*
* Partial copy of Linus' read cache modifications to fs/nfs/file.c
* modified for async RPC by okir@monad.swb.de
- *
- * We do an ugly hack here in order to return proper error codes to the
- * user program when a read request failed: since generic_file_read
- * only checks the return value of inode->i_op->readpage() which is always 0
- * for async RPC, we set the error bit of the page to 1 when an error occurs,
- * and make nfs_readpage transmit requests synchronously when encountering this.
- * This is only a small problem, though, since we now retry all operations
- * within the RPC code when root squashing is suspected.
*/
#include <linux/time.h>
#define NFSDBG_FACILITY NFSDBG_PAGECACHE
-static int nfs_pagein_one(struct list_head *, struct inode *);
+static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int);
+static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int);
static const struct rpc_call_ops nfs_read_partial_ops;
static const struct rpc_call_ops nfs_read_full_ops;
#define MIN_POOL_READ (32)
-struct nfs_read_data *nfs_readdata_alloc(size_t len)
+struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount)
{
- unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_NOFS);
if (p) {
static
int nfs_return_empty_page(struct page *page)
{
- memclear_highpage_flush(page, 0, PAGE_CACHE_SIZE);
+ zero_user_page(page, 0, PAGE_CACHE_SIZE, KM_USER0);
SetPageUptodate(page);
unlock_page(page);
return 0;
pglen = PAGE_CACHE_SIZE - base;
for (;;) {
if (remainder <= pglen) {
- memclear_highpage_flush(*pages, base, remainder);
+ zero_user_page(*pages, base, remainder, KM_USER0);
break;
}
- memclear_highpage_flush(*pages, base, pglen);
+ zero_user_page(*pages, base, pglen, KM_USER0);
pages++;
remainder -= pglen;
pglen = PAGE_CACHE_SIZE;
}
}
-/*
- * Read a page synchronously.
- */
-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 = -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);
-
- /*
- * This works now because the socket layer never tries to DMA
- * into this buffer directly.
- */
- do {
- if (count < rsize)
- 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)->nfs_client->cl_hostname,
- inode->i_sb->s_id,
- (long long)NFS_FILEID(inode),
- (unsigned long long)rdata->args.pgbase,
- rdata->args.count);
-
- lock_kernel();
- result = NFS_PROTO(inode)->read(rdata);
- unlock_kernel();
-
- /*
- * Even if we had a partial success we can't mark the page
- * cache valid.
- */
- if (result < 0) {
- if (result == -EISDIR)
- result = -EINVAL;
- goto io_error;
- }
- count -= 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)
- break;
- } while (count);
- 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 nfs_open_context *ctx, struct inode *inode,
struct page *page)
{
return PTR_ERR(new);
}
if (len < PAGE_CACHE_SIZE)
- memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
+ zero_user_page(page, len, PAGE_CACHE_SIZE - len, KM_USER0);
nfs_list_add_request(new, &one_request);
- nfs_pagein_one(&one_request, inode);
+ if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
+ nfs_pagein_multi(inode, &one_request, 1, len, 0);
+ else
+ nfs_pagein_one(inode, &one_request, 1, len, 0);
return 0;
}
data->task.tk_cookie = (unsigned long)inode;
- dprintk("NFS: %4d initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
+ dprintk("NFS: %5u initiated read call (req %s/%Ld, %u bytes @ offset %Lu)\n",
data->task.tk_pid,
inode->i_sb->s_id,
(long long)NFS_FILEID(inode),
* won't see the new data until our attribute cache is updated. This is more
* or less conventional NFS client behavior.
*/
-static int nfs_pagein_multi(struct list_head *head, struct inode *inode)
+static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
{
struct nfs_page *req = nfs_list_entry(head->next);
struct page *page = req->wb_page;
nfs_list_remove_request(req);
- nbytes = req->wb_bytes;
+ nbytes = count;
do {
size_t len = min(nbytes,rsize);
- data = nfs_readdata_alloc(len);
+ data = nfs_readdata_alloc(1);
if (!data)
goto out_bad;
INIT_LIST_HEAD(&data->pages);
ClearPageError(page);
offset = 0;
- nbytes = req->wb_bytes;
+ nbytes = count;
do {
data = list_entry(list.next, struct nfs_read_data, pages);
list_del_init(&data->pages);
data->pagevec[0] = page;
- if (nbytes > rsize) {
- nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
- rsize, offset);
- offset += rsize;
- nbytes -= rsize;
- } else {
- nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
- nbytes, offset);
- nbytes = 0;
- }
+ if (nbytes < rsize)
+ rsize = nbytes;
+ nfs_read_rpcsetup(req, data, &nfs_read_partial_ops,
+ rsize, offset);
+ offset += rsize;
+ nbytes -= rsize;
nfs_execute_read(data);
} while (nbytes != 0);
return -ENOMEM;
}
-static int nfs_pagein_one(struct list_head *head, struct inode *inode)
+static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags)
{
struct nfs_page *req;
struct page **pages;
struct nfs_read_data *data;
- unsigned int count;
- if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE)
- return nfs_pagein_multi(head, inode);
-
- data = nfs_readdata_alloc(NFS_SERVER(inode)->rsize);
+ data = nfs_readdata_alloc(npages);
if (!data)
goto out_bad;
INIT_LIST_HEAD(&data->pages);
pages = data->pagevec;
- count = 0;
while (!list_empty(head)) {
req = nfs_list_entry(head->next);
nfs_list_remove_request(req);
nfs_list_add_request(req, &data->pages);
ClearPageError(req->wb_page);
*pages++ = req->wb_page;
- count += req->wb_bytes;
}
req = nfs_list_entry(data->pages.next);
return -ENOMEM;
}
-static int
-nfs_pagein_list(struct list_head *head, int rpages)
-{
- LIST_HEAD(one_request);
- struct nfs_page *req;
- int error = 0;
- unsigned int pages = 0;
-
- 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_context->dentry->d_inode);
- if (error < 0)
- break;
- }
- if (error >= 0)
- return pages;
-
- nfs_async_read_error(head);
- return error;
-}
-
/*
* This is the callback from RPC telling us whether a reply was
* received or some error occurred (timeout or socket shutdown).
{
int status;
- dprintk("%s: %4d, (status %d)\n", __FUNCTION__, task->tk_pid,
+ dprintk("NFS: %s: %5u, (status %d)\n", __FUNCTION__, task->tk_pid,
task->tk_status);
status = NFS_PROTO(data->inode)->read_done(task, data);
} else
ctx = get_nfs_open_context((struct nfs_open_context *)
file->private_data);
- if (!IS_SYNC(inode)) {
- error = nfs_readpage_async(ctx, inode, page);
- goto out;
- }
- error = nfs_readpage_sync(ctx, inode, page);
- if (error < 0 && IS_SWAPFILE(inode))
- printk("Aiee.. nfs swap-in of page failed!\n");
-out:
+ error = nfs_readpage_async(ctx, inode, page);
+
put_nfs_open_context(ctx);
return error;
}
struct nfs_readdesc {
- struct list_head *head;
+ struct nfs_pageio_descriptor *pgio;
struct nfs_open_context *ctx;
};
return PTR_ERR(new);
}
if (len < PAGE_CACHE_SIZE)
- memclear_highpage_flush(page, len, PAGE_CACHE_SIZE - len);
- nfs_list_add_request(new, desc->head);
+ zero_user_page(page, len, PAGE_CACHE_SIZE - len, KM_USER0);
+ nfs_pageio_add_request(desc->pgio, new);
return 0;
}
int nfs_readpages(struct file *filp, struct address_space *mapping,
struct list_head *pages, unsigned nr_pages)
{
- LIST_HEAD(head);
+ struct nfs_pageio_descriptor pgio;
struct nfs_readdesc desc = {
- .head = &head,
+ .pgio = &pgio,
};
struct inode *inode = mapping->host;
struct nfs_server *server = NFS_SERVER(inode);
+ size_t rsize = server->rsize;
+ unsigned long npages;
int ret = -ESTALE;
dprintk("NFS: nfs_readpages (%s/%Ld %d)\n",
} else
desc.ctx = get_nfs_open_context((struct nfs_open_context *)
filp->private_data);
+ if (rsize < PAGE_CACHE_SIZE)
+ nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
+ else
+ nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0);
+
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;
- }
+
+ nfs_pageio_complete(&pgio);
+ npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+ nfs_add_stats(inode, NFSIOS_READPAGES, npages);
put_nfs_open_context(desc.ctx);
out:
return ret;