]> err.no Git - linux-2.6/blobdiff - fs/nfs/nfs4proc.c
Merge with /home/shaggy/git/linus-clean/
[linux-2.6] / fs / nfs / nfs4proc.c
index d969dd13e7db883ceed948b8e32ee3a3f0a677ef..1b76f80aedb921c737586fbca6ceeac3f0f522f2 100644 (file)
@@ -123,7 +123,7 @@ static void nfs4_setup_readdir(u64 cookie, u32 *verifier, struct dentry *dentry,
 
        BUG_ON(readdir->count < 80);
        if (cookie > 2) {
-               readdir->cookie = (cookie > 2) ? cookie : 0;
+               readdir->cookie = cookie;
                memcpy(&readdir->verifier, verifier, sizeof(readdir->verifier));
                return;
        }
@@ -269,14 +269,9 @@ static int nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *sta
        int err;
        do {
                err = _nfs4_open_reclaim(sp, state);
-               switch (err) {
-                       case 0:
-                       case -NFS4ERR_STALE_CLIENTID:
-                       case -NFS4ERR_STALE_STATEID:
-                       case -NFS4ERR_EXPIRED:
-                               return err;
-               }
-               err = nfs4_handle_exception(server, err, &exception);
+               if (err != -NFS4ERR_DELAY)
+                       break;
+               nfs4_handle_exception(server, err, &exception);
        } while (exception.retry);
        return err;
 }
@@ -508,6 +503,20 @@ out_stale:
        goto out_nodeleg;
 }
 
+static inline int nfs4_do_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state, struct dentry *dentry)
+{
+       struct nfs_server *server = NFS_SERVER(dentry->d_inode);
+       struct nfs4_exception exception = { };
+       int err;
+
+       do {
+               err = _nfs4_open_expired(sp, state, dentry);
+               if (err == -NFS4ERR_DELAY)
+                       nfs4_handle_exception(server, err, &exception);
+       } while (exception.retry);
+       return err;
+}
+
 static int nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
 {
        struct nfs_inode *nfsi = NFS_I(state->inode);
@@ -520,7 +529,7 @@ static int nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *sta
                        continue;
                get_nfs_open_context(ctx);
                spin_unlock(&state->inode->i_lock);
-               status = _nfs4_open_expired(sp, state, ctx->dentry);
+               status = nfs4_do_open_expired(sp, state, ctx->dentry);
                put_nfs_open_context(ctx);
                return status;
        }
@@ -747,11 +756,10 @@ static int _nfs4_do_setattr(struct nfs_server *server, struct nfs_fattr *fattr,
 
         fattr->valid = 0;
 
-       if (state != NULL)
+       if (state != NULL) {
                msg.rpc_cred = state->owner->so_cred;
-       if (sattr->ia_valid & ATTR_SIZE)
-               nfs4_copy_stateid(&arg.stateid, state, NULL);
-       else
+               nfs4_copy_stateid(&arg.stateid, state, current->files);
+       } else
                memcpy(&arg.stateid, &zero_stateid, sizeof(arg.stateid));
 
        return rpc_call_sync(server->client, &msg, 0);
@@ -1115,47 +1123,31 @@ static int
 nfs4_proc_setattr(struct dentry *dentry, struct nfs_fattr *fattr,
                  struct iattr *sattr)
 {
-       struct inode *          inode = dentry->d_inode;
-       int                     size_change = sattr->ia_valid & ATTR_SIZE;
-       struct nfs4_state       *state = NULL;
-       int need_iput = 0;
+       struct rpc_cred *cred;
+       struct inode *inode = dentry->d_inode;
+       struct nfs4_state *state;
        int status;
 
        fattr->valid = 0;
        
-       if (size_change) {
-               struct rpc_cred *cred = rpcauth_lookupcred(NFS_SERVER(inode)->client->cl_auth, 0);
-               if (IS_ERR(cred))
-                       return PTR_ERR(cred);
+       cred = rpcauth_lookupcred(NFS_SERVER(inode)->client->cl_auth, 0);
+       if (IS_ERR(cred))
+               return PTR_ERR(cred);
+       /* Search for an existing WRITE delegation first */
+       state = nfs4_open_delegated(inode, FMODE_WRITE, cred);
+       if (!IS_ERR(state)) {
+               /* NB: nfs4_open_delegated() bumps the inode->i_count */
+               iput(inode);
+       } else {
+               /* Search for an existing open(O_WRITE) stateid */
                state = nfs4_find_state(inode, cred, FMODE_WRITE);
-               if (state == NULL) {
-                       state = nfs4_open_delegated(dentry->d_inode,
-                                       FMODE_WRITE, cred);
-                       if (IS_ERR(state))
-                               state = nfs4_do_open(dentry->d_parent->d_inode,
-                                               dentry, FMODE_WRITE,
-                                               NULL, cred);
-                       need_iput = 1;
-               }
-               put_rpccred(cred);
-               if (IS_ERR(state))
-                       return PTR_ERR(state);
-
-               if (state->inode != inode) {
-                       printk(KERN_WARNING "nfs: raced in setattr (%p != %p), returning -EIO\n", inode, state->inode);
-                       status = -EIO;
-                       goto out;
-               }
        }
+
        status = nfs4_do_setattr(NFS_SERVER(inode), fattr,
                        NFS_FH(inode), sattr, state);
-out:
-       if (state) {
-               inode = state->inode;
+       if (state != NULL)
                nfs4_close_state(state, FMODE_WRITE);
-               if (need_iput)
-                       iput(inode);
-       }
+       put_rpccred(cred);
        return status;
 }
 
@@ -1730,6 +1722,10 @@ static int _nfs4_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
        };
        int                     status;
 
+       dprintk("%s: dentry = %s/%s, cookie = %Lu\n", __FUNCTION__,
+                       dentry->d_parent->d_name.name,
+                       dentry->d_name.name,
+                       (unsigned long long)cookie);
        lock_kernel();
        nfs4_setup_readdir(cookie, NFS_COOKIEVERF(dir), dentry, &args);
        res.pgbase = args.pgbase;
@@ -1737,6 +1733,7 @@ static int _nfs4_proc_readdir(struct dentry *dentry, struct rpc_cred *cred,
        if (status == 0)
                memcpy(NFS_COOKIEVERF(dir), res.verifier.data, NFS4_VERIFIER_SIZE);
        unlock_kernel();
+       dprintk("%s: returns %d\n", __FUNCTION__, status);
        return status;
 }
 
@@ -2188,9 +2185,75 @@ static void buf_to_pages(const void *buf, size_t buflen,
        }
 }
 
-static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
+struct nfs4_cached_acl {
+       int cached;
+       size_t len;
+       char data[0];
+};
+
+static void nfs4_set_cached_acl(struct inode *inode, struct nfs4_cached_acl *acl)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+
+       spin_lock(&inode->i_lock);
+       kfree(nfsi->nfs4_acl);
+       nfsi->nfs4_acl = acl;
+       spin_unlock(&inode->i_lock);
+}
+
+static void nfs4_zap_acl_attr(struct inode *inode)
+{
+       nfs4_set_cached_acl(inode, NULL);
+}
+
+static inline ssize_t nfs4_read_cached_acl(struct inode *inode, char *buf, size_t buflen)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+       struct nfs4_cached_acl *acl;
+       int ret = -ENOENT;
+
+       spin_lock(&inode->i_lock);
+       acl = nfsi->nfs4_acl;
+       if (acl == NULL)
+               goto out;
+       if (buf == NULL) /* user is just asking for length */
+               goto out_len;
+       if (acl->cached == 0)
+               goto out;
+       ret = -ERANGE; /* see getxattr(2) man page */
+       if (acl->len > buflen)
+               goto out;
+       memcpy(buf, acl->data, acl->len);
+out_len:
+       ret = acl->len;
+out:
+       spin_unlock(&inode->i_lock);
+       return ret;
+}
+
+static void nfs4_write_cached_acl(struct inode *inode, const char *buf, size_t acl_len)
+{
+       struct nfs4_cached_acl *acl;
+
+       if (buf && acl_len <= PAGE_SIZE) {
+               acl = kmalloc(sizeof(*acl) + acl_len, GFP_KERNEL);
+               if (acl == NULL)
+                       goto out;
+               acl->cached = 1;
+               memcpy(acl->data, buf, acl_len);
+       } else {
+               acl = kmalloc(sizeof(*acl), GFP_KERNEL);
+               if (acl == NULL)
+                       goto out;
+               acl->cached = 0;
+       }
+       acl->len = acl_len;
+out:
+       nfs4_set_cached_acl(inode, acl);
+}
+
+static inline ssize_t nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
 {
-       struct nfs_server *server = NFS_SERVER(inode);
        struct page *pages[NFS4ACL_MAXPAGES];
        struct nfs_getaclargs args = {
                .fh = NFS_FH(inode),
@@ -2198,24 +2261,66 @@ static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
                .acl_len = buflen,
        };
        size_t resp_len = buflen;
+       void *resp_buf;
        struct rpc_message msg = {
                .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETACL],
                .rpc_argp = &args,
                .rpc_resp = &resp_len,
        };
+       struct page *localpage = NULL;
        int ret;
 
-       if (!nfs4_server_supports_acls(server))
-               return -EOPNOTSUPP;
-       buf_to_pages(buf, buflen, args.acl_pages, &args.acl_pgbase);
+       if (buflen < PAGE_SIZE) {
+               /* As long as we're doing a round trip to the server anyway,
+                * let's be prepared for a page of acl data. */
+               localpage = alloc_page(GFP_KERNEL);
+               resp_buf = page_address(localpage);
+               if (localpage == NULL)
+                       return -ENOMEM;
+               args.acl_pages[0] = localpage;
+               args.acl_pgbase = 0;
+               args.acl_len = PAGE_SIZE;
+       } else {
+               resp_buf = buf;
+               buf_to_pages(buf, buflen, args.acl_pages, &args.acl_pgbase);
+       }
        ret = rpc_call_sync(NFS_CLIENT(inode), &msg, 0);
-       if (buflen && resp_len > buflen)
-               return -ERANGE;
-       if (ret == 0)
-               ret = resp_len;
+       if (ret)
+               goto out_free;
+       if (resp_len > args.acl_len)
+               nfs4_write_cached_acl(inode, NULL, resp_len);
+       else
+               nfs4_write_cached_acl(inode, resp_buf, resp_len);
+       if (buf) {
+               ret = -ERANGE;
+               if (resp_len > buflen)
+                       goto out_free;
+               if (localpage)
+                       memcpy(buf, resp_buf, resp_len);
+       }
+       ret = resp_len;
+out_free:
+       if (localpage)
+               __free_page(localpage);
        return ret;
 }
 
+static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
+{
+       struct nfs_server *server = NFS_SERVER(inode);
+       int ret;
+
+       if (!nfs4_server_supports_acls(server))
+               return -EOPNOTSUPP;
+       ret = nfs_revalidate_inode(server, inode);
+       if (ret < 0)
+               return ret;
+       ret = nfs4_read_cached_acl(inode, buf, buflen);
+       if (ret != -ENOENT)
+               return ret;
+       return nfs4_get_acl_uncached(inode, buf, buflen);
+}
+
 static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen)
 {
        struct nfs_server *server = NFS_SERVER(inode);
@@ -2236,6 +2341,8 @@ static int nfs4_proc_set_acl(struct inode *inode, const void *buf, size_t buflen
                return -EOPNOTSUPP;
        buf_to_pages(buf, buflen, arg.acl_pages, &arg.acl_pgbase);
        ret = rpc_call_sync(NFS_SERVER(inode)->client, &msg, 0);
+       if (ret == 0)
+               nfs4_write_cached_acl(inode, buf, buflen);
        return ret;
 }
 
@@ -2524,14 +2631,11 @@ static int _nfs4_proc_getlk(struct nfs4_state *state, int cmd, struct file_lock
        down_read(&clp->cl_sem);
        nlo.clientid = clp->cl_clientid;
        down(&state->lock_sema);
-       lsp = nfs4_find_lock_state(state, request->fl_owner);
-       if (lsp)
-               nlo.id = lsp->ls_id; 
-       else {
-               spin_lock(&clp->cl_lock);
-               nlo.id = nfs4_alloc_lockowner_id(clp);
-               spin_unlock(&clp->cl_lock);
-       }
+       status = nfs4_set_lock_state(state, request);
+       if (status != 0)
+               goto out;
+       lsp = request->fl_u.nfs4_fl.owner;
+       nlo.id = lsp->ls_id; 
        arg.u.lockt = &nlo;
        status = rpc_call_sync(server->client, &msg, 0);
        if (!status) {
@@ -2552,8 +2656,7 @@ static int _nfs4_proc_getlk(struct nfs4_state *state, int cmd, struct file_lock
                request->fl_pid = 0;
                status = 0;
        }
-       if (lsp)
-               nfs4_put_lock_state(lsp);
+out:
        up(&state->lock_sema);
        up_read(&clp->cl_sem);
        return status;
@@ -2613,28 +2716,26 @@ static int _nfs4_proc_unlck(struct nfs4_state *state, int cmd, struct file_lock
        };
        struct nfs4_lock_state *lsp;
        struct nfs_locku_opargs luargs;
-       int status = 0;
+       int status;
                        
        down_read(&clp->cl_sem);
        down(&state->lock_sema);
-       lsp = nfs4_find_lock_state(state, request->fl_owner);
-       if (!lsp)
+       status = nfs4_set_lock_state(state, request);
+       if (status != 0)
                goto out;
+       lsp = request->fl_u.nfs4_fl.owner;
        /* We might have lost the locks! */
-       if ((lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) {
-               luargs.seqid = lsp->ls_seqid;
-               memcpy(&luargs.stateid, &lsp->ls_stateid, sizeof(luargs.stateid));
-               arg.u.locku = &luargs;
-               status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
-               nfs4_increment_lock_seqid(status, lsp);
-       }
+       if ((lsp->ls_flags & NFS_LOCK_INITIALIZED) == 0)
+               goto out;
+       luargs.seqid = lsp->ls_seqid;
+       memcpy(&luargs.stateid, &lsp->ls_stateid, sizeof(luargs.stateid));
+       arg.u.locku = &luargs;
+       status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
+       nfs4_increment_lock_seqid(status, lsp);
 
-       if (status == 0) {
+       if (status == 0)
                memcpy(&lsp->ls_stateid,  &res.u.stateid, 
                                sizeof(lsp->ls_stateid));
-               nfs4_notify_unlck(state, request, lsp);
-       }
-       nfs4_put_lock_state(lsp);
 out:
        up(&state->lock_sema);
        if (status == 0)
@@ -2660,7 +2761,7 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *r
 {
        struct inode *inode = state->inode;
        struct nfs_server *server = NFS_SERVER(inode);
-       struct nfs4_lock_state *lsp;
+       struct nfs4_lock_state *lsp = request->fl_u.nfs4_fl.owner;
        struct nfs_lockargs arg = {
                .fh = NFS_FH(inode),
                .type = nfs4_lck_type(cmd, request),
@@ -2682,9 +2783,6 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *r
        };
        int status;
 
-       lsp = nfs4_get_lock_state(state, request->fl_owner);
-       if (lsp == NULL)
-               return -ENOMEM;
        if (!(lsp->ls_flags & NFS_LOCK_INITIALIZED)) {
                struct nfs4_state_owner *owner = state->owner;
                struct nfs_open_to_lock otl = {
@@ -2706,38 +2804,57 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *r
                * seqid mutating errors */
                nfs4_increment_seqid(status, owner);
                up(&owner->so_sema);
+               if (status == 0) {
+                       lsp->ls_flags |= NFS_LOCK_INITIALIZED;
+                       lsp->ls_seqid++;
+               }
        } else {
                struct nfs_exist_lock el = {
                        .seqid = lsp->ls_seqid,
                };
                memcpy(&el.stateid, &lsp->ls_stateid, sizeof(el.stateid));
                largs.u.exist_lock = &el;
-               largs.new_lock_owner = 0;
                arg.u.lock = &largs;
                status = rpc_call_sync(server->client, &msg, RPC_TASK_NOINTR);
+               /* increment seqid on success, and * seqid mutating errors*/
+               nfs4_increment_lock_seqid(status, lsp);
        }
-       /* increment seqid on success, and * seqid mutating errors*/
-       nfs4_increment_lock_seqid(status, lsp);
        /* save the returned stateid. */
-       if (status == 0) {
+       if (status == 0)
                memcpy(&lsp->ls_stateid, &res.u.stateid, sizeof(nfs4_stateid));
-               lsp->ls_flags |= NFS_LOCK_INITIALIZED;
-               if (!reclaim)
-                       nfs4_notify_setlk(state, request, lsp);
-       } else if (status == -NFS4ERR_DENIED)
+       else if (status == -NFS4ERR_DENIED)
                status = -EAGAIN;
-       nfs4_put_lock_state(lsp);
        return status;
 }
 
 static int nfs4_lock_reclaim(struct nfs4_state *state, struct file_lock *request)
 {
-       return _nfs4_do_setlk(state, F_SETLK, request, 1);
+       struct nfs_server *server = NFS_SERVER(state->inode);
+       struct nfs4_exception exception = { };
+       int err;
+
+       do {
+               err = _nfs4_do_setlk(state, F_SETLK, request, 1);
+               if (err != -NFS4ERR_DELAY)
+                       break;
+               nfs4_handle_exception(server, err, &exception);
+       } while (exception.retry);
+       return err;
 }
 
 static int nfs4_lock_expired(struct nfs4_state *state, struct file_lock *request)
 {
-       return _nfs4_do_setlk(state, F_SETLK, request, 0);
+       struct nfs_server *server = NFS_SERVER(state->inode);
+       struct nfs4_exception exception = { };
+       int err;
+
+       do {
+               err = _nfs4_do_setlk(state, F_SETLK, request, 0);
+               if (err != -NFS4ERR_DELAY)
+                       break;
+               nfs4_handle_exception(server, err, &exception);
+       } while (exception.retry);
+       return err;
 }
 
 static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock *request)
@@ -2747,7 +2864,9 @@ static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock
 
        down_read(&clp->cl_sem);
        down(&state->lock_sema);
-       status = _nfs4_do_setlk(state, cmd, request, 0);
+       status = nfs4_set_lock_state(state, request);
+       if (status == 0)
+               status = _nfs4_do_setlk(state, cmd, request, 0);
        up(&state->lock_sema);
        if (status == 0) {
                /* Note: we always want to sleep here! */
@@ -2805,7 +2924,6 @@ nfs4_proc_lock(struct file *filp, int cmd, struct file_lock *request)
                if (signalled())
                        break;
        } while(status < 0);
-
        return status;
 }
 
@@ -2907,6 +3025,7 @@ struct nfs_rpc_ops        nfs_v4_clientops = {
        .file_open      = nfs4_proc_file_open,
        .file_release   = nfs4_proc_file_release,
        .lock           = nfs4_proc_lock,
+       .clear_acl_cache = nfs4_zap_acl_attr,
 };
 
 /*