fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / fs / nfs / read.c
index 053d546..a9c2652 100644 (file)
@@ -15,7 +15,6 @@
  * within the RPC code when root squashing is suspected.
  */
 
-#include <linux/config.h>
 #include <linux/time.h>
 #include <linux/kernel.h>
 #include <linux/errno.h>
@@ -24,7 +23,6 @@
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/pagemap.h>
-#include <linux/mempool.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/nfs_fs.h>
 #include <linux/nfs_page.h>
 
 #include <asm/system.h>
 
+#include "internal.h"
+#include "iostat.h"
+
 #define NFSDBG_FACILITY                NFSDBG_PAGECACHE
 
 static int nfs_pagein_one(struct list_head *, struct inode *);
-static void nfs_readpage_result_partial(struct nfs_read_data *, int);
-static void nfs_readpage_result_full(struct nfs_read_data *, int);
+static const struct rpc_call_ops nfs_read_partial_ops;
+static const struct rpc_call_ops nfs_read_full_ops;
 
-static kmem_cache_t *nfs_rdata_cachep;
+static struct kmem_cache *nfs_rdata_cachep;
 static mempool_t *nfs_rdata_mempool;
 
 #define MIN_POOL_READ  (32)
 
-static __inline__ struct nfs_read_data *nfs_readdata_alloc(void)
+struct nfs_read_data *nfs_readdata_alloc(size_t len)
 {
-       struct nfs_read_data   *p;
-       p = (struct nfs_read_data *)mempool_alloc(nfs_rdata_mempool, SLAB_NOFS);
+       unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
+       struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
+
        if (p) {
                memset(p, 0, sizeof(*p));
                INIT_LIST_HEAD(&p->pages);
+               p->npages = pagecount;
+               if (pagecount <= ARRAY_SIZE(p->page_array))
+                       p->pagevec = p->page_array;
+               else {
+                       p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
+                       if (!p->pagevec) {
+                               mempool_free(p, nfs_rdata_mempool);
+                               p = NULL;
+                       }
+               }
        }
        return p;
 }
 
-static __inline__ void nfs_readdata_free(struct nfs_read_data *p)
+static void nfs_readdata_rcu_free(struct rcu_head *head)
 {
+       struct nfs_read_data *p = container_of(head, struct nfs_read_data, task.u.tk_rcu);
+       if (p && (p->pagevec != &p->page_array[0]))
+               kfree(p->pagevec);
        mempool_free(p, nfs_rdata_mempool);
 }
 
-static void nfs_readdata_release(struct rpc_task *task)
+static void nfs_readdata_free(struct nfs_read_data *rdata)
 {
-        struct nfs_read_data   *data = (struct nfs_read_data *)task->tk_calldata;
-        nfs_readdata_free(data);
+       call_rcu_bh(&rdata->task.u.tk_rcu, nfs_readdata_rcu_free);
 }
 
-static
-unsigned int nfs_page_length(struct inode *inode, struct page *page)
+void nfs_readdata_release(void *data)
 {
-       loff_t i_size = i_size_read(inode);
-       unsigned long idx;
-
-       if (i_size <= 0)
-               return 0;
-       idx = (i_size - 1) >> PAGE_CACHE_SHIFT;
-       if (page->index > idx)
-               return 0;
-       if (page->index != idx)
-               return PAGE_CACHE_SIZE;
-       return 1 + ((i_size - 1) & (PAGE_CACHE_SIZE - 1));
+        nfs_readdata_free(data);
 }
 
 static
@@ -90,30 +93,61 @@ int nfs_return_empty_page(struct page *page)
        return 0;
 }
 
+static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
+{
+       unsigned int remainder = data->args.count - data->res.count;
+       unsigned int base = data->args.pgbase + data->res.count;
+       unsigned int pglen;
+       struct page **pages;
+
+       if (data->res.eof == 0 || remainder == 0)
+               return;
+       /*
+        * Note: "remainder" can never be negative, since we check for
+        *      this in the XDR code.
+        */
+       pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
+       base &= ~PAGE_CACHE_MASK;
+       pglen = PAGE_CACHE_SIZE - base;
+       for (;;) {
+               if (remainder <= pglen) {
+                       memclear_highpage_flush(*pages, base, remainder);
+                       break;
+               }
+               memclear_highpage_flush(*pages, base, pglen);
+               pages++;
+               remainder -= pglen;
+               pglen = PAGE_CACHE_SIZE;
+               base = 0;
+       }
+}
+
 /*
  * Read a page synchronously.
  */
-static int
-nfs_readpage_sync(struct file *file, struct inode *inode, struct page *page)
+static int nfs_readpage_sync(struct nfs_open_context *ctx, struct inode *inode,
+               struct page *page)
 {
        unsigned int    rsize = NFS_SERVER(inode)->rsize;
        unsigned int    count = PAGE_CACHE_SIZE;
-       int             result;
-       struct nfs_read_data    rdata = {
-               .flags          = (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0),
-               .cred           = NULL,
-               .inode          = inode,
-               .args           = {
-                       .fh             = NFS_FH(inode),
-                       .lockowner      = current->files,
-                       .pages          = &page,
-                       .pgbase         = 0UL,
-                       .count          = rsize,
-               },
-               .res            = {
-                       .fattr          = &rdata.fattr,
-               }
-       };
+       int result = -ENOMEM;
+       struct nfs_read_data *rdata;
+
+       rdata = nfs_readdata_alloc(count);
+       if (!rdata)
+               goto out_unlock;
+
+       memset(rdata, 0, sizeof(*rdata));
+       rdata->flags = (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0);
+       rdata->cred = ctx->cred;
+       rdata->inode = inode;
+       INIT_LIST_HEAD(&rdata->pages);
+       rdata->args.fh = NFS_FH(inode);
+       rdata->args.context = ctx;
+       rdata->args.pages = &page;
+       rdata->args.pgbase = 0UL;
+       rdata->args.count = rsize;
+       rdata->res.fattr = &rdata->fattr;
 
        dprintk("NFS: nfs_readpage_sync(%p)\n", page);
 
@@ -123,19 +157,19 @@ nfs_readpage_sync(struct file *file, struct inode *inode, struct page *page)
         */
        do {
                if (count < rsize)
-                       rdata.args.count = count;
-               rdata.res.count = rdata.args.count;
-               rdata.args.offset = page_offset(page) + rdata.args.pgbase;
+                       rdata->args.count = count;
+               rdata->res.count = rdata->args.count;
+               rdata->args.offset = page_offset(page) + rdata->args.pgbase;
 
                dprintk("NFS: nfs_proc_read(%s, (%s/%Ld), %Lu, %u)\n",
-                       NFS_SERVER(inode)->hostname,
+                       NFS_SERVER(inode)->nfs_client->cl_hostname,
                        inode->i_sb->s_id,
                        (long long)NFS_FILEID(inode),
-                       (unsigned long long)rdata.args.pgbase,
-                       rdata.args.count);
+                       (unsigned long long)rdata->args.pgbase,
+                       rdata->args.count);
 
                lock_kernel();
-               result = NFS_PROTO(inode)->read(&rdata, file);
+               result = NFS_PROTO(inode)->read(rdata);
                unlock_kernel();
 
                /*
@@ -148,38 +182,44 @@ nfs_readpage_sync(struct file *file, struct inode *inode, struct page *page)
                        goto io_error;
                }
                count -= result;
-               rdata.args.pgbase += result;
+               rdata->args.pgbase += result;
+               nfs_add_stats(inode, NFSIOS_SERVERREADBYTES, result);
+
                /* Note: result == 0 should only happen if we're caching
                 * a write that extends the file and punches a hole.
                 */
-               if (rdata.res.eof != 0 || result == 0)
+               if (rdata->res.eof != 0 || result == 0)
                        break;
        } while (count);
-       NFS_FLAGS(inode) |= NFS_INO_INVALID_ATIME;
-
-       if (count)
-               memclear_highpage_flush(page, rdata.args.pgbase, count);
-       SetPageUptodate(page);
-       if (PageError(page))
-               ClearPageError(page);
+       spin_lock(&inode->i_lock);
+       NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
+       spin_unlock(&inode->i_lock);
+
+       if (rdata->res.eof || rdata->res.count == rdata->args.count) {
+               SetPageUptodate(page);
+               if (rdata->res.eof && count != 0)
+                       memclear_highpage_flush(page, rdata->args.pgbase, count);
+       }
        result = 0;
 
 io_error:
+       nfs_readdata_free(rdata);
+out_unlock:
        unlock_page(page);
        return result;
 }
 
-static int
-nfs_readpage_async(struct file *file, struct inode *inode, struct page *page)
+static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
+               struct page *page)
 {
        LIST_HEAD(one_request);
        struct nfs_page *new;
        unsigned int len;
 
-       len = nfs_page_length(inode, page);
+       len = nfs_page_length(page);
        if (len == 0)
                return nfs_return_empty_page(page);
-       new = nfs_create_request(file, inode, page, 0, len);
+       new = nfs_create_request(ctx, inode, page, 0, len);
        if (IS_ERR(new)) {
                unlock_page(page);
                return PTR_ERR(new);
@@ -187,7 +227,6 @@ nfs_readpage_async(struct file *file, struct inode *inode, struct page *page)
        if (len < PAGE_CACHE_SIZE)
                memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
 
-       nfs_lock_request(new);
        nfs_list_add_request(new, &one_request);
        nfs_pagein_one(&one_request, inode);
        return 0;
@@ -197,47 +236,47 @@ static void nfs_readpage_release(struct nfs_page *req)
 {
        unlock_page(req->wb_page);
 
-       nfs_clear_request(req);
-       nfs_release_request(req);
-       nfs_unlock_request(req);
-
        dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
-                       req->wb_inode->i_sb->s_id,
-                       (long long)NFS_FILEID(req->wb_inode),
+                       req->wb_context->dentry->d_inode->i_sb->s_id,
+                       (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
                        req->wb_bytes,
                        (long long)req_offset(req));
+       nfs_clear_request(req);
+       nfs_release_request(req);
 }
 
 /*
  * Set up the NFS read request struct
  */
 static void nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data,
+               const struct rpc_call_ops *call_ops,
                unsigned int count, unsigned int offset)
 {
        struct inode            *inode;
+       int flags;
 
        data->req         = req;
-       data->inode       = inode = req->wb_inode;
-       data->cred        = req->wb_cred;
+       data->inode       = inode = req->wb_context->dentry->d_inode;
+       data->cred        = req->wb_context->cred;
 
        data->args.fh     = NFS_FH(inode);
        data->args.offset = req_offset(req) + offset;
        data->args.pgbase = req->wb_pgbase + offset;
        data->args.pages  = data->pagevec;
        data->args.count  = count;
-       data->args.lockowner = req->wb_lockowner;
-       data->args.state  = req->wb_state;
+       data->args.context = req->wb_context;
 
        data->res.fattr   = &data->fattr;
        data->res.count   = count;
        data->res.eof     = 0;
+       nfs_fattr_init(&data->fattr);
 
+       /* Set up the initial task struct. */
+       flags = RPC_TASK_ASYNC | (IS_SWAPFILE(inode)? NFS_RPC_SWAPFLAGS : 0);
+       rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
        NFS_PROTO(inode)->read_setup(data);
 
        data->task.tk_cookie = (unsigned long)inode;
-       data->task.tk_calldata = data;
-       /* Release requests */
-       data->task.tk_release = nfs_readdata_release;
 
        dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
                        data->task.tk_pid,
@@ -269,9 +308,7 @@ static void nfs_execute_read(struct nfs_read_data *data)
        sigset_t oldset;
 
        rpc_clnt_sigmask(clnt, &oldset);
-       lock_kernel();
        rpc_execute(&data->task);
-       unlock_kernel();
        rpc_clnt_sigunmask(clnt, &oldset);
 }
 
@@ -293,24 +330,25 @@ static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
        struct nfs_page *req = nfs_list_entry(head->next);
        struct page *page = req->wb_page;
        struct nfs_read_data *data;
-       unsigned int rsize = NFS_SERVER(inode)->rsize;
-       unsigned int nbytes, offset;
+       size_t rsize = NFS_SERVER(inode)->rsize, nbytes;
+       unsigned int offset;
        int requests = 0;
        LIST_HEAD(list);
 
        nfs_list_remove_request(req);
 
        nbytes = req->wb_bytes;
-       for(;;) {
-               data = nfs_readdata_alloc();
+       do {
+               size_t len = min(nbytes,rsize);
+
+               data = nfs_readdata_alloc(len);
                if (!data)
                        goto out_bad;
+               INIT_LIST_HEAD(&data->pages);
                list_add(&data->pages, &list);
                requests++;
-               if (nbytes <= rsize)
-                       break;
-               nbytes -= rsize;
-       }
+               nbytes -= len;
+       } while(nbytes != 0);
        atomic_set(&req->wb_complete, requests);
 
        ClearPageError(page);
@@ -321,14 +359,15 @@ static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
                list_del_init(&data->pages);
 
                data->pagevec[0] = page;
-               data->complete = nfs_readpage_result_partial;
 
                if (nbytes > rsize) {
-                       nfs_read_rpcsetup(req, data, rsize, offset);
+                       nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
+                                       rsize, offset);
                        offset += rsize;
                        nbytes -= rsize;
                } else {
-                       nfs_read_rpcsetup(req, data, nbytes, offset);
+                       nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
+                                       nbytes, offset);
                        nbytes = 0;
                }
                nfs_execute_read(data);
@@ -357,10 +396,11 @@ static int nfs_pagein_one(struct list_head *head, struct inode *inode)
        if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
                return nfs_pagein_multi(head, inode);
 
-       data = nfs_readdata_alloc();
+       data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
        if (!data)
                goto out_bad;
 
+       INIT_LIST_HEAD(&data->pages);
        pages = data->pagevec;
        count = 0;
        while (!list_empty(head)) {
@@ -373,8 +413,7 @@ static int nfs_pagein_one(struct list_head *head, struct inode *inode)
        }
        req = nfs_list_entry(data->pages.next);
 
-       data->complete = nfs_readpage_result_full;
-       nfs_read_rpcsetup(req, data, count, 0);
+       nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0);
 
        nfs_execute_read(data);
        return 0;
@@ -383,7 +422,7 @@ out_bad:
        return -ENOMEM;
 }
 
-int
+static int
 nfs_pagein_list(struct list_head *head, int rpages)
 {
        LIST_HEAD(one_request);
@@ -394,7 +433,7 @@ nfs_pagein_list(struct list_head *head, int rpages)
        while (!list_empty(head)) {
                pages += nfs_coalesce_requests(head, &one_request, rpages);
                req = nfs_list_entry(one_request.next);
-               error = nfs_pagein_one(&one_request, req->wb_inode);
+               error = nfs_pagein_one(&one_request, req->wb_context->dentry->d_inode);
                if (error < 0)
                        break;
        }
@@ -405,26 +444,74 @@ nfs_pagein_list(struct list_head *head, int rpages)
        return error;
 }
 
+/*
+ * This is the callback from RPC telling us whether a reply was
+ * received or some error occurred (timeout or socket shutdown).
+ */
+int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data)
+{
+       int status;
+
+       dprintk("%s: %4d, (status %d)\n", __FUNCTION__, task->tk_pid,
+                       task->tk_status);
+
+       status = NFS_PROTO(data->inode)->read_done(task, data);
+       if (status != 0)
+               return status;
+
+       nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count);
+
+       if (task->tk_status == -ESTALE) {
+               set_bit(NFS_INO_STALE, &NFS_FLAGS(data->inode));
+               nfs_mark_for_revalidate(data->inode);
+       }
+       spin_lock(&data->inode->i_lock);
+       NFS_I(data->inode)->cache_validity |= NFS_INO_INVALID_ATIME;
+       spin_unlock(&data->inode->i_lock);
+       return 0;
+}
+
+static int nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data)
+{
+       struct nfs_readargs *argp = &data->args;
+       struct nfs_readres *resp = &data->res;
+
+       if (resp->eof || resp->count == argp->count)
+               return 0;
+
+       /* This is a short read! */
+       nfs_inc_stats(data->inode, NFSIOS_SHORTREAD);
+       /* Has the server at least made some progress? */
+       if (resp->count == 0)
+               return 0;
+
+       /* Yes, so retry the read at the end of the data */
+       argp->offset += resp->count;
+       argp->pgbase += resp->count;
+       argp->count -= resp->count;
+       rpc_restart_call(task);
+       return -EAGAIN;
+}
+
 /*
  * Handle a read reply that fills part of a page.
  */
-static void nfs_readpage_result_partial(struct nfs_read_data *data, int status)
+static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata)
 {
+       struct nfs_read_data *data = calldata;
        struct nfs_page *req = data->req;
        struct page *page = req->wb_page;
  
-       if (status >= 0) {
-               unsigned int request = data->args.count;
-               unsigned int result = data->res.count;
-
-               if (result < request) {
-                       memclear_highpage_flush(page,
-                                               data->args.pgbase + result,
-                                               request - result);
-               }
-       } else
-               SetPageError(page);
+       if (nfs_readpage_result(task, data) != 0)
+               return;
 
+       if (likely(task->tk_status >= 0)) {
+               nfs_readpage_truncate_uninitialised_page(data);
+               if (nfs_readpage_retry(task, data) != 0)
+                       return;
+       }
+       if (unlikely(task->tk_status < 0))
+               SetPageError(page);
        if (atomic_dec_and_test(&req->wb_complete)) {
                if (!PageError(page))
                        SetPageUptodate(page);
@@ -432,65 +519,67 @@ static void nfs_readpage_result_partial(struct nfs_read_data *data, int status)
        }
 }
 
+static const struct rpc_call_ops nfs_read_partial_ops = {
+       .rpc_call_done = nfs_readpage_result_partial,
+       .rpc_release = nfs_readdata_release,
+};
+
+static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data)
+{
+       unsigned int count = data->res.count;
+       unsigned int base = data->args.pgbase;
+       struct page **pages;
+
+       if (data->res.eof)
+               count = data->args.count;
+       if (unlikely(count == 0))
+               return;
+       pages = &data->args.pages[base >> PAGE_CACHE_SHIFT];
+       base &= ~PAGE_CACHE_MASK;
+       count += base;
+       for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++)
+               SetPageUptodate(*pages);
+       if (count == 0)
+               return;
+       /* Was this a short read? */
+       if (data->res.eof || data->res.count == data->args.count)
+               SetPageUptodate(*pages);
+}
+
 /*
  * This is the callback from RPC telling us whether a reply was
  * received or some error occurred (timeout or socket shutdown).
  */
-static void nfs_readpage_result_full(struct nfs_read_data *data, int status)
+static void nfs_readpage_result_full(struct rpc_task *task, void *calldata)
 {
-       unsigned int count = data->res.count;
+       struct nfs_read_data *data = calldata;
 
+       if (nfs_readpage_result(task, data) != 0)
+               return;
+       /*
+        * Note: nfs_readpage_retry may change the values of
+        * data->args. In the multi-page case, we therefore need
+        * to ensure that we call nfs_readpage_set_pages_uptodate()
+        * first.
+        */
+       if (likely(task->tk_status >= 0)) {
+               nfs_readpage_truncate_uninitialised_page(data);
+               nfs_readpage_set_pages_uptodate(data);
+               if (nfs_readpage_retry(task, data) != 0)
+                       return;
+       }
        while (!list_empty(&data->pages)) {
                struct nfs_page *req = nfs_list_entry(data->pages.next);
-               struct page *page = req->wb_page;
-               nfs_list_remove_request(req);
 
-               if (status >= 0) {
-                       if (count < PAGE_CACHE_SIZE) {
-                               if (count < req->wb_bytes)
-                                       memclear_highpage_flush(page,
-                                                       req->wb_pgbase + count,
-                                                       req->wb_bytes - count);
-                               count = 0;
-                       } else
-                               count -= PAGE_CACHE_SIZE;
-                       SetPageUptodate(page);
-               } else
-                       SetPageError(page);
+               nfs_list_remove_request(req);
                nfs_readpage_release(req);
        }
 }
 
-/*
- * This is the callback from RPC telling us whether a reply was
- * received or some error occurred (timeout or socket shutdown).
- */
-void nfs_readpage_result(struct rpc_task *task)
-{
-       struct nfs_read_data *data = (struct nfs_read_data *)task->tk_calldata;
-       struct nfs_readargs *argp = &data->args;
-       struct nfs_readres *resp = &data->res;
-       int status = task->tk_status;
-
-       dprintk("NFS: %4d nfs_readpage_result, (status %d)\n",
-               task->tk_pid, status);
-
-       /* Is this a short read? */
-       if (task->tk_status >= 0 && resp->count < argp->count && !resp->eof) {
-               /* Has the server at least made some progress? */
-               if (resp->count != 0) {
-                       /* Yes, so retry the read at the end of the data */
-                       argp->offset += resp->count;
-                       argp->pgbase += resp->count;
-                       argp->count -= resp->count;
-                       rpc_restart_call(task);
-                       return;
-               }
-               task->tk_status = -EIO;
-       }
-       NFS_FLAGS(data->inode) |= NFS_INO_INVALID_ATIME;
-       data->complete(data, status);
-}
+static const struct rpc_call_ops nfs_read_full_ops = {
+       .rpc_call_done = nfs_readpage_result_full,
+       .rpc_release = nfs_readdata_release,
+};
 
 /*
  * Read a page over NFS.
@@ -498,14 +587,17 @@ void nfs_readpage_result(struct rpc_task *task)
  *  -  The error flag is set for this page. This happens only when a
  *     previous async read operation failed.
  */
-int
-nfs_readpage(struct file *file, struct page *page)
+int nfs_readpage(struct file *file, struct page *page)
 {
+       struct nfs_open_context *ctx;
        struct inode *inode = page->mapping->host;
        int             error;
 
        dprintk("NFS: nfs_readpage (%p %ld@%lu)\n",
                page, PAGE_CACHE_SIZE, page->index);
+       nfs_inc_stats(inode, NFSIOS_VFSREADPAGE);
+       nfs_add_stats(inode, NFSIOS_READPAGES, 1);
+
        /*
         * Try to flush any pending writes to the file..
         *
@@ -517,25 +609,38 @@ nfs_readpage(struct file *file, struct page *page)
        if (error)
                goto out_error;
 
+       error = -ESTALE;
+       if (NFS_STALE(inode))
+               goto out_error;
+
+       if (file == NULL) {
+               error = -EBADF;
+               ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
+               if (ctx == NULL)
+                       goto out_error;
+       } else
+               ctx = get_nfs_open_context((struct nfs_open_context *)
+                               file->private_data);
        if (!IS_SYNC(inode)) {
-               error = nfs_readpage_async(file, inode, page);
+               error = nfs_readpage_async(ctx, inode, page);
                goto out;
        }
 
-       error = nfs_readpage_sync(file, inode, page);
+       error = nfs_readpage_sync(ctx, inode, page);
        if (error < 0 && IS_SWAPFILE(inode))
                printk("Aiee.. nfs swap-in of page failed!\n");
 out:
+       put_nfs_open_context(ctx);
        return error;
 
 out_error:
        unlock_page(page);
-       goto out;
+       return error;
 }
 
 struct nfs_readdesc {
        struct list_head *head;
-       struct file *filp;
+       struct nfs_open_context *ctx;
 };
 
 static int
@@ -547,10 +652,10 @@ readpage_async_filler(void *data, struct page *page)
        unsigned int len;
 
        nfs_wb_page(inode, page);
-       len = nfs_page_length(inode, page);
+       len = nfs_page_length(page);
        if (len == 0)
                return nfs_return_empty_page(page);
-       new = nfs_create_request(desc->filp, inode, page, 0, len);
+       new = nfs_create_request(desc->ctx, inode, page, 0, len);
        if (IS_ERR(new)) {
                        SetPageError(page);
                        unlock_page(page);
@@ -558,39 +663,50 @@ readpage_async_filler(void *data, struct page *page)
        }
        if (len < PAGE_CACHE_SIZE)
                memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
-       nfs_lock_request(new);
        nfs_list_add_request(new, desc->head);
        return 0;
 }
 
-int
-nfs_readpages(struct file *filp, struct address_space *mapping,
+int nfs_readpages(struct file *filp, struct address_space *mapping,
                struct list_head *pages, unsigned nr_pages)
 {
        LIST_HEAD(head);
        struct nfs_readdesc desc = {
-               .filp           = filp,
                .head           = &head,
        };
        struct inode *inode = mapping->host;
        struct nfs_server *server = NFS_SERVER(inode);
-       int ret;
+       int ret = -ESTALE;
 
        dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
                        inode->i_sb->s_id,
                        (long long)NFS_FILEID(inode),
                        nr_pages);
+       nfs_inc_stats(inode, NFSIOS_VFSREADPAGES);
+
+       if (NFS_STALE(inode))
+               goto out;
 
+       if (filp == NULL) {
+               desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ);
+               if (desc.ctx == NULL)
+                       return -EBADF;
+       } else
+               desc.ctx = get_nfs_open_context((struct nfs_open_context *)
+                               filp->private_data);
        ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc);
        if (!list_empty(&head)) {
                int err = nfs_pagein_list(&head, server->rpages);
                if (!ret)
+                       nfs_add_stats(inode, NFSIOS_READPAGES, err);
                        ret = err;
        }
+       put_nfs_open_context(desc.ctx);
+out:
        return ret;
 }
 
-int nfs_init_readpagecache(void)
+int __init nfs_init_readpagecache(void)
 {
        nfs_rdata_cachep = kmem_cache_create("nfs_read_data",
                                             sizeof(struct nfs_read_data),
@@ -599,10 +715,8 @@ int nfs_init_readpagecache(void)
        if (nfs_rdata_cachep == NULL)
                return -ENOMEM;
 
-       nfs_rdata_mempool = mempool_create(MIN_POOL_READ,
-                                          mempool_alloc_slab,
-                                          mempool_free_slab,
-                                          nfs_rdata_cachep);
+       nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ,
+                                                    nfs_rdata_cachep);
        if (nfs_rdata_mempool == NULL)
                return -ENOMEM;
 
@@ -612,6 +726,5 @@ int nfs_init_readpagecache(void)
 void nfs_destroy_readpagecache(void)
 {
        mempool_destroy(nfs_rdata_mempool);
-       if (kmem_cache_destroy(nfs_rdata_cachep))
-               printk(KERN_INFO "nfs_read_data: not all structures were freed\n");
+       kmem_cache_destroy(nfs_rdata_cachep);
 }