NFSv4: Ensure that we check lock exclusive/shared type against open modes
[pandora-kernel.git] / fs / nfs / nfs4proc.c
index be2bbac..51f6a40 100644 (file)
@@ -39,6 +39,8 @@
 #include <linux/delay.h>
 #include <linux/errno.h>
 #include <linux/string.h>
+#include <linux/ratelimit.h>
+#include <linux/printk.h>
 #include <linux/slab.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/gss_api.h>
@@ -255,15 +257,28 @@ static int nfs4_handle_exception(struct nfs_server *server, int errorcode, struc
 {
        struct nfs_client *clp = server->nfs_client;
        struct nfs4_state *state = exception->state;
+       struct inode *inode = exception->inode;
        int ret = errorcode;
 
        exception->retry = 0;
        switch(errorcode) {
                case 0:
                        return 0;
+               case -NFS4ERR_OPENMODE:
+                       if (nfs_have_delegation(inode, FMODE_READ)) {
+                               nfs_inode_return_delegation(inode);
+                               exception->retry = 1;
+                               return 0;
+                       }
+                       if (state == NULL)
+                               break;
+                       nfs4_schedule_stateid_recovery(server, state);
+                       goto wait_on_recovery;
+               case -NFS4ERR_DELEG_REVOKED:
                case -NFS4ERR_ADMIN_REVOKED:
                case -NFS4ERR_BAD_STATEID:
-               case -NFS4ERR_OPENMODE:
+                       if (state != NULL)
+                               nfs_remove_bad_delegation(state->inode);
                        if (state == NULL)
                                break;
                        nfs4_schedule_stateid_recovery(server, state);
@@ -894,6 +909,8 @@ out:
 
 static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode)
 {
+       if (delegation == NULL)
+               return 0;
        if ((delegation->type & fmode) != fmode)
                return 0;
        if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags))
@@ -1036,8 +1053,7 @@ static struct nfs4_state *nfs4_try_open_cached(struct nfs4_opendata *opendata)
                }
                rcu_read_lock();
                delegation = rcu_dereference(nfsi->delegation);
-               if (delegation == NULL ||
-                   !can_open_delegated(delegation, fmode)) {
+               if (!can_open_delegated(delegation, fmode)) {
                        rcu_read_unlock();
                        break;
                }
@@ -1091,7 +1107,12 @@ static struct nfs4_state *nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data
                if (delegation)
                        delegation_flags = delegation->flags;
                rcu_read_unlock();
-               if ((delegation_flags & 1UL<<NFS_DELEGATION_NEED_RECLAIM) == 0)
+               if (data->o_arg.claim == NFS4_OPEN_CLAIM_DELEGATE_CUR) {
+                       pr_err_ratelimited("NFS: Broken NFSv4 server %s is "
+                                       "returning a delegation for "
+                                       "OPEN(CLAIM_DELEGATE_CUR)\n",
+                                       NFS_CLIENT(inode)->cl_server);
+               } else if ((delegation_flags & 1UL<<NFS_DELEGATION_NEED_RECLAIM) == 0)
                        nfs_inode_set_delegation(state->inode,
                                        data->owner->so_cred,
                                        &data->o_res);
@@ -1308,8 +1329,11 @@ int nfs4_open_delegation_recall(struct nfs_open_context *ctx, struct nfs4_state
                                 * The show must go on: exit, but mark the
                                 * stateid as needing recovery.
                                 */
+                       case -NFS4ERR_DELEG_REVOKED:
                        case -NFS4ERR_ADMIN_REVOKED:
                        case -NFS4ERR_BAD_STATEID:
+                               nfs_inode_find_state_and_recover(state->inode,
+                                               stateid);
                                nfs4_schedule_stateid_recovery(server, state);
                        case -EKEYEXPIRED:
                                /*
@@ -1423,11 +1447,9 @@ static void nfs4_open_prepare(struct rpc_task *task, void *calldata)
                        goto out_no_action;
                rcu_read_lock();
                delegation = rcu_dereference(NFS_I(data->state->inode)->delegation);
-               if (delegation != NULL &&
-                   test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) == 0) {
-                       rcu_read_unlock();
-                       goto out_no_action;
-               }
+               if (data->o_arg.claim != NFS4_OPEN_CLAIM_DELEGATE_CUR &&
+                   can_open_delegated(delegation, data->o_arg.fmode))
+                       goto unlock_no_action;
                rcu_read_unlock();
        }
        /* Update sequence id. */
@@ -1444,6 +1466,8 @@ static void nfs4_open_prepare(struct rpc_task *task, void *calldata)
                return;
        rpc_call_start(task);
        return;
+unlock_no_action:
+       rcu_read_unlock();
 out_no_action:
        task->tk_action = NULL;
 
@@ -1814,7 +1838,7 @@ static struct nfs4_state *nfs4_do_open(struct inode *dir, struct dentry *dentry,
                 * the user though...
                 */
                if (status == -NFS4ERR_BAD_SEQID) {
-                       printk(KERN_WARNING "NFS: v4 server %s "
+                       pr_warn_ratelimited("NFS: v4 server %s "
                                        " returned a bad sequence-id error!\n",
                                        NFS_SERVER(dir)->nfs_client->cl_hostname);
                        exception.retry = 1;
@@ -1885,7 +1909,10 @@ static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
                           struct nfs4_state *state)
 {
        struct nfs_server *server = NFS_SERVER(inode);
-       struct nfs4_exception exception = { };
+       struct nfs4_exception exception = {
+               .state = state,
+               .inode = inode,
+       };
        int err;
        do {
                err = nfs4_handle_exception(server,
@@ -2215,11 +2242,12 @@ static int nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
                switch (err) {
                case 0:
                case -NFS4ERR_WRONGSEC:
-                       break;
+                       goto out;
                default:
                        err = nfs4_handle_exception(server, err, &exception);
                }
        } while (exception.retry);
+out:
        return err;
 }
 
@@ -3422,19 +3450,6 @@ static inline int nfs4_server_supports_acls(struct nfs_server *server)
  */
 #define NFS4ACL_MAXPAGES (XATTR_SIZE_MAX >> PAGE_CACHE_SHIFT)
 
-static void buf_to_pages(const void *buf, size_t buflen,
-               struct page **pages, unsigned int *pgbase)
-{
-       const void *p = buf;
-
-       *pgbase = offset_in_page(buf);
-       p -= *pgbase;
-       while (p < buf + buflen) {
-               *(pages++) = virt_to_page(p);
-               p += PAGE_CACHE_SIZE;
-       }
-}
-
 static int buf_to_pages_noslab(const void *buf, size_t buflen,
                struct page **pages, unsigned int *pgbase)
 {
@@ -3531,9 +3546,19 @@ out:
        nfs4_set_cached_acl(inode, acl);
 }
 
+/*
+ * The getxattr API returns the required buffer length when called with a
+ * NULL buf. The NFSv4 acl tool then calls getxattr again after allocating
+ * the required buf.  On a NULL buf, we send a page of data to the server
+ * guessing that the ACL request can be serviced by a page. If so, we cache
+ * up to the page of ACL data, and the 2nd call to getxattr is serviced by
+ * the cache. If not so, we throw away the page, and cache the required
+ * length. The next getxattr call will then produce another round trip to
+ * the server, this time with the input buf of the required size.
+ */
 static ssize_t __nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t buflen)
 {
-       struct page *pages[NFS4ACL_MAXPAGES];
+       struct page *pages[NFS4ACL_MAXPAGES] = {NULL, };
        struct nfs_getaclargs args = {
                .fh = NFS_FH(inode),
                .acl_pages = pages,
@@ -3548,41 +3573,60 @@ static ssize_t __nfs4_get_acl_uncached(struct inode *inode, void *buf, size_t bu
                .rpc_argp = &args,
                .rpc_resp = &res,
        };
-       struct page *localpage = NULL;
-       int ret;
+       int ret = -ENOMEM, npages, i, acl_len = 0;
 
-       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);
+       npages = (buflen + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       /* As long as we're doing a round trip to the server anyway,
+        * let's be prepared for a page of acl data. */
+       if (npages == 0)
+               npages = 1;
+
+       for (i = 0; i < npages; i++) {
+               pages[i] = alloc_page(GFP_KERNEL);
+               if (!pages[i])
+                       goto out_free;
        }
-       ret = nfs4_call_sync(NFS_SERVER(inode)->client, NFS_SERVER(inode), &msg, &args.seq_args, &res.seq_res, 0);
+       if (npages > 1) {
+               /* for decoding across pages */
+               res.acl_scratch = alloc_page(GFP_KERNEL);
+               if (!res.acl_scratch)
+                       goto out_free;
+       }
+       args.acl_len = npages * PAGE_SIZE;
+       args.acl_pgbase = 0;
+       /* Let decode_getfacl know not to fail if the ACL data is larger than
+        * the page we send as a guess */
+       if (buf == NULL)
+               res.acl_flags |= NFS4_ACL_LEN_REQUEST;
+       resp_buf = page_address(pages[0]);
+
+       dprintk("%s  buf %p buflen %ld npages %d args.acl_len %ld\n",
+               __func__, buf, buflen, npages, args.acl_len);
+       ret = nfs4_call_sync(NFS_SERVER(inode)->client, NFS_SERVER(inode),
+                            &msg, &args.seq_args, &res.seq_res, 0);
        if (ret)
                goto out_free;
-       if (res.acl_len > args.acl_len)
-               nfs4_write_cached_acl(inode, NULL, res.acl_len);
+
+       acl_len = res.acl_len - res.acl_data_offset;
+       if (acl_len > args.acl_len)
+               nfs4_write_cached_acl(inode, NULL, acl_len);
        else
-               nfs4_write_cached_acl(inode, resp_buf, res.acl_len);
+               nfs4_write_cached_acl(inode, resp_buf + res.acl_data_offset,
+                                     acl_len);
        if (buf) {
                ret = -ERANGE;
-               if (res.acl_len > buflen)
+               if (acl_len > buflen)
                        goto out_free;
-               if (localpage)
-                       memcpy(buf, resp_buf, res.acl_len);
+               _copy_from_pages(buf, pages, res.acl_data_offset,
+                               acl_len);
        }
-       ret = res.acl_len;
+       ret = acl_len;
 out_free:
-       if (localpage)
-               __free_page(localpage);
+       for (i = 0; i < npages; i++)
+               if (pages[i])
+                       __free_page(pages[i]);
+       if (res.acl_scratch)
+               __free_page(res.acl_scratch);
        return ret;
 }
 
@@ -3613,6 +3657,8 @@ static ssize_t nfs4_proc_get_acl(struct inode *inode, void *buf, size_t buflen)
                nfs_zap_acl_cache(inode);
        ret = nfs4_read_cached_acl(inode, buf, buflen);
        if (ret != -ENOENT)
+               /* -ENOENT is returned if there is no ACL or if there is an ACL
+                * but no cached acl data, just the acl length */
                return ret;
        return nfs4_get_acl_uncached(inode, buf, buflen);
 }
@@ -3681,8 +3727,11 @@ nfs4_async_handle_error(struct rpc_task *task, const struct nfs_server *server,
        if (task->tk_status >= 0)
                return 0;
        switch(task->tk_status) {
+               case -NFS4ERR_DELEG_REVOKED:
                case -NFS4ERR_ADMIN_REVOKED:
                case -NFS4ERR_BAD_STATEID:
+                       if (state != NULL)
+                               nfs_remove_bad_delegation(state->inode);
                case -NFS4ERR_OPENMODE:
                        if (state == NULL)
                                break;
@@ -4404,7 +4453,9 @@ static int _nfs4_do_setlk(struct nfs4_state *state, int cmd, struct file_lock *f
 static int nfs4_lock_reclaim(struct nfs4_state *state, struct file_lock *request)
 {
        struct nfs_server *server = NFS_SERVER(state->inode);
-       struct nfs4_exception exception = { };
+       struct nfs4_exception exception = {
+               .inode = state->inode,
+       };
        int err;
 
        do {
@@ -4422,7 +4473,9 @@ static int nfs4_lock_reclaim(struct nfs4_state *state, struct file_lock *request
 static int nfs4_lock_expired(struct nfs4_state *state, struct file_lock *request)
 {
        struct nfs_server *server = NFS_SERVER(state->inode);
-       struct nfs4_exception exception = { };
+       struct nfs4_exception exception = {
+               .inode = state->inode,
+       };
        int err;
 
        err = nfs4_set_lock_state(state, request);
@@ -4500,7 +4553,10 @@ out:
 
 static int nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock *request)
 {
-       struct nfs4_exception exception = { };
+       struct nfs4_exception exception = {
+               .state = state,
+               .inode = state->inode,
+       };
        int err;
 
        do {
@@ -4545,6 +4601,20 @@ nfs4_proc_lock(struct file *filp, int cmd, struct file_lock *request)
 
        if (state == NULL)
                return -ENOLCK;
+       /*
+        * Don't rely on the VFS having checked the file open mode,
+        * since it won't do this for flock() locks.
+        */
+       switch (request->fl_type & (F_RDLCK|F_WRLCK|F_UNLCK)) {
+       case F_RDLCK:
+               if (!(filp->f_mode & FMODE_READ))
+                       return -EBADF;
+               break;
+       case F_WRLCK:
+               if (!(filp->f_mode & FMODE_WRITE))
+                       return -EBADF;
+       }
+
        do {
                status = nfs4_proc_setlk(state, cmd, request);
                if ((status != -EAGAIN) || IS_SETLK(cmd))
@@ -4593,6 +4663,7 @@ int nfs4_lock_delegation_recall(struct nfs4_state *state, struct file_lock *fl)
                                 * The show must go on: exit, but mark the
                                 * stateid as needing recovery.
                                 */
+                       case -NFS4ERR_DELEG_REVOKED:
                        case -NFS4ERR_ADMIN_REVOKED:
                        case -NFS4ERR_BAD_STATEID:
                        case -NFS4ERR_OPENMODE:
@@ -4850,8 +4921,10 @@ int nfs4_proc_exchange_id(struct nfs_client *clp, struct rpc_cred *cred)
                                clp->cl_rpcclient->cl_auth->au_flavor);
 
        res.server_scope = kzalloc(sizeof(struct server_scope), GFP_KERNEL);
-       if (unlikely(!res.server_scope))
-               return -ENOMEM;
+       if (unlikely(!res.server_scope)) {
+               status = -ENOMEM;
+               goto out;
+       }
 
        status = rpc_call_sync(clp->cl_rpcclient, &msg, RPC_TASK_TIMEOUT);
        if (!status)
@@ -4868,12 +4941,13 @@ int nfs4_proc_exchange_id(struct nfs_client *clp, struct rpc_cred *cred)
                        clp->server_scope = NULL;
                }
 
-               if (!clp->server_scope)
+               if (!clp->server_scope) {
                        clp->server_scope = res.server_scope;
-               else
-                       kfree(res.server_scope);
+                       goto out;
+               }
        }
-
+       kfree(res.server_scope);
+out:
        dprintk("<-- %s status= %d\n", __func__, status);
        return status;
 }
@@ -5928,21 +6002,22 @@ nfs4_layoutcommit_done(struct rpc_task *task, void *calldata)
                return;
 
        switch (task->tk_status) { /* Just ignore these failures */
-       case NFS4ERR_DELEG_REVOKED: /* layout was recalled */
-       case NFS4ERR_BADIOMODE:     /* no IOMODE_RW layout for range */
-       case NFS4ERR_BADLAYOUT:     /* no layout */
-       case NFS4ERR_GRACE:         /* loca_recalim always false */
+       case -NFS4ERR_DELEG_REVOKED: /* layout was recalled */
+       case -NFS4ERR_BADIOMODE:     /* no IOMODE_RW layout for range */
+       case -NFS4ERR_BADLAYOUT:     /* no layout */
+       case -NFS4ERR_GRACE:        /* loca_recalim always false */
                task->tk_status = 0;
-       }
-
-       if (nfs4_async_handle_error(task, server, NULL) == -EAGAIN) {
-               rpc_restart_call_prepare(task);
-               return;
-       }
-
-       if (task->tk_status == 0)
+               break;
+       case 0:
                nfs_post_op_update_inode_force_wcc(data->args.inode,
                                                   data->res.fattr);
+               break;
+       default:
+               if (nfs4_async_handle_error(task, server, NULL) == -EAGAIN) {
+                       rpc_restart_call_prepare(task);
+                       return;
+               }
+       }
 }
 
 static void nfs4_layoutcommit_release(void *calldata)
@@ -6045,11 +6120,12 @@ nfs41_proc_secinfo_no_name(struct nfs_server *server, struct nfs_fh *fhandle,
                case 0:
                case -NFS4ERR_WRONGSEC:
                case -NFS4ERR_NOTSUPP:
-                       break;
+                       goto out;
                default:
                        err = nfs4_handle_exception(server, err, &exception);
                }
        } while (exception.retry);
+out:
        return err;
 }