nfsd: move most of nfsfh.h to fs/nfsd
[safe/jmp/linux-2.6] / fs / nfsd / nfs4proc.c
index 795ad6c..e2b5666 100644 (file)
  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *
- * Note: some routines in this file are just trivial wrappers
- * (e.g. nfsd4_lookup()) defined solely for the sake of consistent
- * naming.  Since all such routines have been declared "inline",
- * there shouldn't be any associated overhead.  At some point in
- * the future, I might inline these "by hand" to clean up a
- * little.
  */
-
-#include <linux/param.h>
-#include <linux/major.h>
-#include <linux/slab.h>
 #include <linux/file.h>
 
-#include <linux/sunrpc/svc.h>
-#include <linux/nfsd/nfsd.h>
-#include <linux/nfsd/cache.h>
-#include <linux/nfs4.h>
-#include <linux/nfsd/state.h>
-#include <linux/nfsd/xdr4.h>
-#include <linux/nfs4_acl.h>
+#include "cache.h"
+#include "xdr4.h"
+#include "vfs.h"
 
 #define NFSDDBG_FACILITY               NFSDDBG_PROC
 
+static u32 nfsd_attrmask[] = {
+       NFSD_WRITEABLE_ATTRS_WORD0,
+       NFSD_WRITEABLE_ATTRS_WORD1,
+       NFSD_WRITEABLE_ATTRS_WORD2
+};
+
+static u32 nfsd41_ex_attrmask[] = {
+       NFSD_SUPPATTR_EXCLCREAT_WORD0,
+       NFSD_SUPPATTR_EXCLCREAT_WORD1,
+       NFSD_SUPPATTR_EXCLCREAT_WORD2
+};
+
+static __be32
+check_attr_support(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+                  u32 *bmval, u32 *writable)
+{
+       struct dentry *dentry = cstate->current_fh.fh_dentry;
+
+       /*
+        * Check about attributes are supported by the NFSv4 server or not.
+        * According to spec, unsupported attributes return ERR_ATTRNOTSUPP.
+        */
+       if ((bmval[0] & ~nfsd_suppattrs0(cstate->minorversion)) ||
+           (bmval[1] & ~nfsd_suppattrs1(cstate->minorversion)) ||
+           (bmval[2] & ~nfsd_suppattrs2(cstate->minorversion)))
+               return nfserr_attrnotsupp;
+
+       /*
+        * Check FATTR4_WORD0_ACL can be supported
+        * in current environment or not.
+        */
+       if (bmval[0] & FATTR4_WORD0_ACL) {
+               if (!IS_POSIXACL(dentry->d_inode))
+                       return nfserr_attrnotsupp;
+       }
+
+       /*
+        * According to spec, read-only attributes return ERR_INVAL.
+        */
+       if (writable) {
+               if ((bmval[0] & ~writable[0]) || (bmval[1] & ~writable[1]) ||
+                   (bmval[2] & ~writable[2]))
+                       return nfserr_inval;
+       }
+
+       return nfs_ok;
+}
+
+static __be32
+nfsd4_check_open_attributes(struct svc_rqst *rqstp,
+       struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
+{
+       __be32 status = nfs_ok;
+
+       if (open->op_create == NFS4_OPEN_CREATE) {
+               if (open->op_createmode == NFS4_CREATE_UNCHECKED
+                   || open->op_createmode == NFS4_CREATE_GUARDED)
+                       status = check_attr_support(rqstp, cstate,
+                                       open->op_bmval, nfsd_attrmask);
+               else if (open->op_createmode == NFS4_CREATE_EXCLUSIVE4_1)
+                       status = check_attr_support(rqstp, cstate,
+                                       open->op_bmval, nfsd41_ex_attrmask);
+       }
+
+       return status;
+}
+
+static int
+is_create_with_attrs(struct nfsd4_open *open)
+{
+       return open->op_create == NFS4_OPEN_CREATE
+               && (open->op_createmode == NFS4_CREATE_UNCHECKED
+                   || open->op_createmode == NFS4_CREATE_GUARDED
+                   || open->op_createmode == NFS4_CREATE_EXCLUSIVE4_1);
+}
+
+/*
+ * if error occurs when setting the acl, just clear the acl bit
+ * in the returned attr bitmap.
+ */
+static void
+do_set_nfs4_acl(struct svc_rqst *rqstp, struct svc_fh *fhp,
+               struct nfs4_acl *acl, u32 *bmval)
+{
+       __be32 status;
+
+       status = nfsd4_set_nfs4_acl(rqstp, fhp, acl);
+       if (status)
+               /*
+                * We should probably fail the whole open at this point,
+                * but we've already created the file, so it's too late;
+                * So this seems the least of evils:
+                */
+               bmval[0] &= ~FATTR4_WORD0_ACL;
+}
+
 static inline void
 fh_dup2(struct svc_fh *dst, struct svc_fh *src)
 {
@@ -67,37 +148,53 @@ fh_dup2(struct svc_fh *dst, struct svc_fh *src)
        *dst = *src;
 }
 
-static int
+static __be32
 do_open_permission(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open, int accmode)
 {
-       int status;
+       __be32 status;
 
        if (open->op_truncate &&
                !(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
                return nfserr_inval;
 
        if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
-               accmode |= MAY_READ;
+               accmode |= NFSD_MAY_READ;
        if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
-               accmode |= (MAY_WRITE | MAY_TRUNC);
-       if (open->op_share_deny & NFS4_SHARE_DENY_WRITE)
-               accmode |= MAY_WRITE;
+               accmode |= (NFSD_MAY_WRITE | NFSD_MAY_TRUNC);
+       if (open->op_share_deny & NFS4_SHARE_DENY_READ)
+               accmode |= NFSD_MAY_WRITE;
 
        status = fh_verify(rqstp, current_fh, S_IFREG, accmode);
 
        return status;
 }
 
-static int
+static __be32
 do_open_lookup(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
 {
        struct svc_fh resfh;
-       int status;
+       __be32 status;
+       int created = 0;
 
        fh_init(&resfh, NFS4_FHSIZE);
        open->op_truncate = 0;
 
        if (open->op_create) {
+               /* FIXME: check session persistence and pnfs flags.
+                * The nfsv4.1 spec requires the following semantics:
+                *
+                * Persistent   | pNFS   | Server REQUIRED | Client Allowed
+                * Reply Cache  | server |                 |
+                * -------------+--------+-----------------+--------------------
+                * no           | no     | EXCLUSIVE4_1    | EXCLUSIVE4_1
+                *              |        |                 | (SHOULD)
+                *              |        | and EXCLUSIVE4  | or EXCLUSIVE4
+                *              |        |                 | (SHOULD NOT)
+                * no           | yes    | EXCLUSIVE4_1    | EXCLUSIVE4_1
+                * yes          | no     | GUARDED4        | GUARDED4
+                * yes          | yes    | GUARDED4        | GUARDED4
+                */
+
                /*
                 * Note: create modes (UNCHECKED,GUARDED...) are the same
                 * in NFSv4 as in v3.
@@ -105,36 +202,47 @@ do_open_lookup(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_o
                status = nfsd_create_v3(rqstp, current_fh, open->op_fname.data,
                                        open->op_fname.len, &open->op_iattr,
                                        &resfh, open->op_createmode,
-                                       (u32 *)open->op_verf.data, &open->op_truncate);
-       }
-       else {
+                                       (u32 *)open->op_verf.data,
+                                       &open->op_truncate, &created);
+
+               /*
+                * Following rfc 3530 14.2.16, use the returned bitmask
+                * to indicate which attributes we used to store the
+                * verifier:
+                */
+               if (open->op_createmode == NFS4_CREATE_EXCLUSIVE && status == 0)
+                       open->op_bmval[1] = (FATTR4_WORD1_TIME_ACCESS |
+                                               FATTR4_WORD1_TIME_MODIFY);
+       } else {
                status = nfsd_lookup(rqstp, current_fh,
                                     open->op_fname.data, open->op_fname.len, &resfh);
                fh_unlock(current_fh);
        }
+       if (status)
+               goto out;
 
-       if (!status) {
-               set_change_info(&open->op_cinfo, current_fh);
+       if (is_create_with_attrs(open) && open->op_acl != NULL)
+               do_set_nfs4_acl(rqstp, &resfh, open->op_acl, open->op_bmval);
 
-               /* set reply cache */
-               fh_dup2(current_fh, &resfh);
-               open->op_stateowner->so_replay.rp_openfh_len =
-                       resfh.fh_handle.fh_size;
-               memcpy(open->op_stateowner->so_replay.rp_openfh,
-                               &resfh.fh_handle.fh_base,
-                               resfh.fh_handle.fh_size);
+       set_change_info(&open->op_cinfo, current_fh);
+       fh_dup2(current_fh, &resfh);
 
-               status = do_open_permission(rqstp, current_fh, open, MAY_NOP);
-       }
+       /* set reply cache */
+       fh_copy_shallow(&open->op_stateowner->so_replay.rp_openfh,
+                       &resfh.fh_handle);
+       if (!created)
+               status = do_open_permission(rqstp, current_fh, open,
+                                           NFSD_MAY_NOP);
 
+out:
        fh_put(&resfh);
        return status;
 }
 
-static int
+static __be32
 do_open_fhandle(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
 {
-       int status;
+       __be32 status;
 
        /* Only reclaims from previously confirmed clients are valid */
        if ((status = nfs4_check_open_reclaim(&open->op_clientid)))
@@ -147,24 +255,35 @@ do_open_fhandle(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_
        memset(&open->op_cinfo, 0, sizeof(struct nfsd4_change_info));
 
        /* set replay cache */
-       open->op_stateowner->so_replay.rp_openfh_len = current_fh->fh_handle.fh_size;
-       memcpy(open->op_stateowner->so_replay.rp_openfh,
-               &current_fh->fh_handle.fh_base,
-               current_fh->fh_handle.fh_size);
+       fh_copy_shallow(&open->op_stateowner->so_replay.rp_openfh,
+                       &current_fh->fh_handle);
 
        open->op_truncate = (open->op_iattr.ia_valid & ATTR_SIZE) &&
                (open->op_iattr.ia_size == 0);
 
-       status = do_open_permission(rqstp, current_fh, open, MAY_OWNER_OVERRIDE);
+       status = do_open_permission(rqstp, current_fh, open,
+                                   NFSD_MAY_OWNER_OVERRIDE);
 
        return status;
 }
 
+static void
+copy_clientid(clientid_t *clid, struct nfsd4_session *session)
+{
+       struct nfsd4_sessionid *sid =
+                       (struct nfsd4_sessionid *)session->se_sessionid.data;
+
+       clid->cl_boot = sid->clientid.cl_boot;
+       clid->cl_id = sid->clientid.cl_id;
+}
 
-static inline int
-nfsd4_open(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open, struct nfs4_stateowner **replay_owner)
+static __be32
+nfsd4_open(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+          struct nfsd4_open *open)
 {
-       int status;
+       __be32 status;
+       struct nfsd4_compoundres *resp;
+
        dprintk("NFSD: nfsd4_open filename %.*s op_stateowner %p\n",
                (int)open->op_fname.len, open->op_fname.data,
                open->op_stateowner);
@@ -173,41 +292,44 @@ nfsd4_open(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open
        if (open->op_create && open->op_claim_type != NFS4_OPEN_CLAIM_NULL)
                return nfserr_inval;
 
+       if (nfsd4_has_session(cstate))
+               copy_clientid(&open->op_clientid, cstate->session);
+
        nfs4_lock_state();
 
        /* check seqid for replay. set nfs4_owner */
-       status = nfsd4_process_open1(open);
-       if (status == NFSERR_REPLAY_ME) {
+       resp = rqstp->rq_resp;
+       status = nfsd4_process_open1(&resp->cstate, open);
+       if (status == nfserr_replay_me) {
                struct nfs4_replay *rp = &open->op_stateowner->so_replay;
-               fh_put(current_fh);
-               current_fh->fh_handle.fh_size = rp->rp_openfh_len;
-               memcpy(&current_fh->fh_handle.fh_base, rp->rp_openfh,
-                               rp->rp_openfh_len);
-               status = fh_verify(rqstp, current_fh, 0, MAY_NOP);
+               fh_put(&cstate->current_fh);
+               fh_copy_shallow(&cstate->current_fh.fh_handle,
+                               &rp->rp_openfh);
+               status = fh_verify(rqstp, &cstate->current_fh, 0, NFSD_MAY_NOP);
                if (status)
                        dprintk("nfsd4_open: replay failed"
                                " restoring previous filehandle\n");
                else
-                       status = NFSERR_REPLAY_ME;
+                       status = nfserr_replay_me;
        }
        if (status)
                goto out;
 
+       status = nfsd4_check_open_attributes(rqstp, cstate, open);
+       if (status)
+               goto out;
+
        /* Openowner is now set, so sequence id will get bumped.  Now we need
         * these checks before we do any creates: */
        status = nfserr_grace;
-       if (nfs4_in_grace() && open->op_claim_type != NFS4_OPEN_CLAIM_PREVIOUS)
+       if (locks_in_grace() && open->op_claim_type != NFS4_OPEN_CLAIM_PREVIOUS)
                goto out;
        status = nfserr_no_grace;
-       if (!nfs4_in_grace() && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
+       if (!locks_in_grace() && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
                goto out;
 
        switch (open->op_claim_type) {
                case NFS4_OPEN_CLAIM_DELEGATE_CUR:
-                       status = nfserr_inval;
-                       if (open->op_create)
-                               goto out;
-                       /* fall through */
                case NFS4_OPEN_CLAIM_NULL:
                        /*
                         * (1) set CURRENT_FH to the file being opened,
@@ -215,7 +337,8 @@ nfsd4_open(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open
                         * (3) set open->op_truncate if the file is to be
                         * truncated after opening, (4) do permission checking.
                         */
-                       status = do_open_lookup(rqstp, current_fh, open);
+                       status = do_open_lookup(rqstp, &cstate->current_fh,
+                                               open);
                        if (status)
                                goto out;
                        break;
@@ -227,18 +350,19 @@ nfsd4_open(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open
                         * open->op_truncate if the file is to be truncated
                         * after opening, (3) do permission checking.
                        */
-                       status = do_open_fhandle(rqstp, current_fh, open);
+                       status = do_open_fhandle(rqstp, &cstate->current_fh,
+                                                open);
                        if (status)
                                goto out;
                        break;
                case NFS4_OPEN_CLAIM_DELEGATE_PREV:
                        open->op_stateowner->so_confirmed = 1;
-                       printk("NFSD: unsupported OPEN claim type %d\n",
+                       dprintk("NFSD: unsupported OPEN claim type %d\n",
                                open->op_claim_type);
                        status = nfserr_notsupp;
                        goto out;
                default:
-                       printk("NFSD: Invalid OPEN claim type %d\n",
+                       dprintk("NFSD: Invalid OPEN claim type %d\n",
                                open->op_claim_type);
                        status = nfserr_inval;
                        goto out;
@@ -248,11 +372,11 @@ nfsd4_open(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open
         * successful, it (1) truncates the file if open->op_truncate was
         * set, (2) sets open->op_stateid, (3) sets open->op_delegation.
         */
-       status = nfsd4_process_open2(rqstp, current_fh, open);
+       status = nfsd4_process_open2(rqstp, &cstate->current_fh, open);
 out:
        if (open->op_stateowner) {
                nfs4_get_stateowner(open->op_stateowner);
-               *replay_owner = open->op_stateowner;
+               cstate->replay_owner = open->op_stateowner;
        }
        nfs4_unlock_state();
        return status;
@@ -261,99 +385,115 @@ out:
 /*
  * filehandle-manipulating ops.
  */
-static inline int
-nfsd4_getfh(struct svc_fh *current_fh, struct svc_fh **getfh)
+static __be32
+nfsd4_getfh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+           struct svc_fh **getfh)
 {
-       if (!current_fh->fh_dentry)
+       if (!cstate->current_fh.fh_dentry)
                return nfserr_nofilehandle;
 
-       *getfh = current_fh;
+       *getfh = &cstate->current_fh;
        return nfs_ok;
 }
 
-static inline int
-nfsd4_putfh(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_putfh *putfh)
+static __be32
+nfsd4_putfh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+           struct nfsd4_putfh *putfh)
 {
-       fh_put(current_fh);
-       current_fh->fh_handle.fh_size = putfh->pf_fhlen;
-       memcpy(&current_fh->fh_handle.fh_base, putfh->pf_fhval, putfh->pf_fhlen);
-       return fh_verify(rqstp, current_fh, 0, MAY_NOP);
+       fh_put(&cstate->current_fh);
+       cstate->current_fh.fh_handle.fh_size = putfh->pf_fhlen;
+       memcpy(&cstate->current_fh.fh_handle.fh_base, putfh->pf_fhval,
+              putfh->pf_fhlen);
+       return fh_verify(rqstp, &cstate->current_fh, 0, NFSD_MAY_NOP);
 }
 
-static inline int
-nfsd4_putrootfh(struct svc_rqst *rqstp, struct svc_fh *current_fh)
+static __be32
+nfsd4_putrootfh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+               void *arg)
 {
-       int status;
+       __be32 status;
 
-       fh_put(current_fh);
-       status = exp_pseudoroot(rqstp->rq_client, current_fh,
-                             &rqstp->rq_chandle);
+       fh_put(&cstate->current_fh);
+       status = exp_pseudoroot(rqstp, &cstate->current_fh);
        return status;
 }
 
-static inline int
-nfsd4_restorefh(struct svc_fh *current_fh, struct svc_fh *save_fh)
+static __be32
+nfsd4_restorefh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+               void *arg)
 {
-       if (!save_fh->fh_dentry)
+       if (!cstate->save_fh.fh_dentry)
                return nfserr_restorefh;
 
-       fh_dup2(current_fh, save_fh);
+       fh_dup2(&cstate->current_fh, &cstate->save_fh);
        return nfs_ok;
 }
 
-static inline int
-nfsd4_savefh(struct svc_fh *current_fh, struct svc_fh *save_fh)
+static __be32
+nfsd4_savefh(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            void *arg)
 {
-       if (!current_fh->fh_dentry)
+       if (!cstate->current_fh.fh_dentry)
                return nfserr_nofilehandle;
 
-       fh_dup2(save_fh, current_fh);
+       fh_dup2(&cstate->save_fh, &cstate->current_fh);
        return nfs_ok;
 }
 
 /*
  * misc nfsv4 ops
  */
-static inline int
-nfsd4_access(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_access *access)
+static __be32
+nfsd4_access(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_access *access)
 {
        if (access->ac_req_access & ~NFS3_ACCESS_FULL)
                return nfserr_inval;
 
        access->ac_resp_access = access->ac_req_access;
-       return nfsd_access(rqstp, current_fh, &access->ac_resp_access, &access->ac_supported);
+       return nfsd_access(rqstp, &cstate->current_fh, &access->ac_resp_access,
+                          &access->ac_supported);
 }
 
-static inline int
-nfsd4_commit(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_commit *commit)
+static __be32
+nfsd4_commit(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_commit *commit)
 {
-       int status;
+       __be32 status;
 
        u32 *p = (u32 *)commit->co_verf.data;
        *p++ = nfssvc_boot.tv_sec;
        *p++ = nfssvc_boot.tv_usec;
 
-       status = nfsd_commit(rqstp, current_fh, commit->co_offset, commit->co_count);
+       status = nfsd_commit(rqstp, &cstate->current_fh, commit->co_offset,
+                            commit->co_count);
        if (status == nfserr_symlink)
                status = nfserr_inval;
        return status;
 }
 
-static int
-nfsd4_create(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_create *create)
+static __be32
+nfsd4_create(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_create *create)
 {
        struct svc_fh resfh;
-       int status;
+       __be32 status;
        dev_t rdev;
 
        fh_init(&resfh, NFS4_FHSIZE);
 
-       status = fh_verify(rqstp, current_fh, S_IFDIR, MAY_CREATE);
+       status = fh_verify(rqstp, &cstate->current_fh, S_IFDIR,
+                          NFSD_MAY_CREATE);
        if (status == nfserr_symlink)
                status = nfserr_notdir;
        if (status)
                return status;
 
+       status = check_attr_support(rqstp, cstate, create->cr_bmval,
+                                   nfsd_attrmask);
+       if (status)
+               return status;
+
        switch (create->cr_type) {
        case NF4LNK:
                /* ugh! we have to null-terminate the linktext, or
@@ -365,9 +505,10 @@ nfsd4_create(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_cre
                 */
                create->cr_linkname[create->cr_linklen] = 0;
 
-               status = nfsd_symlink(rqstp, current_fh, create->cr_name,
-                                     create->cr_namelen, create->cr_linkname,
-                                     create->cr_linklen, &resfh, &create->cr_iattr);
+               status = nfsd_symlink(rqstp, &cstate->current_fh,
+                                     create->cr_name, create->cr_namelen,
+                                     create->cr_linkname, create->cr_linklen,
+                                     &resfh, &create->cr_iattr);
                break;
 
        case NF4BLK:
@@ -375,9 +516,9 @@ nfsd4_create(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_cre
                if (MAJOR(rdev) != create->cr_specdata1 ||
                    MINOR(rdev) != create->cr_specdata2)
                        return nfserr_inval;
-               status = nfsd_create(rqstp, current_fh, create->cr_name,
-                                    create->cr_namelen, &create->cr_iattr,
-                                    S_IFBLK, rdev, &resfh);
+               status = nfsd_create(rqstp, &cstate->current_fh,
+                                    create->cr_name, create->cr_namelen,
+                                    &create->cr_iattr, S_IFBLK, rdev, &resfh);
                break;
 
        case NF4CHR:
@@ -385,105 +526,119 @@ nfsd4_create(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_cre
                if (MAJOR(rdev) != create->cr_specdata1 ||
                    MINOR(rdev) != create->cr_specdata2)
                        return nfserr_inval;
-               status = nfsd_create(rqstp, current_fh, create->cr_name,
-                                    create->cr_namelen, &create->cr_iattr,
-                                    S_IFCHR, rdev, &resfh);
+               status = nfsd_create(rqstp, &cstate->current_fh,
+                                    create->cr_name, create->cr_namelen,
+                                    &create->cr_iattr,S_IFCHR, rdev, &resfh);
                break;
 
        case NF4SOCK:
-               status = nfsd_create(rqstp, current_fh, create->cr_name,
-                                    create->cr_namelen, &create->cr_iattr,
-                                    S_IFSOCK, 0, &resfh);
+               status = nfsd_create(rqstp, &cstate->current_fh,
+                                    create->cr_name, create->cr_namelen,
+                                    &create->cr_iattr, S_IFSOCK, 0, &resfh);
                break;
 
        case NF4FIFO:
-               status = nfsd_create(rqstp, current_fh, create->cr_name,
-                                    create->cr_namelen, &create->cr_iattr,
-                                    S_IFIFO, 0, &resfh);
+               status = nfsd_create(rqstp, &cstate->current_fh,
+                                    create->cr_name, create->cr_namelen,
+                                    &create->cr_iattr, S_IFIFO, 0, &resfh);
                break;
 
        case NF4DIR:
                create->cr_iattr.ia_valid &= ~ATTR_SIZE;
-               status = nfsd_create(rqstp, current_fh, create->cr_name,
-                                    create->cr_namelen, &create->cr_iattr,
-                                    S_IFDIR, 0, &resfh);
+               status = nfsd_create(rqstp, &cstate->current_fh,
+                                    create->cr_name, create->cr_namelen,
+                                    &create->cr_iattr, S_IFDIR, 0, &resfh);
                break;
 
        default:
                status = nfserr_badtype;
        }
 
-       if (!status) {
-               fh_unlock(current_fh);
-               set_change_info(&create->cr_cinfo, current_fh);
-               fh_dup2(current_fh, &resfh);
-       }
+       if (status)
+               goto out;
+
+       if (create->cr_acl != NULL)
+               do_set_nfs4_acl(rqstp, &resfh, create->cr_acl,
+                               create->cr_bmval);
 
+       fh_unlock(&cstate->current_fh);
+       set_change_info(&create->cr_cinfo, &cstate->current_fh);
+       fh_dup2(&cstate->current_fh, &resfh);
+out:
        fh_put(&resfh);
        return status;
 }
 
-static inline int
-nfsd4_getattr(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_getattr *getattr)
+static __be32
+nfsd4_getattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             struct nfsd4_getattr *getattr)
 {
-       int status;
+       __be32 status;
 
-       status = fh_verify(rqstp, current_fh, 0, MAY_NOP);
+       status = fh_verify(rqstp, &cstate->current_fh, 0, NFSD_MAY_NOP);
        if (status)
                return status;
 
        if (getattr->ga_bmval[1] & NFSD_WRITEONLY_ATTRS_WORD1)
                return nfserr_inval;
 
-       getattr->ga_bmval[0] &= NFSD_SUPPORTED_ATTRS_WORD0;
-       getattr->ga_bmval[1] &= NFSD_SUPPORTED_ATTRS_WORD1;
+       getattr->ga_bmval[0] &= nfsd_suppattrs0(cstate->minorversion);
+       getattr->ga_bmval[1] &= nfsd_suppattrs1(cstate->minorversion);
+       getattr->ga_bmval[2] &= nfsd_suppattrs2(cstate->minorversion);
 
-       getattr->ga_fhp = current_fh;
+       getattr->ga_fhp = &cstate->current_fh;
        return nfs_ok;
 }
 
-static inline int
-nfsd4_link(struct svc_rqst *rqstp, struct svc_fh *current_fh,
-          struct svc_fh *save_fh, struct nfsd4_link *link)
+static __be32
+nfsd4_link(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+          struct nfsd4_link *link)
 {
-       int status = nfserr_nofilehandle;
+       __be32 status = nfserr_nofilehandle;
 
-       if (!save_fh->fh_dentry)
+       if (!cstate->save_fh.fh_dentry)
                return status;
-       status = nfsd_link(rqstp, current_fh, link->li_name, link->li_namelen, save_fh);
+       status = nfsd_link(rqstp, &cstate->current_fh,
+                          link->li_name, link->li_namelen, &cstate->save_fh);
        if (!status)
-               set_change_info(&link->li_cinfo, current_fh);
+               set_change_info(&link->li_cinfo, &cstate->current_fh);
        return status;
 }
 
-static int
-nfsd4_lookupp(struct svc_rqst *rqstp, struct svc_fh *current_fh)
+static __be32
+nfsd4_lookupp(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             void *arg)
 {
        struct svc_fh tmp_fh;
-       int ret;
+       __be32 ret;
 
        fh_init(&tmp_fh, NFS4_FHSIZE);
-       if((ret = exp_pseudoroot(rqstp->rq_client, &tmp_fh,
-                             &rqstp->rq_chandle)) != 0)
+       ret = exp_pseudoroot(rqstp, &tmp_fh);
+       if (ret)
                return ret;
-       if (tmp_fh.fh_dentry == current_fh->fh_dentry) {
+       if (tmp_fh.fh_dentry == cstate->current_fh.fh_dentry) {
                fh_put(&tmp_fh);
                return nfserr_noent;
        }
        fh_put(&tmp_fh);
-       return nfsd_lookup(rqstp, current_fh, "..", 2, current_fh);
+       return nfsd_lookup(rqstp, &cstate->current_fh,
+                          "..", 2, &cstate->current_fh);
 }
 
-static inline int
-nfsd4_lookup(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_lookup *lookup)
+static __be32
+nfsd4_lookup(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_lookup *lookup)
 {
-       return nfsd_lookup(rqstp, current_fh, lookup->lo_name, lookup->lo_len, current_fh);
+       return nfsd_lookup(rqstp, &cstate->current_fh,
+                          lookup->lo_name, lookup->lo_len,
+                          &cstate->current_fh);
 }
 
-static inline int
-nfsd4_read(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_read *read)
+static __be32
+nfsd4_read(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+          struct nfsd4_read *read)
 {
-       int status;
+       __be32 status;
 
        /* no need to check permission - this will be done in nfsd_read() */
 
@@ -493,8 +648,8 @@ nfsd4_read(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_read
 
        nfs4_lock_state();
        /* check stateid */
-       if ((status = nfs4_preprocess_stateid_op(current_fh, &read->rd_stateid,
-                               CHECK_FH | RD_STATE, &read->rd_filp))) {
+       if ((status = nfs4_preprocess_stateid_op(cstate, &read->rd_stateid,
+                                                RD_STATE, &read->rd_filp))) {
                dprintk("NFSD: nfsd4_read: couldn't process stateid!\n");
                goto out;
        }
@@ -504,12 +659,13 @@ nfsd4_read(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_read
 out:
        nfs4_unlock_state();
        read->rd_rqstp = rqstp;
-       read->rd_fhp = current_fh;
+       read->rd_fhp = &cstate->current_fh;
        return status;
 }
 
-static inline int
-nfsd4_readdir(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_readdir *readdir)
+static __be32
+nfsd4_readdir(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             struct nfsd4_readdir *readdir)
 {
        u64 cookie = readdir->rd_cookie;
        static const nfs4_verifier zeroverf;
@@ -519,56 +675,60 @@ nfsd4_readdir(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_re
        if (readdir->rd_bmval[1] & NFSD_WRITEONLY_ATTRS_WORD1)
                return nfserr_inval;
 
-       readdir->rd_bmval[0] &= NFSD_SUPPORTED_ATTRS_WORD0;
-       readdir->rd_bmval[1] &= NFSD_SUPPORTED_ATTRS_WORD1;
+       readdir->rd_bmval[0] &= nfsd_suppattrs0(cstate->minorversion);
+       readdir->rd_bmval[1] &= nfsd_suppattrs1(cstate->minorversion);
+       readdir->rd_bmval[2] &= nfsd_suppattrs2(cstate->minorversion);
 
        if ((cookie > ~(u32)0) || (cookie == 1) || (cookie == 2) ||
            (cookie == 0 && memcmp(readdir->rd_verf.data, zeroverf.data, NFS4_VERIFIER_SIZE)))
                return nfserr_bad_cookie;
 
        readdir->rd_rqstp = rqstp;
-       readdir->rd_fhp = current_fh;
+       readdir->rd_fhp = &cstate->current_fh;
        return nfs_ok;
 }
 
-static inline int
-nfsd4_readlink(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_readlink *readlink)
+static __be32
+nfsd4_readlink(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+              struct nfsd4_readlink *readlink)
 {
        readlink->rl_rqstp = rqstp;
-       readlink->rl_fhp = current_fh;
+       readlink->rl_fhp = &cstate->current_fh;
        return nfs_ok;
 }
 
-static inline int
-nfsd4_remove(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_remove *remove)
+static __be32
+nfsd4_remove(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_remove *remove)
 {
-       int status;
+       __be32 status;
 
-       if (nfs4_in_grace())
+       if (locks_in_grace())
                return nfserr_grace;
-       status = nfsd_unlink(rqstp, current_fh, 0, remove->rm_name, remove->rm_namelen);
+       status = nfsd_unlink(rqstp, &cstate->current_fh, 0,
+                            remove->rm_name, remove->rm_namelen);
        if (status == nfserr_symlink)
                return nfserr_notdir;
        if (!status) {
-               fh_unlock(current_fh);
-               set_change_info(&remove->rm_cinfo, current_fh);
+               fh_unlock(&cstate->current_fh);
+               set_change_info(&remove->rm_cinfo, &cstate->current_fh);
        }
        return status;
 }
 
-static inline int
-nfsd4_rename(struct svc_rqst *rqstp, struct svc_fh *current_fh,
-            struct svc_fh *save_fh, struct nfsd4_rename *rename)
+static __be32
+nfsd4_rename(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_rename *rename)
 {
-       int status = nfserr_nofilehandle;
+       __be32 status = nfserr_nofilehandle;
 
-       if (!save_fh->fh_dentry)
+       if (!cstate->save_fh.fh_dentry)
                return status;
-       if (nfs4_in_grace() && !(save_fh->fh_export->ex_flags
+       if (locks_in_grace() && !(cstate->save_fh.fh_export->ex_flags
                                        & NFSEXP_NOSUBTREECHECK))
                return nfserr_grace;
-       status = nfsd_rename(rqstp, save_fh, rename->rn_sname,
-                            rename->rn_snamelen, current_fh,
+       status = nfsd_rename(rqstp, &cstate->save_fh, rename->rn_sname,
+                            rename->rn_snamelen, &cstate->current_fh,
                             rename->rn_tname, rename->rn_tnamelen);
 
        /* the underlying filesystem returns different error's than required
@@ -576,51 +736,90 @@ nfsd4_rename(struct svc_rqst *rqstp, struct svc_fh *current_fh,
        if (status == nfserr_isdir)
                status = nfserr_exist;
        else if ((status == nfserr_notdir) &&
-                  (S_ISDIR(save_fh->fh_dentry->d_inode->i_mode) &&
-                   S_ISDIR(current_fh->fh_dentry->d_inode->i_mode)))
+                  (S_ISDIR(cstate->save_fh.fh_dentry->d_inode->i_mode) &&
+                   S_ISDIR(cstate->current_fh.fh_dentry->d_inode->i_mode)))
                status = nfserr_exist;
        else if (status == nfserr_symlink)
                status = nfserr_notdir;
 
        if (!status) {
-               set_change_info(&rename->rn_sinfo, current_fh);
-               set_change_info(&rename->rn_tinfo, save_fh);
+               set_change_info(&rename->rn_sinfo, &cstate->current_fh);
+               set_change_info(&rename->rn_tinfo, &cstate->save_fh);
        }
        return status;
 }
 
-static inline int
-nfsd4_setattr(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_setattr *setattr)
+static __be32
+nfsd4_secinfo(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             struct nfsd4_secinfo *secinfo)
+{
+       struct svc_fh resfh;
+       struct svc_export *exp;
+       struct dentry *dentry;
+       __be32 err;
+
+       fh_init(&resfh, NFS4_FHSIZE);
+       err = nfsd_lookup_dentry(rqstp, &cstate->current_fh,
+                                   secinfo->si_name, secinfo->si_namelen,
+                                   &exp, &dentry);
+       if (err)
+               return err;
+       if (dentry->d_inode == NULL) {
+               exp_put(exp);
+               err = nfserr_noent;
+       } else
+               secinfo->si_exp = exp;
+       dput(dentry);
+       return err;
+}
+
+static __be32
+nfsd4_setattr(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             struct nfsd4_setattr *setattr)
 {
-       int status = nfs_ok;
+       __be32 status = nfs_ok;
 
        if (setattr->sa_iattr.ia_valid & ATTR_SIZE) {
                nfs4_lock_state();
-               status = nfs4_preprocess_stateid_op(current_fh,
-                       &setattr->sa_stateid, CHECK_FH | WR_STATE, NULL);
+               status = nfs4_preprocess_stateid_op(cstate,
+                       &setattr->sa_stateid, WR_STATE, NULL);
                nfs4_unlock_state();
                if (status) {
                        dprintk("NFSD: nfsd4_setattr: couldn't process stateid!\n");
                        return status;
                }
        }
+       status = mnt_want_write(cstate->current_fh.fh_export->ex_path.mnt);
+       if (status)
+               return status;
        status = nfs_ok;
+
+       status = check_attr_support(rqstp, cstate, setattr->sa_bmval,
+                                   nfsd_attrmask);
+       if (status)
+               goto out;
+
        if (setattr->sa_acl != NULL)
-               status = nfsd4_set_nfs4_acl(rqstp, current_fh, setattr->sa_acl);
+               status = nfsd4_set_nfs4_acl(rqstp, &cstate->current_fh,
+                                           setattr->sa_acl);
        if (status)
-               return status;
-       status = nfsd_setattr(rqstp, current_fh, &setattr->sa_iattr,
+               goto out;
+       status = nfsd_setattr(rqstp, &cstate->current_fh, &setattr->sa_iattr,
                                0, (time_t)0);
+out:
+       mnt_drop_write(cstate->current_fh.fh_export->ex_path.mnt);
        return status;
 }
 
-static inline int
-nfsd4_write(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_write *write)
+static __be32
+nfsd4_write(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+           struct nfsd4_write *write)
 {
        stateid_t *stateid = &write->wr_stateid;
        struct file *filp = NULL;
        u32 *p;
-       int status = nfs_ok;
+       __be32 status = nfs_ok;
+       unsigned long cnt;
 
        /* no need to check permission - this will be done in nfsd_write() */
 
@@ -628,8 +827,7 @@ nfsd4_write(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_writ
                return nfserr_inval;
 
        nfs4_lock_state();
-       status = nfs4_preprocess_stateid_op(current_fh, stateid,
-                                       CHECK_FH | WR_STATE, &filp);
+       status = nfs4_preprocess_stateid_op(cstate, stateid, WR_STATE, &filp);
        if (filp)
                get_file(filp);
        nfs4_unlock_state();
@@ -639,18 +837,20 @@ nfsd4_write(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_writ
                return status;
        }
 
-       write->wr_bytes_written = write->wr_buflen;
+       cnt = write->wr_buflen;
        write->wr_how_written = write->wr_stable_how;
        p = (u32 *)write->wr_verifier.data;
        *p++ = nfssvc_boot.tv_sec;
        *p++ = nfssvc_boot.tv_usec;
 
-       status =  nfsd_write(rqstp, current_fh, filp, write->wr_offset,
-                       rqstp->rq_vec, write->wr_vlen, write->wr_buflen,
-                       &write->wr_how_written);
+       status =  nfsd_write(rqstp, &cstate->current_fh, filp,
+                            write->wr_offset, rqstp->rq_vec, write->wr_vlen,
+                            &cnt, &write->wr_how_written);
        if (filp)
                fput(filp);
 
+       write->wr_bytes_written = cnt;
+
        if (status == nfserr_symlink)
                status = nfserr_inval;
        return status;
@@ -661,20 +861,22 @@ nfsd4_write(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_writ
  * attributes matched.  VERIFY is implemented by mapping NFSERR_SAME
  * to NFS_OK after the call; NVERIFY by mapping NFSERR_NOT_SAME to NFS_OK.
  */
-static int
-nfsd4_verify(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_verify *verify)
+static __be32
+_nfsd4_verify(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_verify *verify)
 {
-       u32 *buf, *p;
+       __be32 *buf, *p;
        int count;
-       int status;
+       __be32 status;
 
-       status = fh_verify(rqstp, current_fh, 0, MAY_NOP);
+       status = fh_verify(rqstp, &cstate->current_fh, 0, NFSD_MAY_NOP);
+       if (status)
+               return status;
+
+       status = check_attr_support(rqstp, cstate, verify->ve_bmval, NULL);
        if (status)
                return status;
 
-       if ((verify->ve_bmval[0] & ~NFSD_SUPPORTED_ATTRS_WORD0)
-           || (verify->ve_bmval[1] & ~NFSD_SUPPORTED_ATTRS_WORD1))
-               return nfserr_attrnotsupp;
        if ((verify->ve_bmval[0] & FATTR4_WORD0_RDATTR_ERROR)
            || (verify->ve_bmval[1] & NFSD_WRITEONLY_ATTRS_WORD1))
                return nfserr_inval;
@@ -689,10 +891,11 @@ nfsd4_verify(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_ver
        if (!buf)
                return nfserr_resource;
 
-       status = nfsd4_encode_fattr(current_fh, current_fh->fh_export,
-                                   current_fh->fh_dentry, buf,
+       status = nfsd4_encode_fattr(&cstate->current_fh,
+                                   cstate->current_fh.fh_export,
+                                   cstate->current_fh.fh_dentry, buf,
                                    &count, verify->ve_bmval,
-                                   rqstp);
+                                   rqstp, 0);
 
        /* this means that nfsd4_encode_fattr() ran out of space */
        if (status == nfserr_resource && count == 0)
@@ -700,7 +903,8 @@ nfsd4_verify(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_ver
        if (status)
                goto out_kfree;
 
-       p = buf + 3;
+       /* skip bitmap */
+       p = buf + 1 + ntohl(buf[0]);
        status = nfserr_not_same;
        if (ntohl(*p++) != verify->ve_attrlen)
                goto out_kfree;
@@ -712,6 +916,26 @@ out_kfree:
        return status;
 }
 
+static __be32
+nfsd4_nverify(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+             struct nfsd4_verify *verify)
+{
+       __be32 status;
+
+       status = _nfsd4_verify(rqstp, cstate, verify);
+       return status == nfserr_not_same ? nfs_ok : status;
+}
+
+static __be32
+nfsd4_verify(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
+            struct nfsd4_verify *verify)
+{
+       __be32 status;
+
+       status = _nfsd4_verify(rqstp, cstate, verify);
+       return status == nfserr_same ? nfs_ok : status;
+}
+
 /*
  * NULL call.
  */
@@ -727,6 +951,40 @@ static inline void nfsd4_increment_op_stats(u32 opnum)
                nfsdstats.nfs4_opcount[opnum]++;
 }
 
+typedef __be32(*nfsd4op_func)(struct svc_rqst *, struct nfsd4_compound_state *,
+                             void *);
+enum nfsd4_op_flags {
+       ALLOWED_WITHOUT_FH = 1 << 0,    /* No current filehandle required */
+       ALLOWED_ON_ABSENT_FS = 2 << 0,  /* ops processed on absent fs */
+       ALLOWED_AS_FIRST_OP = 3 << 0,   /* ops reqired first in compound */
+};
+
+struct nfsd4_operation {
+       nfsd4op_func op_func;
+       u32 op_flags;
+       char *op_name;
+};
+
+static struct nfsd4_operation nfsd4_ops[];
+
+static const char *nfsd4_op_name(unsigned opnum);
+
+/*
+ * Enforce NFSv4.1 COMPOUND ordering rules.
+ *
+ * TODO:
+ * - enforce NFS4ERR_NOT_ONLY_OP,
+ * - DESTROY_SESSION MUST be the final operation in the COMPOUND request.
+ */
+static bool nfs41_op_ordering_ok(struct nfsd4_compoundargs *args)
+{
+       if (args->minorversion && args->opcnt > 0) {
+               struct nfsd4_op *op = &args->ops[0];
+               return (op->status == nfserr_op_illegal) ||
+                      (nfsd4_ops[op->opnum].op_flags & ALLOWED_AS_FIRST_OP);
+       }
+       return true;
+}
 
 /*
  * COMPOUND call.
@@ -737,24 +995,14 @@ nfsd4_proc_compound(struct svc_rqst *rqstp,
                    struct nfsd4_compoundres *resp)
 {
        struct nfsd4_op *op;
-       struct svc_fh   *current_fh = NULL;
-       struct svc_fh   *save_fh = NULL;
-       struct nfs4_stateowner *replay_owner = NULL;
-       int             slack_space;    /* in words, not bytes! */
-       int             status;
-
-       status = nfserr_resource;
-       current_fh = kmalloc(sizeof(*current_fh), GFP_KERNEL);
-       if (current_fh == NULL)
-               goto out;
-       fh_init(current_fh, NFS4_FHSIZE);
-       save_fh = kmalloc(sizeof(*save_fh), GFP_KERNEL);
-       if (save_fh == NULL)
-               goto out;
-       fh_init(save_fh, NFS4_FHSIZE);
+       struct nfsd4_operation *opdesc;
+       struct nfsd4_compound_state *cstate = &resp->cstate;
+       int             slack_bytes;
+       __be32          status;
 
        resp->xbuf = &rqstp->rq_res;
-       resp->p = rqstp->rq_res.head[0].iov_base + rqstp->rq_res.head[0].iov_len;
+       resp->p = rqstp->rq_res.head[0].iov_base +
+                                               rqstp->rq_res.head[0].iov_len;
        resp->tagp = resp->p;
        /* reserve space for: taglen, tag, and opcnt */
        resp->p += 2 + XDR_QUADLEN(args->taglen);
@@ -763,20 +1011,33 @@ nfsd4_proc_compound(struct svc_rqst *rqstp,
        resp->tag = args->tag;
        resp->opcnt = 0;
        resp->rqstp = rqstp;
+       resp->cstate.minorversion = args->minorversion;
+       resp->cstate.replay_owner = NULL;
+       fh_init(&resp->cstate.current_fh, NFS4_FHSIZE);
+       fh_init(&resp->cstate.save_fh, NFS4_FHSIZE);
+       /* Use the deferral mechanism only for NFSv4.0 compounds */
+       rqstp->rq_usedeferral = (args->minorversion == 0);
 
        /*
         * According to RFC3010, this takes precedence over all other errors.
         */
        status = nfserr_minor_vers_mismatch;
-       if (args->minorversion > NFSD_SUPPORTED_MINOR_VERSION)
+       if (args->minorversion > nfsd_supported_minorversion)
                goto out;
 
+       if (!nfs41_op_ordering_ok(args)) {
+               op = &args->ops[0];
+               op->status = nfserr_sequence_pos;
+               goto encode_op;
+       }
+
        status = nfs_ok;
        while (!status && resp->opcnt < args->opcnt) {
                op = &args->ops[resp->opcnt++];
 
-               dprintk("nfsv4 compound op #%d: %d\n", resp->opcnt, op->opnum);
-
+               dprintk("nfsv4 compound op #%d/%d: %d (%s)\n",
+                       resp->opcnt, args->opcnt, op->opnum,
+                       nfsd4_op_name(op->opnum));
                /*
                 * The XDR decode routines may have pre-set op->status;
                 * for example, if there is a miscellaneous XDR error
@@ -790,164 +1051,55 @@ nfsd4_proc_compound(struct svc_rqst *rqstp,
                 * failed response to the next operation.  If we don't
                 * have enough room, fail with ERR_RESOURCE.
                 */
-/* FIXME - is slack_space *really* words, or bytes??? - neilb */
-               slack_space = (char *)resp->end - (char *)resp->p;
-               if (slack_space < COMPOUND_SLACK_SPACE + COMPOUND_ERR_SLACK_SPACE) {
-                       BUG_ON(slack_space < COMPOUND_ERR_SLACK_SPACE);
+               slack_bytes = (char *)resp->end - (char *)resp->p;
+               if (slack_bytes < COMPOUND_SLACK_SPACE
+                               + COMPOUND_ERR_SLACK_SPACE) {
+                       BUG_ON(slack_bytes < COMPOUND_ERR_SLACK_SPACE);
                        op->status = nfserr_resource;
                        goto encode_op;
                }
 
-               /* All operations except RENEW, SETCLIENTID, RESTOREFH
-               * SETCLIENTID_CONFIRM, PUTFH and PUTROOTFH
-               * require a valid current filehandle
-               */
-               if (!current_fh->fh_dentry) {
-                       if (!((op->opnum == OP_PUTFH) ||
-                             (op->opnum == OP_PUTROOTFH) ||
-                             (op->opnum == OP_SETCLIENTID) ||
-                             (op->opnum == OP_SETCLIENTID_CONFIRM) ||
-                             (op->opnum == OP_RENEW) ||
-                             (op->opnum == OP_RESTOREFH) ||
-                             (op->opnum == OP_RELEASE_LOCKOWNER))) {
+               opdesc = &nfsd4_ops[op->opnum];
+
+               if (!cstate->current_fh.fh_dentry) {
+                       if (!(opdesc->op_flags & ALLOWED_WITHOUT_FH)) {
                                op->status = nfserr_nofilehandle;
                                goto encode_op;
                        }
-               }
-               /* Check must be done at start of each operation, except
-                * for GETATTR and ops not listed as returning NFS4ERR_MOVED
-                */
-               else if (current_fh->fh_export->ex_fslocs.migrated &&
-                        !((op->opnum == OP_GETATTR) ||
-                          (op->opnum == OP_PUTROOTFH) ||
-                          (op->opnum == OP_PUTPUBFH) ||
-                          (op->opnum == OP_RENEW) ||
-                          (op->opnum == OP_SETCLIENTID) ||
-                          (op->opnum == OP_RELEASE_LOCKOWNER))) {
+               } else if (cstate->current_fh.fh_export->ex_fslocs.migrated &&
+                         !(opdesc->op_flags & ALLOWED_ON_ABSENT_FS)) {
                        op->status = nfserr_moved;
                        goto encode_op;
                }
-               switch (op->opnum) {
-               case OP_ACCESS:
-                       op->status = nfsd4_access(rqstp, current_fh, &op->u.access);
-                       break;
-               case OP_CLOSE:
-                       op->status = nfsd4_close(rqstp, current_fh, &op->u.close, &replay_owner);
-                       break;
-               case OP_COMMIT:
-                       op->status = nfsd4_commit(rqstp, current_fh, &op->u.commit);
-                       break;
-               case OP_CREATE:
-                       op->status = nfsd4_create(rqstp, current_fh, &op->u.create);
-                       break;
-               case OP_DELEGRETURN:
-                       op->status = nfsd4_delegreturn(rqstp, current_fh, &op->u.delegreturn);
-                       break;
-               case OP_GETATTR:
-                       op->status = nfsd4_getattr(rqstp, current_fh, &op->u.getattr);
-                       break;
-               case OP_GETFH:
-                       op->status = nfsd4_getfh(current_fh, &op->u.getfh);
-                       break;
-               case OP_LINK:
-                       op->status = nfsd4_link(rqstp, current_fh, save_fh, &op->u.link);
-                       break;
-               case OP_LOCK:
-                       op->status = nfsd4_lock(rqstp, current_fh, &op->u.lock, &replay_owner);
-                       break;
-               case OP_LOCKT:
-                       op->status = nfsd4_lockt(rqstp, current_fh, &op->u.lockt);
-                       break;
-               case OP_LOCKU:
-                       op->status = nfsd4_locku(rqstp, current_fh, &op->u.locku, &replay_owner);
-                       break;
-               case OP_LOOKUP:
-                       op->status = nfsd4_lookup(rqstp, current_fh, &op->u.lookup);
-                       break;
-               case OP_LOOKUPP:
-                       op->status = nfsd4_lookupp(rqstp, current_fh);
-                       break;
-               case OP_NVERIFY:
-                       op->status = nfsd4_verify(rqstp, current_fh, &op->u.nverify);
-                       if (op->status == nfserr_not_same)
-                               op->status = nfs_ok;
-                       break;
-               case OP_OPEN:
-                       op->status = nfsd4_open(rqstp, current_fh, &op->u.open, &replay_owner);
-                       break;
-               case OP_OPEN_CONFIRM:
-                       op->status = nfsd4_open_confirm(rqstp, current_fh, &op->u.open_confirm, &replay_owner);
-                       break;
-               case OP_OPEN_DOWNGRADE:
-                       op->status = nfsd4_open_downgrade(rqstp, current_fh, &op->u.open_downgrade, &replay_owner);
-                       break;
-               case OP_PUTFH:
-                       op->status = nfsd4_putfh(rqstp, current_fh, &op->u.putfh);
-                       break;
-               case OP_PUTROOTFH:
-                       op->status = nfsd4_putrootfh(rqstp, current_fh);
-                       break;
-               case OP_READ:
-                       op->status = nfsd4_read(rqstp, current_fh, &op->u.read);
-                       break;
-               case OP_READDIR:
-                       op->status = nfsd4_readdir(rqstp, current_fh, &op->u.readdir);
-                       break;
-               case OP_READLINK:
-                       op->status = nfsd4_readlink(rqstp, current_fh, &op->u.readlink);
-                       break;
-               case OP_REMOVE:
-                       op->status = nfsd4_remove(rqstp, current_fh, &op->u.remove);
-                       break;
-               case OP_RENAME:
-                       op->status = nfsd4_rename(rqstp, current_fh, save_fh, &op->u.rename);
-                       break;
-               case OP_RENEW:
-                       op->status = nfsd4_renew(&op->u.renew);
-                       break;
-               case OP_RESTOREFH:
-                       op->status = nfsd4_restorefh(current_fh, save_fh);
-                       break;
-               case OP_SAVEFH:
-                       op->status = nfsd4_savefh(current_fh, save_fh);
-                       break;
-               case OP_SETATTR:
-                       op->status = nfsd4_setattr(rqstp, current_fh, &op->u.setattr);
-                       break;
-               case OP_SETCLIENTID:
-                       op->status = nfsd4_setclientid(rqstp, &op->u.setclientid);
-                       break;
-               case OP_SETCLIENTID_CONFIRM:
-                       op->status = nfsd4_setclientid_confirm(rqstp, &op->u.setclientid_confirm);
-                       break;
-               case OP_VERIFY:
-                       op->status = nfsd4_verify(rqstp, current_fh, &op->u.verify);
-                       if (op->status == nfserr_same)
-                               op->status = nfs_ok;
-                       break;
-               case OP_WRITE:
-                       op->status = nfsd4_write(rqstp, current_fh, &op->u.write);
-                       break;
-               case OP_RELEASE_LOCKOWNER:
-                       op->status = nfsd4_release_lockowner(rqstp, &op->u.release_lockowner);
-                       break;
-               default:
+
+               if (opdesc->op_func)
+                       op->status = opdesc->op_func(rqstp, cstate, &op->u);
+               else
                        BUG_ON(op->status == nfs_ok);
-                       break;
-               }
 
 encode_op:
-               if (op->status == NFSERR_REPLAY_ME) {
-                       op->replay = &replay_owner->so_replay;
+               /* Only from SEQUENCE */
+               if (resp->cstate.status == nfserr_replay_cache) {
+                       dprintk("%s NFS4.1 replay from cache\n", __func__);
+                       status = op->status;
+                       goto out;
+               }
+               if (op->status == nfserr_replay_me) {
+                       op->replay = &cstate->replay_owner->so_replay;
                        nfsd4_encode_replay(resp, op);
                        status = op->status = op->replay->rp_status;
                } else {
                        nfsd4_encode_operation(resp, op);
                        status = op->status;
                }
-               if (replay_owner && (replay_owner != (void *)(-1))) {
-                       nfs4_put_stateowner(replay_owner);
-                       replay_owner = NULL;
+
+               dprintk("nfsv4 compound op %p opcnt %d #%d: %d: status %d\n",
+                       args->ops, args->opcnt, resp->opcnt, op->opnum,
+                       be32_to_cpu(status));
+
+               if (cstate->replay_owner) {
+                       nfs4_put_stateowner(cstate->replay_owner);
+                       cstate->replay_owner = NULL;
                }
                /* XXX Ugh, we need to get rid of this kind of special case: */
                if (op->opnum == OP_READ && op->u.read.rd_filp)
@@ -955,36 +1107,207 @@ encode_op:
 
                nfsd4_increment_op_stats(op->opnum);
        }
+       if (!rqstp->rq_usedeferral && status == nfserr_dropit) {
+               dprintk("%s Dropit - send NFS4ERR_DELAY\n", __func__);
+               status = nfserr_jukebox;
+       }
 
+       resp->cstate.status = status;
+       fh_put(&resp->cstate.current_fh);
+       fh_put(&resp->cstate.save_fh);
+       BUG_ON(resp->cstate.replay_owner);
 out:
        nfsd4_release_compoundargs(args);
-       if (current_fh)
-               fh_put(current_fh);
-       kfree(current_fh);
-       if (save_fh)
-               fh_put(save_fh);
-       kfree(save_fh);
+       /* Reset deferral mechanism for RPC deferrals */
+       rqstp->rq_usedeferral = 1;
+       dprintk("nfsv4 compound returned %d\n", ntohl(status));
        return status;
 }
 
-#define nfs4svc_decode_voidargs                NULL
-#define nfs4svc_release_void           NULL
+static struct nfsd4_operation nfsd4_ops[] = {
+       [OP_ACCESS] = {
+               .op_func = (nfsd4op_func)nfsd4_access,
+               .op_name = "OP_ACCESS",
+       },
+       [OP_CLOSE] = {
+               .op_func = (nfsd4op_func)nfsd4_close,
+               .op_name = "OP_CLOSE",
+       },
+       [OP_COMMIT] = {
+               .op_func = (nfsd4op_func)nfsd4_commit,
+               .op_name = "OP_COMMIT",
+       },
+       [OP_CREATE] = {
+               .op_func = (nfsd4op_func)nfsd4_create,
+               .op_name = "OP_CREATE",
+       },
+       [OP_DELEGRETURN] = {
+               .op_func = (nfsd4op_func)nfsd4_delegreturn,
+               .op_name = "OP_DELEGRETURN",
+       },
+       [OP_GETATTR] = {
+               .op_func = (nfsd4op_func)nfsd4_getattr,
+               .op_flags = ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_GETATTR",
+       },
+       [OP_GETFH] = {
+               .op_func = (nfsd4op_func)nfsd4_getfh,
+               .op_name = "OP_GETFH",
+       },
+       [OP_LINK] = {
+               .op_func = (nfsd4op_func)nfsd4_link,
+               .op_name = "OP_LINK",
+       },
+       [OP_LOCK] = {
+               .op_func = (nfsd4op_func)nfsd4_lock,
+               .op_name = "OP_LOCK",
+       },
+       [OP_LOCKT] = {
+               .op_func = (nfsd4op_func)nfsd4_lockt,
+               .op_name = "OP_LOCKT",
+       },
+       [OP_LOCKU] = {
+               .op_func = (nfsd4op_func)nfsd4_locku,
+               .op_name = "OP_LOCKU",
+       },
+       [OP_LOOKUP] = {
+               .op_func = (nfsd4op_func)nfsd4_lookup,
+               .op_name = "OP_LOOKUP",
+       },
+       [OP_LOOKUPP] = {
+               .op_func = (nfsd4op_func)nfsd4_lookupp,
+               .op_name = "OP_LOOKUPP",
+       },
+       [OP_NVERIFY] = {
+               .op_func = (nfsd4op_func)nfsd4_nverify,
+               .op_name = "OP_NVERIFY",
+       },
+       [OP_OPEN] = {
+               .op_func = (nfsd4op_func)nfsd4_open,
+               .op_name = "OP_OPEN",
+       },
+       [OP_OPEN_CONFIRM] = {
+               .op_func = (nfsd4op_func)nfsd4_open_confirm,
+               .op_name = "OP_OPEN_CONFIRM",
+       },
+       [OP_OPEN_DOWNGRADE] = {
+               .op_func = (nfsd4op_func)nfsd4_open_downgrade,
+               .op_name = "OP_OPEN_DOWNGRADE",
+       },
+       [OP_PUTFH] = {
+               .op_func = (nfsd4op_func)nfsd4_putfh,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_PUTFH",
+       },
+       [OP_PUTPUBFH] = {
+               .op_func = (nfsd4op_func)nfsd4_putrootfh,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_PUTPUBFH",
+       },
+       [OP_PUTROOTFH] = {
+               .op_func = (nfsd4op_func)nfsd4_putrootfh,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_PUTROOTFH",
+       },
+       [OP_READ] = {
+               .op_func = (nfsd4op_func)nfsd4_read,
+               .op_name = "OP_READ",
+       },
+       [OP_READDIR] = {
+               .op_func = (nfsd4op_func)nfsd4_readdir,
+               .op_name = "OP_READDIR",
+       },
+       [OP_READLINK] = {
+               .op_func = (nfsd4op_func)nfsd4_readlink,
+               .op_name = "OP_READLINK",
+       },
+       [OP_REMOVE] = {
+               .op_func = (nfsd4op_func)nfsd4_remove,
+               .op_name = "OP_REMOVE",
+       },
+       [OP_RENAME] = {
+               .op_name = "OP_RENAME",
+               .op_func = (nfsd4op_func)nfsd4_rename,
+       },
+       [OP_RENEW] = {
+               .op_func = (nfsd4op_func)nfsd4_renew,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_RENEW",
+       },
+       [OP_RESTOREFH] = {
+               .op_func = (nfsd4op_func)nfsd4_restorefh,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_RESTOREFH",
+       },
+       [OP_SAVEFH] = {
+               .op_func = (nfsd4op_func)nfsd4_savefh,
+               .op_name = "OP_SAVEFH",
+       },
+       [OP_SECINFO] = {
+               .op_func = (nfsd4op_func)nfsd4_secinfo,
+               .op_name = "OP_SECINFO",
+       },
+       [OP_SETATTR] = {
+               .op_func = (nfsd4op_func)nfsd4_setattr,
+               .op_name = "OP_SETATTR",
+       },
+       [OP_SETCLIENTID] = {
+               .op_func = (nfsd4op_func)nfsd4_setclientid,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_SETCLIENTID",
+       },
+       [OP_SETCLIENTID_CONFIRM] = {
+               .op_func = (nfsd4op_func)nfsd4_setclientid_confirm,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_SETCLIENTID_CONFIRM",
+       },
+       [OP_VERIFY] = {
+               .op_func = (nfsd4op_func)nfsd4_verify,
+               .op_name = "OP_VERIFY",
+       },
+       [OP_WRITE] = {
+               .op_func = (nfsd4op_func)nfsd4_write,
+               .op_name = "OP_WRITE",
+       },
+       [OP_RELEASE_LOCKOWNER] = {
+               .op_func = (nfsd4op_func)nfsd4_release_lockowner,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_ON_ABSENT_FS,
+               .op_name = "OP_RELEASE_LOCKOWNER",
+       },
+
+       /* NFSv4.1 operations */
+       [OP_EXCHANGE_ID] = {
+               .op_func = (nfsd4op_func)nfsd4_exchange_id,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_AS_FIRST_OP,
+               .op_name = "OP_EXCHANGE_ID",
+       },
+       [OP_CREATE_SESSION] = {
+               .op_func = (nfsd4op_func)nfsd4_create_session,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_AS_FIRST_OP,
+               .op_name = "OP_CREATE_SESSION",
+       },
+       [OP_DESTROY_SESSION] = {
+               .op_func = (nfsd4op_func)nfsd4_destroy_session,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_AS_FIRST_OP,
+               .op_name = "OP_DESTROY_SESSION",
+       },
+       [OP_SEQUENCE] = {
+               .op_func = (nfsd4op_func)nfsd4_sequence,
+               .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_AS_FIRST_OP,
+               .op_name = "OP_SEQUENCE",
+       },
+};
+
+static const char *nfsd4_op_name(unsigned opnum)
+{
+       if (opnum < ARRAY_SIZE(nfsd4_ops))
+               return nfsd4_ops[opnum].op_name;
+       return "unknown_operation";
+}
+
 #define nfsd4_voidres                  nfsd4_voidargs
-#define nfs4svc_release_compound       NULL
 struct nfsd4_voidargs { int dummy; };
 
-#define PROC(name, argt, rest, relt, cache, respsize)  \
- { (svc_procfunc) nfsd4_proc_##name,           \
-   (kxdrproc_t) nfs4svc_decode_##argt##args,   \
-   (kxdrproc_t) nfs4svc_encode_##rest##res,    \
-   (kxdrproc_t) nfs4svc_release_##relt,                \
-   sizeof(struct nfsd4_##argt##args),          \
-   sizeof(struct nfsd4_##rest##res),           \
-   0,                                          \
-   cache,                                      \
-   respsize,                                   \
- }
-
 /*
  * TODO: At the present time, the NFSv4 server does not do XID caching
  * of requests.  Implementing XID caching would not be a serious problem,
@@ -996,8 +1319,23 @@ struct nfsd4_voidargs { int dummy; };
  * better XID's.
  */
 static struct svc_procedure            nfsd_procedures4[2] = {
-  PROC(null,    void,          void,           void,     RC_NOCACHE, 1),
-  PROC(compound, compound,     compound,       compound, RC_NOCACHE, NFSD_BUFSIZE/4)
+       [NFSPROC4_NULL] = {
+               .pc_func = (svc_procfunc) nfsd4_proc_null,
+               .pc_encode = (kxdrproc_t) nfs4svc_encode_voidres,
+               .pc_argsize = sizeof(struct nfsd4_voidargs),
+               .pc_ressize = sizeof(struct nfsd4_voidres),
+               .pc_cachetype = RC_NOCACHE,
+               .pc_xdrressize = 1,
+       },
+       [NFSPROC4_COMPOUND] = {
+               .pc_func = (svc_procfunc) nfsd4_proc_compound,
+               .pc_decode = (kxdrproc_t) nfs4svc_decode_compoundargs,
+               .pc_encode = (kxdrproc_t) nfs4svc_encode_compoundres,
+               .pc_argsize = sizeof(struct nfsd4_compoundargs),
+               .pc_ressize = sizeof(struct nfsd4_compoundres),
+               .pc_cachetype = RC_NOCACHE,
+               .pc_xdrressize = NFSD_BUFSIZE/4,
+       },
 };
 
 struct svc_version     nfsd_version4 = {