clp = alloc_client(name);
-       if (clp == NULL)
-               return NULL;
-       memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
-       atomic_set(&clp->cl_count, 1);
-       atomic_set(&clp->cl_callback.cb_set, 0);
-       INIT_LIST_HEAD(&clp->cl_idhash);
-       INIT_LIST_HEAD(&clp->cl_strhash);
-       INIT_LIST_HEAD(&clp->cl_openowners);
-       INIT_LIST_HEAD(&clp->cl_delegations);
-       INIT_LIST_HEAD(&clp->cl_sessions);
-       INIT_LIST_HEAD(&clp->cl_lru);
-       return clp;
-}
-
 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
 {
        memcpy(target->cl_verifier.data, source->data,
@@ -787,6 +800,46 @@ static void gen_confirm(struct nfs4_client *clp)
        *p++ = i++;
 }
 
+static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
+               struct svc_rqst *rqstp, nfs4_verifier *verf)
+{
+       struct nfs4_client *clp;
+       struct sockaddr *sa = svc_addr(rqstp);
+       char *princ;
+
+       clp = alloc_client(name);
+       if (clp == NULL)
+               return NULL;
+
+       princ = svc_gss_principal(rqstp);
+       if (princ) {
+               clp->cl_principal = kstrdup(princ, GFP_KERNEL);
+               if (clp->cl_principal == NULL) {
+                       free_client(clp);
+                       return NULL;
+               }
+       }
+
+       memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
+       atomic_set(&clp->cl_count, 1);
+       atomic_set(&clp->cl_cb_conn.cb_set, 0);
+       INIT_LIST_HEAD(&clp->cl_idhash);
+       INIT_LIST_HEAD(&clp->cl_strhash);
+       INIT_LIST_HEAD(&clp->cl_openowners);
+       INIT_LIST_HEAD(&clp->cl_delegations);
+       INIT_LIST_HEAD(&clp->cl_sessions);
+       INIT_LIST_HEAD(&clp->cl_lru);
+       clear_bit(0, &clp->cl_cb_slot_busy);
+       rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
+       copy_verf(clp, verf);
+       rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
+       clp->cl_flavor = rqstp->rq_flavor;
+       copy_cred(&clp->cl_cred, &rqstp->rq_cred);
+       gen_confirm(clp);
+
+       return clp;
+}
+
 static int check_name(struct xdr_netobj name)
 {
        if (name.len == 0) 
@@ -894,92 +947,40 @@ find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
        return NULL;
 }
 
-/* a helper function for parse_callback */
-static int
-parse_octet(unsigned int *lenp, char **addrp)
-{
-       unsigned int len = *lenp;
-       char *p = *addrp;
-       int n = -1;
-       char c;
-
-       for (;;) {
-               if (!len)
-                       break;
-               len--;
-               c = *p++;
-               if (c == '.')
-                       break;
-               if ((c < '0') || (c > '9')) {
-                       n = -1;
-                       break;
-               }
-               if (n < 0)
-                       n = 0;
-               n = (n * 10) + (c - '0');
-               if (n > 255) {
-                       n = -1;
-                       break;
-               }
-       }
-       *lenp = len;
-       *addrp = p;
-       return n;
-}
-
-/* parse and set the setclientid ipv4 callback address */
-static int
-parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
-{
-       int temp = 0;
-       u32 cbaddr = 0;
-       u16 cbport = 0;
-       u32 addrlen = addr_len;
-       char *addr = addr_val;
-       int i, shift;
-
-       /* ipaddress */
-       shift = 24;
-       for(i = 4; i > 0  ; i--) {
-               if ((temp = parse_octet(&addrlen, &addr)) < 0) {
-                       return 0;
-               }
-               cbaddr |= (temp << shift);
-               if (shift > 0)
-               shift -= 8;
-       }
-       *cbaddrp = cbaddr;
-
-       /* port */
-       shift = 8;
-       for(i = 2; i > 0  ; i--) {
-               if ((temp = parse_octet(&addrlen, &addr)) < 0) {
-                       return 0;
-               }
-               cbport |= (temp << shift);
-               if (shift > 0)
-                       shift -= 8;
-       }
-       *cbportp = cbport;
-       return 1;
-}
-
 static void
-gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
-{
-       struct nfs4_callback *cb = &clp->cl_callback;
-
-       /* Currently, we only support tcp for the callback channel */
-       if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
+gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
+{
+       struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
+       unsigned short expected_family;
+
+       /* Currently, we only support tcp and tcp6 for the callback channel */
+       if (se->se_callback_netid_len == 3 &&
+           !memcmp(se->se_callback_netid_val, "tcp", 3))
+               expected_family = AF_INET;
+       else if (se->se_callback_netid_len == 4 &&
+                !memcmp(se->se_callback_netid_val, "tcp6", 4))
+               expected_family = AF_INET6;
+       else
                goto out_err;
 
-       if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
-                        &cb->cb_addr, &cb->cb_port)))
+       cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
+                                           se->se_callback_addr_len,
+                                           (struct sockaddr *) &cb->cb_addr,
+                                           sizeof(cb->cb_addr));
+
+       if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
                goto out_err;
+
+       if (cb->cb_addr.ss_family == AF_INET6)
+               ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
+
+       cb->cb_minorversion = 0;
        cb->cb_prog = se->se_callback_prog;
        cb->cb_ident = se->se_callback_ident;
        return;
 out_err:
+       cb->cb_addr.ss_family = AF_UNSPEC;
+       cb->cb_addrlen = 0;
        dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
                "will not receive delegations\n",
                clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
@@ -987,175 +988,87 @@ out_err:
        return;
 }
 
-void
-nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
-{
-       struct nfsd4_compoundres *resp = rqstp->rq_resp;
-
-       resp->cstate.statp = statp;
-}
-
 /*
- * Dereference the result pages.
+ * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
  */
-static void
-nfsd4_release_respages(struct page **respages, short resused)
+void
+nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
 {
-       int i;
+       struct nfsd4_slot *slot = resp->cstate.slot;
+       unsigned int base;
 
-       dprintk("--> %s\n", __func__);
-       for (i = 0; i < resused; i++) {
-               if (!respages[i])
-                       continue;
-               put_page(respages[i]);
-               respages[i] = NULL;
-       }
-}
+       dprintk("--> %s slot %p\n", __func__, slot);
 
-static void
-nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
-{
-       int i;
+       slot->sl_opcnt = resp->opcnt;
+       slot->sl_status = resp->cstate.status;
 
-       for (i = 0; i < count; i++) {
-               topages[i] = frompages[i];
-               if (!topages[i])
-                       continue;
-               get_page(topages[i]);
+       if (nfsd4_not_cached(resp)) {
+               slot->sl_datalen = 0;
+               return;
        }
+       slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
+       base = (char *)resp->cstate.datap -
+                                       (char *)resp->xbuf->head[0].iov_base;
+       if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
+                                   slot->sl_datalen))
+               WARN("%s: sessions DRC could not cache compound\n", __func__);
+       return;
 }
 
 /*
- * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
- * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
- * length of the XDR response is less than se_fmaxresp_cached
- * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
- * of the reply (e.g. readdir).
+ * Encode the replay sequence operation from the slot values.
+ * If cachethis is FALSE encode the uncached rep error on the next
+ * operation which sets resp->p and increments resp->opcnt for
+ * nfs4svc_encode_compoundres.
  *
- * Store the base and length of the rq_req.head[0] page
- * of the NFSv4.1 data, just past the rpc header.
  */
-void
-nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
+static __be32
+nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
+                         struct nfsd4_compoundres *resp)
 {
-       struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
-       struct svc_rqst *rqstp = resp->rqstp;
-       struct nfsd4_compoundargs *args = rqstp->rq_argp;
-       struct nfsd4_op *op = &args->ops[resp->opcnt];
-       struct kvec *resv = &rqstp->rq_res.head[0];
+       struct nfsd4_op *op;
+       struct nfsd4_slot *slot = resp->cstate.slot;
 
-       dprintk("--> %s entry %p\n", __func__, entry);
-
-       /* Don't cache a failed OP_SEQUENCE. */
-       if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
-               return;
+       dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
+               resp->opcnt, resp->cstate.slot->sl_cachethis);
 
-       nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
-       entry->ce_opcnt = resp->opcnt;
-       entry->ce_status = resp->cstate.status;
+       /* Encode the replayed sequence operation */
+       op = &args->ops[resp->opcnt - 1];
+       nfsd4_encode_operation(resp, op);
 
-       /*
-        * Don't need a page to cache just the sequence operation - the slot
-        * does this for us!
-        */
-
-       if (nfsd4_not_cached(resp)) {
-               entry->ce_resused = 0;
-               entry->ce_rpchdrlen = 0;
-               dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
-                       resp->cstate.slot->sl_cache_entry.ce_cachethis);
-               return;
-       }
-       entry->ce_resused = rqstp->rq_resused;
-       if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
-               entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
-       nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
-                        entry->ce_resused);
-       entry->ce_datav.iov_base = resp->cstate.statp;
-       entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
-                               (char *)page_address(rqstp->rq_respages[0]));
-       /* Current request rpc header length*/
-       entry->ce_rpchdrlen = (char *)resp->cstate.statp -
-                               (char *)page_address(rqstp->rq_respages[0]);
-}
-
-/*
- * We keep the rpc header, but take the nfs reply from the replycache.
- */
-static int
-nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
-                       struct nfsd4_cache_entry *entry)
-{
-       struct svc_rqst *rqstp = resp->rqstp;
-       struct kvec *resv = &resp->rqstp->rq_res.head[0];
-       int len;
-
-       /* Current request rpc header length*/
-       len = (char *)resp->cstate.statp -
-                       (char *)page_address(rqstp->rq_respages[0]);
-       if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
-               dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
-                       entry->ce_datav.iov_len);
-               return 0;
+       /* Return nfserr_retry_uncached_rep in next operation. */
+       if (args->opcnt > 1 && slot->sl_cachethis == 0) {
+               op = &args->ops[resp->opcnt++];
+               op->status = nfserr_retry_uncached_rep;
+               nfsd4_encode_operation(resp, op);
        }
-       /* copy the cached reply nfsd data past the current rpc header */
-       memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
-               entry->ce_datav.iov_len);
-       resv->iov_len = len + entry->ce_datav.iov_len;
-       return 1;
+       return op->status;
 }
 
 /*
- * Keep the first page of the replay. Copy the NFSv4.1 data from the first
- * cached page.  Replace any futher replay pages from the cache.
+ * The sequence operation is not cached because we can use the slot and
+ * session values.
  */
 __be32
 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
                         struct nfsd4_sequence *seq)
 {
-       struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
+       struct nfsd4_slot *slot = resp->cstate.slot;
        __be32 status;
 
-       dprintk("--> %s entry %p\n", __func__, entry);
+       dprintk("--> %s slot %p\n", __func__, slot);
 
-       /*
-        * If this is just the sequence operation, we did not keep
-        * a page in the cache entry because we can just use the
-        * slot info stored in struct nfsd4_sequence that was checked
-        * against the slot in nfsd4_sequence().
-        *
-        * This occurs when seq->cachethis is FALSE, or when the client
-        * session inactivity timer fires and a solo sequence operation
-        * is sent (lease renewal).
-        */
-       if (seq && nfsd4_not_cached(resp)) {
-               seq->maxslots = resp->cstate.session->se_fnumslots;
-               return nfs_ok;
-       }
-
-       if (!nfsd41_copy_replay_data(resp, entry)) {
-               /*
-                * Not enough room to use the replay rpc header, send the
-                * cached header. Release all the allocated result pages.
-                */
-               svc_free_res_pages(resp->rqstp);
-               nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
-                       entry->ce_resused);
-       } else {
-               /* Release all but the first allocated result page */
-
-               resp->rqstp->rq_resused--;
-               svc_free_res_pages(resp->rqstp);
+       /* Either returns 0 or nfserr_retry_uncached */
+       status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
+       if (status == nfserr_retry_uncached_rep)
+               return status;
 
-               nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
-                                &entry->ce_respages[1],
-                                entry->ce_resused - 1);
-       }
+       /* The sequence operation has been encoded, cstate->datap set. */
+       memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
 
-       resp->rqstp->rq_resused = entry->ce_resused;
-       resp->opcnt = entry->ce_opcnt;
-       resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
-       status = entry->ce_status;
+       resp->opcnt = slot->sl_opcnt;
+       resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
+       status = slot->sl_status;
 
        return status;
 }
@@ -1185,13 +1098,15 @@ nfsd4_exchange_id(struct svc_rqst *rqstp,
        int status;
        unsigned int            strhashval;
        char                    dname[HEXDIR_LEN];
+       char                    addr_str[INET6_ADDRSTRLEN];
        nfs4_verifier           verf = exid->verifier;
-       u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
+       struct sockaddr         *sa = svc_addr(rqstp);
 
+       rpc_ntop(sa, addr_str, sizeof(addr_str));
        dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
-               " ip_addr=%u flags %x, spa_how %d\n",
+               "ip_addr=%s flags %x, spa_how %d\n",
                __func__, rqstp, exid, exid->clname.len, exid->clname.data,
-               ip_addr, exid->flags, exid->spa_how);
+               addr_str, exid->flags, exid->spa_how);
 
        if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
                return nfserr_inval;
@@ -1239,12 +1154,6 @@ nfsd4_exchange_id(struct svc_rqst *rqstp,
                        expire_client(conf);
                        goto out_new;
                }
-               if (ip_addr != conf->cl_addr &&
-                   !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
-                       /* Client collision. 18.35.4 case 3 */
-                       status = nfserr_clid_inuse;
-                       goto out;
-               }
                /*
                 * Set bit when the owner id and verifier map to an already
                 * confirmed client id (18.35.3).
@@ -1258,12 +1167,12 @@ nfsd4_exchange_id(struct svc_rqst *rqstp,
                copy_verf(conf, &verf);
                new = conf;
                goto out_copy;
-       } else {
-               /* 18.35.4 case 7 */
-               if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
-                       status = nfserr_noent;
-                       goto out;
-               }
+       }
+
+       /* 18.35.4 case 7 */
+       if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
+               status = nfserr_noent;
+               goto out;
        }
 
        unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
@@ -1278,28 +1187,23 @@ nfsd4_exchange_id(struct svc_rqst *rqstp,
 
 out_new:
        /* Normal case */
-       new = create_client(exid->clname, dname);
+       new = create_client(exid->clname, dname, rqstp, &verf);
        if (new == NULL) {
-               status = nfserr_resource;
+               status = nfserr_serverfault;
                goto out;
        }
 
-       copy_verf(new, &verf);
-       copy_cred(&new->cl_cred, &rqstp->rq_cred);
-       new->cl_addr = ip_addr;
        gen_clid(new);
-       gen_confirm(new);
        add_to_unconfirmed(new, strhashval);
 out_copy:
        exid->clientid.cl_boot = new->cl_clientid.cl_boot;
        exid->clientid.cl_id = new->cl_clientid.cl_id;
 
-       new->cl_slot.sl_seqid = 0;
        exid->seqid = 1;
        nfsd4_set_ex_flags(new, exid);
 
        dprintk("nfsd4_exchange_id seqid %d flags %x\n",
-               new->cl_slot.sl_seqid, new->cl_exchange_flags);
+               new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
        status = nfs_ok;
 
 out:
@@ -1310,40 +1214,60 @@ error:
 }
 
 static int
-check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
+check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
 {
-       dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
-               slot->sl_seqid);
+       dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
+               slot_seqid);
 
        /* The slot is in use, and no response has been sent. */
-       if (slot->sl_inuse) {
-               if (seqid == slot->sl_seqid)
+       if (slot_inuse) {
+               if (seqid == slot_seqid)
                        return nfserr_jukebox;
                else
                        return nfserr_seq_misordered;
        }
        /* Normal */
-       if (likely(seqid == slot->sl_seqid + 1))
+       if (likely(seqid == slot_seqid + 1))
                return nfs_ok;
        /* Replay */
-       if (seqid == slot->sl_seqid)
+       if (seqid == slot_seqid)
                return nfserr_replay_cache;
        /* Wraparound */
-       if (seqid == 1 && (slot->sl_seqid + 1) == 0)
+       if (seqid == 1 && (slot_seqid + 1) == 0)
                return nfs_ok;
        /* Misordered replay or misordered new request */
        return nfserr_seq_misordered;
 }
 
+/*
+ * Cache the create session result into the create session single DRC
+ * slot cache by saving the xdr structure. sl_seqid has been set.
+ * Do this for solo or embedded create session operations.
+ */
+static void
+nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
+                          struct nfsd4_clid_slot *slot, int nfserr)
+{
+       slot->sl_status = nfserr;
+       memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
+}
+
+static __be32
+nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
+                           struct nfsd4_clid_slot *slot)
+{
+       memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
+       return slot->sl_status;
+}
+
 __be32
 nfsd4_create_session(struct svc_rqst *rqstp,
                     struct nfsd4_compound_state *cstate,
                     struct nfsd4_create_session *cr_ses)
 {
-       u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
-       struct nfsd4_compoundres *resp = rqstp->rq_resp;
+       struct sockaddr *sa = svc_addr(rqstp);
        struct nfs4_client *conf, *unconf;
-       struct nfsd4_slot *slot = NULL;
+       struct nfsd4_clid_slot *cs_slot = NULL;
        int status = 0;
 
        nfs4_lock_state();
@@ -1351,66 +1275,76 @@ nfsd4_create_session(struct svc_rqst *rqstp,
        conf = find_confirmed_client(&cr_ses->clientid);
 
        if (conf) {
-               slot = &conf->cl_slot;
-               status = check_slot_seqid(cr_ses->seqid, slot);
+               cs_slot = &conf->cl_cs_slot;
+               status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
                if (status == nfserr_replay_cache) {
                        dprintk("Got a create_session replay! seqid= %d\n",
-                               slot->sl_seqid);
-                       cstate->slot = slot;
-                       cstate->status = status;
+                               cs_slot->sl_seqid);
                        /* Return the cached reply status */
-                       status = nfsd4_replay_cache_entry(resp, NULL);
+                       status = nfsd4_replay_create_session(cr_ses, cs_slot);
                        goto out;
-               } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
+               } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
                        status = nfserr_seq_misordered;
                        dprintk("Sequence misordered!\n");
                        dprintk("Expected seqid= %d but got seqid= %d\n",
-                               slot->sl_seqid, cr_ses->seqid);
+                               cs_slot->sl_seqid, cr_ses->seqid);
                        goto out;
                }
-               conf->cl_slot.sl_seqid++;
+               cs_slot->sl_seqid++;
        } else if (unconf) {
                if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
-                   (ip_addr != unconf->cl_addr)) {
+                   !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
                        status = nfserr_clid_inuse;
                        goto out;
                }
 
-               slot = &unconf->cl_slot;
-               status = check_slot_seqid(cr_ses->seqid, slot);
+               cs_slot = &unconf->cl_cs_slot;
+               status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
                if (status) {
                        /* an unconfirmed replay returns misordered */
                        status = nfserr_seq_misordered;
-                       goto out;
+                       goto out_cache;
                }
 
-               slot->sl_seqid++; /* from 0 to 1 */
+               cs_slot->sl_seqid++; /* from 0 to 1 */
                move_to_confirmed(unconf);
 
-               /*
-                * We do not support RDMA or persistent sessions
-                */
-               cr_ses->flags &= ~SESSION4_PERSIST;
-               cr_ses->flags &= ~SESSION4_RDMA;
-
+               if (cr_ses->flags & SESSION4_BACK_CHAN) {
+                       unconf->cl_cb_xprt = rqstp->rq_xprt;
+                       svc_xprt_get(unconf->cl_cb_xprt);
+                       rpc_copy_addr(
+                               (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
+                               sa);
+                       unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
+                       unconf->cl_cb_conn.cb_minorversion =
+                               cstate->minorversion;
+                       unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
+                       unconf->cl_cb_seq_nr = 1;
+                       nfsd4_probe_callback(unconf);
+               }
                conf = unconf;
        } else {
                status = nfserr_stale_clientid;
                goto out;
        }
 
+       /*
+        * We do not support RDMA or persistent sessions
+        */
+       cr_ses->flags &= ~SESSION4_PERSIST;
+       cr_ses->flags &= ~SESSION4_RDMA;
+
        status = alloc_init_session(rqstp, conf, cr_ses);
        if (status)
                goto out;
 
        memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
               NFS4_MAX_SESSIONID_LEN);
-       cr_ses->seqid = slot->sl_seqid;
+       cr_ses->seqid = cs_slot->sl_seqid;
 
-       slot->sl_inuse = true;
-       cstate->slot = slot;
-       /* Ensure a page is used for the cache */
-       slot->sl_cache_entry.ce_cachethis = 1;
+out_cache:
+       /* cache solo and embedded create sessions under the state lock */
+       nfsd4_cache_create_session(cr_ses, cs_slot, status);
 out:
        nfs4_unlock_state();
        dprintk("%s returns %d\n", __func__, ntohl(status));
@@ -1444,7 +1378,7 @@ nfsd4_destroy_session(struct svc_rqst *r,
        spin_unlock(&sessionid_lock);
 
        /* wait for callbacks */
-       shutdown_callback_client(ses->se_client);
+       nfsd4_set_callback_client(ses->se_client, NULL);
        nfsd4_put_session(ses);
        status = nfs_ok;
 out:
@@ -1472,21 +1406,26 @@ nfsd4_sequence(struct svc_rqst *rqstp,
                goto out;
 
        status = nfserr_badslot;
-       if (seq->slotid >= session->se_fnumslots)
+       if (seq->slotid >= session->se_fchannel.maxreqs)
                goto out;
 
-       slot = &session->se_slots[seq->slotid];
+       slot = session->se_slots[seq->slotid];
        dprintk("%s: slotid %d\n", __func__, seq->slotid);
 
-       status = check_slot_seqid(seq->seqid, slot);
+       /* We do not negotiate the number of slots yet, so set the
+        * maxslots to the session maxreqs which is used to encode
+        * sr_highest_slotid and the sr_target_slot id to maxslots */
+       seq->maxslots = session->se_fchannel.maxreqs;
+
+       status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
        if (status == nfserr_replay_cache) {
                cstate->slot = slot;
                cstate->session = session;
                /* Return the cached reply status and set cstate->status
-                * for nfsd4_svc_encode_compoundres processing */
+                * for nfsd4_proc_compound processing */
                status = nfsd4_replay_cache_entry(resp, seq);
                cstate->status = nfserr_replay_cache;
-               goto replay_cache;
+               goto out;
        }
        if (status)
                goto out;
@@ -1494,23 +1433,23 @@ nfsd4_sequence(struct svc_rqst *rqstp,
        /* Success! bump slot seqid */
        slot->sl_inuse = true;
        slot->sl_seqid = seq->seqid;
-       slot->sl_cache_entry.ce_cachethis = seq->cachethis;
-       /* Always set the cache entry cachethis for solo sequence */
-       if (nfsd4_is_solo_sequence(resp))
-               slot->sl_cache_entry.ce_cachethis = 1;
+       slot->sl_cachethis = seq->cachethis;
 
        cstate->slot = slot;
        cstate->session = session;
 
-replay_cache:
-       /* Renew the clientid on success and on replay.
-        * Hold a session reference until done processing the compound:
+       /* Hold a session reference until done processing the compound:
         * nfsd4_put_session called only if the cstate slot is set.
         */
-       renew_client(session->se_client);
        nfsd4_get_session(session);
 out:
        spin_unlock(&sessionid_lock);
+       /* Renew the clientid on success and on replay */
+       if (cstate->session) {
+               nfs4_lock_state();
+               renew_client(session->se_client);
+               nfs4_unlock_state();
+       }
        dprintk("%s: return %d\n", __func__, ntohl(status));
        return status;
 }
@@ -1519,7 +1458,7 @@ __be32
 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                  struct nfsd4_setclientid *setclid)
 {
-       struct sockaddr_in      *sin = svc_addr_in(rqstp);
+       struct sockaddr         *sa = svc_addr(rqstp);
        struct xdr_netobj       clname = { 
                .len = setclid->se_namelen,
                .data = setclid->se_name,
@@ -1528,7 +1467,6 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        unsigned int            strhashval;
        struct nfs4_client      *conf, *unconf, *new;
        __be32                  status;
-       char                    *princ;
        char                    dname[HEXDIR_LEN];
        
        if (!check_name(clname))
@@ -1551,8 +1489,11 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                /* RFC 3530 14.2.33 CASE 0: */
                status = nfserr_clid_inuse;
                if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
-                       dprintk("NFSD: setclientid: string in use by client"
-                               " at %pI4\n", &conf->cl_addr);
+                       char addr_str[INET6_ADDRSTRLEN];
+                       rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
+                                sizeof(addr_str));
+                       dprintk("NFSD: setclientid: string in use by client "
+                               "at %s\n", addr_str);
                        goto out;
                }
        }
@@ -1570,7 +1511,7 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                 */
                if (unconf)
                        expire_client(unconf);
-               new = create_client(clname, dname);
+               new = create_client(clname, dname, rqstp, &clverifier);
                if (new == NULL)
                        goto out;
                gen_clid(new);
@@ -1587,7 +1528,7 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                         */
                        expire_client(unconf);
                }
-               new = create_client(clname, dname);
+               new = create_client(clname, dname, rqstp, &clverifier);
                if (new == NULL)
                        goto out;
                copy_clid(new, conf);
@@ -1597,7 +1538,7 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                 * probable client reboot; state will be removed if
                 * confirmed.
                 */
-               new = create_client(clname, dname);
+               new = create_client(clname, dname, rqstp, &clverifier);
                if (new == NULL)
                        goto out;
                gen_clid(new);
@@ -1608,25 +1549,12 @@ nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
                 * confirmed.
                 */
                expire_client(unconf);
-               new = create_client(clname, dname);
+               new = create_client(clname, dname, rqstp, &clverifier);
                if (new == NULL)
                        goto out;
                gen_clid(new);
        }
-       copy_verf(new, &clverifier);
-       new->cl_addr = sin->sin_addr.s_addr;
-       new->cl_flavor = rqstp->rq_flavor;
-       princ = svc_gss_principal(rqstp);
-       if (princ) {
-               new->cl_principal = kstrdup(princ, GFP_KERNEL);
-               if (new->cl_principal == NULL) {
-                       free_client(new);
-                       goto out;
-               }
-       }
-       copy_cred(&new->cl_cred, &rqstp->rq_cred);
-       gen_confirm(new);
-       gen_callback(new, setclid);
+       gen_callback(new, setclid, rpc_get_scope_id(sa));
        add_to_unconfirmed(new, strhashval);
        setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
        setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
@@ -1648,7 +1576,7 @@ nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
                         struct nfsd4_compound_state *cstate,
                         struct nfsd4_setclientid_confirm *setclientid_confirm)
 {
-       struct sockaddr_in *sin = svc_addr_in(rqstp);
+       struct sockaddr *sa = svc_addr(rqstp);
        struct nfs4_client *conf, *unconf;
        nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
        clientid_t * clid = &setclientid_confirm->sc_clientid;
@@ -1667,9 +1595,9 @@ nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
        unconf = find_unconfirmed_client(clid);
 
        status = nfserr_clid_inuse;
-       if (conf && conf->cl_addr != sin->sin_addr.s_addr)
+       if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
                goto out;
-       if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
+       if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
                goto out;
 
        /*
@@ -1687,9 +1615,7 @@ nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
                else {
                        /* XXX: We just turn off callbacks until we can handle
                          * change request correctly. */
-                       atomic_set(&conf->cl_callback.cb_set, 0);
-                       gen_confirm(conf);
-                       nfsd4_remove_clid_dir(unconf);
+                       atomic_set(&conf->cl_cb_conn.cb_set, 0);
                        expire_client(unconf);
                        status = nfs_ok;
 
@@ -1883,7 +1809,7 @@ init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *
        stp->st_stateowner = sop;
        get_nfs4_file(fp);
        stp->st_file = fp;
-       stp->st_stateid.si_boot = boot_time;
+       stp->st_stateid.si_boot = get_seconds();
        stp->st_stateid.si_stateownerid = sop->so_id;
        stp->st_stateid.si_fileid = fp->fi_id;
        stp->st_stateid.si_generation = 0;
@@ -2055,24 +1981,13 @@ nfs4_file_downgrade(struct file *filp, unsigned int share_access)
 {
        if (share_access & NFS4_SHARE_ACCESS_WRITE) {
                drop_file_write_access(filp);
+               spin_lock(&filp->f_lock);
                filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
+               spin_unlock(&filp->f_lock);
        }
 }
 
 /*
- * Recall a delegation
- */
-static int
-do_recall(void *__dp)
-{
-       struct nfs4_delegation *dp = __dp;
-
-       dp->dl_file->fi_had_conflict = true;
-       nfsd4_cb_recall(dp);
-       return 0;
-}
-
-/*
  * Spawn a thread to perform a recall on the delegation represented
  * by the lease (file_lock)
  *
@@ -2083,8 +1998,7 @@ do_recall(void *__dp)
 static
 void nfsd_break_deleg_cb(struct file_lock *fl)
 {
-       struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
-       struct task_struct *t;
+       struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
 
        dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
        if (!dp)
@@ -2112,16 +2026,8 @@ void nfsd_break_deleg_cb(struct file_lock *fl)
         */
        fl->fl_break_time = 0;
 
-       t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
-       if (IS_ERR(t)) {
-               struct nfs4_client *clp = dp->dl_client;
-
-               printk(KERN_INFO "NFSD: Callback thread failed for "
-                       "for client (clientid %08x/%08x)\n",
-                       clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
-               put_nfs4_client(dp->dl_client);
-               nfs4_put_delegation(dp);
-       }
+       dp->dl_file->fi_had_conflict = true;
+       nfsd4_cb_recall(dp);
 }
 
 /*
@@ -2184,7 +2090,7 @@ int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
                return -EAGAIN;
 }
 
-static struct lock_manager_operations nfsd_lease_mng_ops = {
+static const struct lock_manager_operations nfsd_lease_mng_ops = {
        .fl_break = nfsd_break_deleg_cb,
        .fl_release_private = nfsd_release_deleg_cb,
        .fl_copy_lock = nfsd_copy_lock_deleg_cb,
@@ -2423,7 +2329,7 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
 {
        struct nfs4_delegation *dp;
        struct nfs4_stateowner *sop = stp->st_stateowner;
-       struct nfs4_callback *cb = &sop->so_client->cl_callback;
+       struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
        struct file_lock fl, *flp = &fl;
        int status, flag = 0;
 
@@ -2479,11 +2385,8 @@ nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_sta
 
        memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
 
-       dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
-                    dp->dl_stateid.si_boot,
-                    dp->dl_stateid.si_stateownerid,
-                    dp->dl_stateid.si_fileid,
-                    dp->dl_stateid.si_generation);
+       dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
+               STATEID_VAL(&dp->dl_stateid));
 out:
        if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
                        && flag == NFS4_OPEN_DELEGATE_NONE
@@ -2562,8 +2465,10 @@ nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nf
        }
        memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
 
-       if (nfsd4_has_session(&resp->cstate))
+       if (nfsd4_has_session(&resp->cstate)) {
                open->op_stateowner->so_confirmed = 1;
+               nfsd4_create_clid_dir(open->op_stateowner->so_client);
+       }
 
        /*
        * Attempt to hand out a delegation. No error return, because the
@@ -2573,9 +2478,8 @@ nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nf
 
        status = nfs_ok;
 
-       dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
-                   stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
-                   stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
+       dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
+               STATEID_VAL(&stp->st_stateid));
 out:
        if (fp)
                put_nfs4_file(fp);
@@ -2615,7 +2519,7 @@ nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        renew_client(clp);
        status = nfserr_cb_path_down;
        if (!list_empty(&clp->cl_delegations)
-                       && !atomic_read(&clp->cl_callback.cb_set))
+                       && !atomic_read(&clp->cl_cb_conn.cb_set))
                goto out;
        status = nfs_ok;
 out:
@@ -2632,6 +2536,12 @@ nfsd4_end_grace(void)
        dprintk("NFSD: end of grace period\n");
        nfsd4_recdir_purge_old();
        locks_end_grace(&nfsd4_manager);
+       /*
+        * Now that every NFSv4 client has had the chance to recover and
+        * to see the (possibly new, possibly shorter) lease time, we
+        * can safely set the next grace time to the current lease time:
+        */
+       nfsd4_grace = nfsd4_lease;
 }
 
 static time_t
@@ -2641,9 +2551,9 @@ nfs4_laundromat(void)
        struct nfs4_stateowner *sop;
        struct nfs4_delegation *dp;
        struct list_head *pos, *next, reaplist;
-       time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
-       time_t t, clientid_val = NFSD_LEASE_TIME;
-       time_t u, test_val = NFSD_LEASE_TIME;
+       time_t cutoff = get_seconds() - nfsd4_lease;
+       time_t t, clientid_val = nfsd4_lease;
+       time_t u, test_val = nfsd4_lease;
 
        nfs4_lock_state();
 
@@ -2683,7 +2593,7 @@ nfs4_laundromat(void)
                list_del_init(&dp->dl_recall_lru);
                unhash_delegation(dp);
        }
-       test_val = NFSD_LEASE_TIME;
+       test_val = nfsd4_lease;
        list_for_each_safe(pos, next, &close_lru) {
                sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
                if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
@@ -2739,12 +2649,39 @@ nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
 static int
 STALE_STATEID(stateid_t *stateid)
 {
-       if (stateid->si_boot == boot_time)
-               return 0;
-       dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
-               stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
-               stateid->si_generation);
-       return 1;
+       if (time_after((unsigned long)boot_time,
+                       (unsigned long)stateid->si_boot)) {
+               dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
+                       STATEID_VAL(stateid));
+               return 1;
+       }
+       return 0;
+}
+
+static int
+EXPIRED_STATEID(stateid_t *stateid)
+{
+       if (time_before((unsigned long)boot_time,
+                       ((unsigned long)stateid->si_boot)) &&
+           time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
+               dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
+                       STATEID_VAL(stateid));
+               return 1;
+       }
+       return 0;
+}
+
+static __be32
+stateid_error_map(stateid_t *stateid)
+{
+       if (STALE_STATEID(stateid))
+               return nfserr_stale_stateid;
+       if (EXPIRED_STATEID(stateid))
+               return nfserr_expired;
+
+       dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
+               STATEID_VAL(stateid));
+       return nfserr_bad_stateid;
 }
 
 static inline int
@@ -2868,8 +2805,10 @@ nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
        status = nfserr_bad_stateid;
        if (is_delegation_stateid(stateid)) {
                dp = find_delegation_stateid(ino, stateid);
-               if (!dp)
+               if (!dp) {
+                       status = stateid_error_map(stateid);
                        goto out;
+               }
                status = check_stateid_generation(stateid, &dp->dl_stateid,
                                                  flags);
                if (status)
@@ -2882,8 +2821,10 @@ nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
                        *filpp = dp->dl_vfs_file;
        } else { /* open or lock stateid */
                stp = find_stateid(stateid, flags);
-               if (!stp)
+               if (!stp) {
+                       status = stateid_error_map(stateid);
                        goto out;
+               }
                if (nfs4_check_fh(current_fh, stp))
                        goto out;
                if (!stp->st_stateowner->so_confirmed)
@@ -2925,10 +2866,8 @@ nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
        struct svc_fh *current_fh = &cstate->current_fh;
        __be32 status;
 
-       dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
-                       "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
-               stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
-               stateid->si_generation);
+       dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
+               seqid, STATEID_VAL(stateid));
 
        *stpp = NULL;
        *sopp = NULL;
@@ -2957,7 +2896,7 @@ nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
                 */
                sop = search_close_lru(stateid->si_stateownerid, flags);
                if (sop == NULL)
-                       return nfserr_bad_stateid;
+                       return stateid_error_map(stateid);
                *sopp = sop;
                goto check_replay;
        }
@@ -3060,12 +2999,8 @@ nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        sop->so_confirmed = 1;
        update_stateid(&stp->st_stateid);
        memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
-       dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
-               "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
-                        stp->st_stateid.si_boot,
-                        stp->st_stateid.si_stateownerid,
-                        stp->st_stateid.si_fileid,
-                        stp->st_stateid.si_generation);
+       dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
+               __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
 
        nfsd4_create_clid_dir(sop->so_client);
 out:
@@ -3228,8 +3163,10 @@ nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
        if (!is_delegation_stateid(stateid))
                goto out;
        dp = find_delegation_stateid(inode, stateid);
-       if (!dp)
+       if (!dp) {
+               status = stateid_error_map(stateid);
                goto out;
+       }
        status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
        if (status)
                goto out;
@@ -3322,9 +3259,8 @@ find_delegation_stateid(struct inode *ino, stateid_t *stid)
        struct nfs4_file *fp;
        struct nfs4_delegation *dl;
 
-       dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
-                    stid->si_boot, stid->si_stateownerid,
-                    stid->si_fileid, stid->si_generation);
+       dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
+               STATEID_VAL(stid));
 
        fp = find_file(ino);
        if (!fp)
@@ -3353,7 +3289,7 @@ nfs4_transform_lock_offset(struct file_lock *lock)
 
 /* Hack!: For now, we're defining this just so we can use a pointer to it
  * as a unique cookie to identify our (NFSv4's) posix locks. */
-static struct lock_manager_operations nfsd_posix_mng_ops  = {
+static const struct lock_manager_operations nfsd_posix_mng_ops  = {
 };
 
 static inline void
@@ -3456,7 +3392,7 @@ alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struc
        stp->st_stateowner = sop;
        get_nfs4_file(fp);
        stp->st_file = fp;
-       stp->st_stateid.si_boot = boot_time;
+       stp->st_stateid.si_boot = get_seconds();
        stp->st_stateid.si_stateownerid = sop->so_id;
        stp->st_stateid.si_fileid = fp->fi_id;
        stp->st_stateid.si_generation = 0;
@@ -3988,6 +3924,7 @@ nfs4_state_init(void)
                INIT_LIST_HEAD(&conf_str_hashtbl[i]);
                INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
                INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
+               INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
        }
        for (i = 0; i < SESSION_HASH_SIZE; i++)
                INIT_LIST_HEAD(&sessionid_hashtbl[i]);
@@ -4010,8 +3947,6 @@ nfs4_state_init(void)
        INIT_LIST_HEAD(&close_lru);
        INIT_LIST_HEAD(&client_lru);
        INIT_LIST_HEAD(&del_recall_lru);
-       for (i = 0; i < CLIENT_HASH_SIZE; i++)
-               INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
        reclaim_str_hashtbl_size = 0;
        return 0;
 }
@@ -4029,12 +3964,6 @@ nfsd4_load_reboot_recovery_data(void)
                printk("NFSD: Failure reading reboot recovery data\n");
 }
 
-unsigned long
-get_nfs4_grace_period(void)
-{
-       return max(user_lease_time, lease_time) * HZ;
-}
-
 /*
  * Since the lifetime of a delegation isn't limited to that of an open, a
  * client may quite reasonably hang on to a delegation as long as it has
@@ -4058,37 +3987,45 @@ set_max_delegations(void)
 
 /* initialization to perform when the nfsd service is started: */
 
-static void
+static int
 __nfs4_state_start(void)
 {
-       unsigned long grace_time;
+       int ret;
 
        boot_time = get_seconds();
-       grace_time = get_nfs4_grace_period();
-       lease_time = user_lease_time;
        locks_start_grace(&nfsd4_manager);
        printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
-              grace_time/HZ);
+              nfsd4_grace);
+       ret = set_callback_cred();
+       if (ret)
+               return -ENOMEM;
        laundry_wq = create_singlethread_workqueue("nfsd4");
-       queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
+       if (laundry_wq == NULL)
+               return -ENOMEM;
+       ret = nfsd4_create_callback_queue();
+       if (ret)
+               goto out_free_laundry;
+       queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
        set_max_delegations();
+       return 0;
+out_free_laundry:
+       destroy_workqueue(laundry_wq);
+       return ret;
 }
 
-void
+int
 nfs4_state_start(void)
 {
+       int ret;
+
        if (nfs4_init)
-               return;
+               return 0;
        nfsd4_load_reboot_recovery_data();
-       __nfs4_state_start();
+       ret = __nfs4_state_start();
+       if (ret)
+               return ret;
        nfs4_init = 1;
-       return;
-}
-
-time_t
-nfs4_lease_time(void)
-{
-       return lease_time;
+       return 0;
 }
 
 static void
@@ -4135,6 +4072,7 @@ nfs4_state_shutdown(void)
        nfs4_lock_state();
        nfs4_release_reclaim();
        __nfs4_state_shutdown();
+       nfsd4_destroy_callback_queue();
        nfs4_unlock_state();
 }
 
@@ -4174,21 +4112,3 @@ nfs4_recoverydir(void)
 {
        return user_recovery_dirname;
 }
-
-/*
- * Called when leasetime is changed.
- *
- * The only way the protocol gives us to handle on-the-fly lease changes is to
- * simulate a reboot.  Instead of doing that, we just wait till the next time
- * we start to register any changes in lease time.  If the administrator
- * really wants to change the lease time *now*, they can go ahead and bring
- * nfsd down and then back up again after changing the lease time.
- *
- * user_lease_time is protected by nfsd_mutex since it's only really accessed
- * when nfsd is starting
- */
-void
-nfs4_reset_lease(time_t leasetime)
-{
-       user_lease_time = leasetime;
-}