xref: /openbmc/linux/fs/nfs/write.c (revision 03fa9e84)
11da177e4SLinus Torvalds /*
21da177e4SLinus Torvalds  * linux/fs/nfs/write.c
31da177e4SLinus Torvalds  *
47c85d900STrond Myklebust  * Write file data over NFS.
51da177e4SLinus Torvalds  *
61da177e4SLinus Torvalds  * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
71da177e4SLinus Torvalds  */
81da177e4SLinus Torvalds 
91da177e4SLinus Torvalds #include <linux/types.h>
101da177e4SLinus Torvalds #include <linux/slab.h>
111da177e4SLinus Torvalds #include <linux/mm.h>
121da177e4SLinus Torvalds #include <linux/pagemap.h>
131da177e4SLinus Torvalds #include <linux/file.h>
141da177e4SLinus Torvalds #include <linux/writeback.h>
1589a09141SPeter Zijlstra #include <linux/swap.h>
161da177e4SLinus Torvalds 
171da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h>
181da177e4SLinus Torvalds #include <linux/nfs_fs.h>
191da177e4SLinus Torvalds #include <linux/nfs_mount.h>
201da177e4SLinus Torvalds #include <linux/nfs_page.h>
213fcfab16SAndrew Morton #include <linux/backing-dev.h>
223fcfab16SAndrew Morton 
231da177e4SLinus Torvalds #include <asm/uaccess.h>
241da177e4SLinus Torvalds 
251da177e4SLinus Torvalds #include "delegation.h"
2649a70f27STrond Myklebust #include "internal.h"
2791d5b470SChuck Lever #include "iostat.h"
281da177e4SLinus Torvalds 
291da177e4SLinus Torvalds #define NFSDBG_FACILITY		NFSDBG_PAGECACHE
301da177e4SLinus Torvalds 
311da177e4SLinus Torvalds #define MIN_POOL_WRITE		(32)
321da177e4SLinus Torvalds #define MIN_POOL_COMMIT		(4)
331da177e4SLinus Torvalds 
341da177e4SLinus Torvalds /*
351da177e4SLinus Torvalds  * Local function declarations
361da177e4SLinus Torvalds  */
371da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*,
381da177e4SLinus Torvalds 					    struct page *,
391da177e4SLinus Torvalds 					    unsigned int, unsigned int);
40c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
41c63c7b05STrond Myklebust 				  struct inode *inode, int ioflags);
42f8512ad0SFred Isaman static void nfs_redirty_request(struct nfs_page *req);
43788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops;
44788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops;
45788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops;
461da177e4SLinus Torvalds 
47e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep;
483feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool;
491da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool;
501da177e4SLinus Torvalds 
51c9d8f89dSTrond Myklebust struct nfs_write_data *nfs_commitdata_alloc(void)
521da177e4SLinus Torvalds {
53e6b4f8daSChristoph Lameter 	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
5440859d7eSChuck Lever 
551da177e4SLinus Torvalds 	if (p) {
561da177e4SLinus Torvalds 		memset(p, 0, sizeof(*p));
571da177e4SLinus Torvalds 		INIT_LIST_HEAD(&p->pages);
581da177e4SLinus Torvalds 	}
591da177e4SLinus Torvalds 	return p;
601da177e4SLinus Torvalds }
611da177e4SLinus Torvalds 
625e4424afSTrond Myklebust void nfs_commit_free(struct nfs_write_data *p)
631da177e4SLinus Torvalds {
6440859d7eSChuck Lever 	if (p && (p->pagevec != &p->page_array[0]))
6540859d7eSChuck Lever 		kfree(p->pagevec);
661da177e4SLinus Torvalds 	mempool_free(p, nfs_commit_mempool);
671da177e4SLinus Torvalds }
681da177e4SLinus Torvalds 
698d5658c9STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
703feb2d49STrond Myklebust {
71e6b4f8daSChristoph Lameter 	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
723feb2d49STrond Myklebust 
733feb2d49STrond Myklebust 	if (p) {
743feb2d49STrond Myklebust 		memset(p, 0, sizeof(*p));
753feb2d49STrond Myklebust 		INIT_LIST_HEAD(&p->pages);
76e9f7bee1STrond Myklebust 		p->npages = pagecount;
770d0b5cb3SChuck Lever 		if (pagecount <= ARRAY_SIZE(p->page_array))
780d0b5cb3SChuck Lever 			p->pagevec = p->page_array;
793feb2d49STrond Myklebust 		else {
800d0b5cb3SChuck Lever 			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
810d0b5cb3SChuck Lever 			if (!p->pagevec) {
823feb2d49STrond Myklebust 				mempool_free(p, nfs_wdata_mempool);
833feb2d49STrond Myklebust 				p = NULL;
843feb2d49STrond Myklebust 			}
853feb2d49STrond Myklebust 		}
863feb2d49STrond Myklebust 	}
873feb2d49STrond Myklebust 	return p;
883feb2d49STrond Myklebust }
893feb2d49STrond Myklebust 
905e4424afSTrond Myklebust static void nfs_writedata_free(struct nfs_write_data *p)
913feb2d49STrond Myklebust {
923feb2d49STrond Myklebust 	if (p && (p->pagevec != &p->page_array[0]))
933feb2d49STrond Myklebust 		kfree(p->pagevec);
943feb2d49STrond Myklebust 	mempool_free(p, nfs_wdata_mempool);
953feb2d49STrond Myklebust }
963feb2d49STrond Myklebust 
97383ba719STrond Myklebust void nfs_writedata_release(void *data)
981da177e4SLinus Torvalds {
99383ba719STrond Myklebust 	struct nfs_write_data *wdata = data;
100383ba719STrond Myklebust 
101383ba719STrond Myklebust 	put_nfs_open_context(wdata->args.context);
1021da177e4SLinus Torvalds 	nfs_writedata_free(wdata);
1031da177e4SLinus Torvalds }
1041da177e4SLinus Torvalds 
1057b159fc1STrond Myklebust static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
1067b159fc1STrond Myklebust {
1077b159fc1STrond Myklebust 	ctx->error = error;
1087b159fc1STrond Myklebust 	smp_wmb();
1097b159fc1STrond Myklebust 	set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
1107b159fc1STrond Myklebust }
1117b159fc1STrond Myklebust 
112277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page)
113277459d2STrond Myklebust {
114277459d2STrond Myklebust 	struct nfs_page *req = NULL;
115277459d2STrond Myklebust 
116277459d2STrond Myklebust 	if (PagePrivate(page)) {
117277459d2STrond Myklebust 		req = (struct nfs_page *)page_private(page);
118277459d2STrond Myklebust 		if (req != NULL)
119c03b4024STrond Myklebust 			kref_get(&req->wb_kref);
120277459d2STrond Myklebust 	}
121277459d2STrond Myklebust 	return req;
122277459d2STrond Myklebust }
123277459d2STrond Myklebust 
124277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page)
125277459d2STrond Myklebust {
126587142f8STrond Myklebust 	struct inode *inode = page->mapping->host;
127277459d2STrond Myklebust 	struct nfs_page *req = NULL;
128277459d2STrond Myklebust 
129587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
130277459d2STrond Myklebust 	req = nfs_page_find_request_locked(page);
131587142f8STrond Myklebust 	spin_unlock(&inode->i_lock);
132277459d2STrond Myklebust 	return req;
133277459d2STrond Myklebust }
134277459d2STrond Myklebust 
1351da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */
1361da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
1371da177e4SLinus Torvalds {
1381da177e4SLinus Torvalds 	struct inode *inode = page->mapping->host;
1391da177e4SLinus Torvalds 	loff_t end, i_size = i_size_read(inode);
140ca52fec1STrond Myklebust 	pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
1411da177e4SLinus Torvalds 
1421da177e4SLinus Torvalds 	if (i_size > 0 && page->index < end_index)
1431da177e4SLinus Torvalds 		return;
1441da177e4SLinus Torvalds 	end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
1451da177e4SLinus Torvalds 	if (i_size >= end)
1461da177e4SLinus Torvalds 		return;
14791d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
1481da177e4SLinus Torvalds 	i_size_write(inode, end);
1491da177e4SLinus Torvalds }
1501da177e4SLinus Torvalds 
151a301b777STrond Myklebust /* A writeback failed: mark the page as bad, and invalidate the page cache */
152a301b777STrond Myklebust static void nfs_set_pageerror(struct page *page)
153a301b777STrond Myklebust {
154a301b777STrond Myklebust 	SetPageError(page);
155a301b777STrond Myklebust 	nfs_zap_mapping(page->mapping->host, page->mapping);
156a301b777STrond Myklebust }
157a301b777STrond Myklebust 
1581da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request
1591da177e4SLinus Torvalds  * covers the full page.
1601da177e4SLinus Torvalds  */
1611da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
1621da177e4SLinus Torvalds {
1631da177e4SLinus Torvalds 	if (PageUptodate(page))
1641da177e4SLinus Torvalds 		return;
1651da177e4SLinus Torvalds 	if (base != 0)
1661da177e4SLinus Torvalds 		return;
16749a70f27STrond Myklebust 	if (count != nfs_page_length(page))
1681da177e4SLinus Torvalds 		return;
1691da177e4SLinus Torvalds 	SetPageUptodate(page);
1701da177e4SLinus Torvalds }
1711da177e4SLinus Torvalds 
172e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1731da177e4SLinus Torvalds 		unsigned int offset, unsigned int count)
1741da177e4SLinus Torvalds {
1751da177e4SLinus Torvalds 	struct nfs_page	*req;
176e21195a7STrond Myklebust 	int ret;
1771da177e4SLinus Torvalds 
178e21195a7STrond Myklebust 	for (;;) {
179e21195a7STrond Myklebust 		req = nfs_update_request(ctx, page, offset, count);
180e21195a7STrond Myklebust 		if (!IS_ERR(req))
181e21195a7STrond Myklebust 			break;
182e21195a7STrond Myklebust 		ret = PTR_ERR(req);
183e21195a7STrond Myklebust 		if (ret != -EBUSY)
184e21195a7STrond Myklebust 			return ret;
185e21195a7STrond Myklebust 		ret = nfs_wb_page(page->mapping->host, page);
186e21195a7STrond Myklebust 		if (ret != 0)
187e21195a7STrond Myklebust 			return ret;
188e21195a7STrond Myklebust 	}
1891da177e4SLinus Torvalds 	/* Update file length */
1901da177e4SLinus Torvalds 	nfs_grow_file(page, offset, count);
191acee478aSTrond Myklebust 	nfs_clear_page_tag_locked(req);
192abd3e641STrond Myklebust 	return 0;
1931da177e4SLinus Torvalds }
1941da177e4SLinus Torvalds 
1951da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc)
1961da177e4SLinus Torvalds {
1971da177e4SLinus Torvalds 	if (wbc->for_reclaim)
198c63c7b05STrond Myklebust 		return FLUSH_HIGHPRI | FLUSH_STABLE;
1991da177e4SLinus Torvalds 	if (wbc->for_kupdate)
2001da177e4SLinus Torvalds 		return FLUSH_LOWPRI;
2011da177e4SLinus Torvalds 	return 0;
2021da177e4SLinus Torvalds }
2031da177e4SLinus Torvalds 
2041da177e4SLinus Torvalds /*
20589a09141SPeter Zijlstra  * NFS congestion control
20689a09141SPeter Zijlstra  */
20789a09141SPeter Zijlstra 
20889a09141SPeter Zijlstra int nfs_congestion_kb;
20989a09141SPeter Zijlstra 
21089a09141SPeter Zijlstra #define NFS_CONGESTION_ON_THRESH 	(nfs_congestion_kb >> (PAGE_SHIFT-10))
21189a09141SPeter Zijlstra #define NFS_CONGESTION_OFF_THRESH	\
21289a09141SPeter Zijlstra 	(NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
21389a09141SPeter Zijlstra 
2145a6d41b3STrond Myklebust static int nfs_set_page_writeback(struct page *page)
21589a09141SPeter Zijlstra {
2165a6d41b3STrond Myklebust 	int ret = test_set_page_writeback(page);
2175a6d41b3STrond Myklebust 
2185a6d41b3STrond Myklebust 	if (!ret) {
21989a09141SPeter Zijlstra 		struct inode *inode = page->mapping->host;
22089a09141SPeter Zijlstra 		struct nfs_server *nfss = NFS_SERVER(inode);
22189a09141SPeter Zijlstra 
222277866a0SPeter Zijlstra 		if (atomic_long_inc_return(&nfss->writeback) >
22389a09141SPeter Zijlstra 				NFS_CONGESTION_ON_THRESH)
22489a09141SPeter Zijlstra 			set_bdi_congested(&nfss->backing_dev_info, WRITE);
22589a09141SPeter Zijlstra 	}
2265a6d41b3STrond Myklebust 	return ret;
22789a09141SPeter Zijlstra }
22889a09141SPeter Zijlstra 
22989a09141SPeter Zijlstra static void nfs_end_page_writeback(struct page *page)
23089a09141SPeter Zijlstra {
23189a09141SPeter Zijlstra 	struct inode *inode = page->mapping->host;
23289a09141SPeter Zijlstra 	struct nfs_server *nfss = NFS_SERVER(inode);
23389a09141SPeter Zijlstra 
23489a09141SPeter Zijlstra 	end_page_writeback(page);
235c4dc4beeSPeter Zijlstra 	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
23689a09141SPeter Zijlstra 		clear_bdi_congested(&nfss->backing_dev_info, WRITE);
23789a09141SPeter Zijlstra }
23889a09141SPeter Zijlstra 
23989a09141SPeter Zijlstra /*
240e261f51fSTrond Myklebust  * Find an associated nfs write request, and prepare to flush it out
2419cccef95STrond Myklebust  * May return an error if the user signalled nfs_wait_on_request().
242e261f51fSTrond Myklebust  */
243c63c7b05STrond Myklebust static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
244c63c7b05STrond Myklebust 				struct page *page)
245e261f51fSTrond Myklebust {
246587142f8STrond Myklebust 	struct inode *inode = page->mapping->host;
247e261f51fSTrond Myklebust 	struct nfs_page *req;
248e261f51fSTrond Myklebust 	int ret;
249e261f51fSTrond Myklebust 
250587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
251e261f51fSTrond Myklebust 	for(;;) {
252e261f51fSTrond Myklebust 		req = nfs_page_find_request_locked(page);
253e261f51fSTrond Myklebust 		if (req == NULL) {
254587142f8STrond Myklebust 			spin_unlock(&inode->i_lock);
2559cccef95STrond Myklebust 			return 0;
256e261f51fSTrond Myklebust 		}
257acee478aSTrond Myklebust 		if (nfs_set_page_tag_locked(req))
258e261f51fSTrond Myklebust 			break;
259e261f51fSTrond Myklebust 		/* Note: If we hold the page lock, as is the case in nfs_writepage,
260acee478aSTrond Myklebust 		 *	 then the call to nfs_set_page_tag_locked() will always
261e261f51fSTrond Myklebust 		 *	 succeed provided that someone hasn't already marked the
262e261f51fSTrond Myklebust 		 *	 request as dirty (in which case we don't care).
263e261f51fSTrond Myklebust 		 */
264587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
265e261f51fSTrond Myklebust 		ret = nfs_wait_on_request(req);
266e261f51fSTrond Myklebust 		nfs_release_request(req);
267e261f51fSTrond Myklebust 		if (ret != 0)
268e261f51fSTrond Myklebust 			return ret;
269587142f8STrond Myklebust 		spin_lock(&inode->i_lock);
270e261f51fSTrond Myklebust 	}
271612c9384STrond Myklebust 	if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
272612c9384STrond Myklebust 		/* This request is marked for commit */
273587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
274acee478aSTrond Myklebust 		nfs_clear_page_tag_locked(req);
275c63c7b05STrond Myklebust 		nfs_pageio_complete(pgio);
2769cccef95STrond Myklebust 		return 0;
277612c9384STrond Myklebust 	}
278c63c7b05STrond Myklebust 	if (nfs_set_page_writeback(page) != 0) {
279587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
280c63c7b05STrond Myklebust 		BUG();
281c63c7b05STrond Myklebust 	}
282587142f8STrond Myklebust 	spin_unlock(&inode->i_lock);
283f8512ad0SFred Isaman 	if (!nfs_pageio_add_request(pgio, req)) {
284f8512ad0SFred Isaman 		nfs_redirty_request(req);
285f8512ad0SFred Isaman 		return pgio->pg_error;
286f8512ad0SFred Isaman 	}
2879cccef95STrond Myklebust 	return 0;
288e261f51fSTrond Myklebust }
289e261f51fSTrond Myklebust 
290f758c885STrond Myklebust static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
291f758c885STrond Myklebust {
292f758c885STrond Myklebust 	struct inode *inode = page->mapping->host;
293f758c885STrond Myklebust 
294f758c885STrond Myklebust 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
295f758c885STrond Myklebust 	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
296f758c885STrond Myklebust 
297f758c885STrond Myklebust 	nfs_pageio_cond_complete(pgio, page->index);
298f758c885STrond Myklebust 	return nfs_page_async_flush(pgio, page);
299f758c885STrond Myklebust }
300f758c885STrond Myklebust 
301e261f51fSTrond Myklebust /*
3021da177e4SLinus Torvalds  * Write an mmapped page to the server.
3031da177e4SLinus Torvalds  */
3044d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
3051da177e4SLinus Torvalds {
306f758c885STrond Myklebust 	struct nfs_pageio_descriptor pgio;
307e261f51fSTrond Myklebust 	int err;
3081da177e4SLinus Torvalds 
309f758c885STrond Myklebust 	nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
310f758c885STrond Myklebust 	err = nfs_do_writepage(page, wbc, &pgio);
311f758c885STrond Myklebust 	nfs_pageio_complete(&pgio);
312f758c885STrond Myklebust 	if (err < 0)
3134d770ccfSTrond Myklebust 		return err;
314f758c885STrond Myklebust 	if (pgio.pg_error < 0)
315f758c885STrond Myklebust 		return pgio.pg_error;
316f758c885STrond Myklebust 	return 0;
3174d770ccfSTrond Myklebust }
3184d770ccfSTrond Myklebust 
3194d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc)
3204d770ccfSTrond Myklebust {
321f758c885STrond Myklebust 	int ret;
3224d770ccfSTrond Myklebust 
323f758c885STrond Myklebust 	ret = nfs_writepage_locked(page, wbc);
3241da177e4SLinus Torvalds 	unlock_page(page);
325f758c885STrond Myklebust 	return ret;
326f758c885STrond Myklebust }
327f758c885STrond Myklebust 
328f758c885STrond Myklebust static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
329f758c885STrond Myklebust {
330f758c885STrond Myklebust 	int ret;
331f758c885STrond Myklebust 
332f758c885STrond Myklebust 	ret = nfs_do_writepage(page, wbc, data);
333f758c885STrond Myklebust 	unlock_page(page);
334f758c885STrond Myklebust 	return ret;
3351da177e4SLinus Torvalds }
3361da177e4SLinus Torvalds 
3371da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
3381da177e4SLinus Torvalds {
3391da177e4SLinus Torvalds 	struct inode *inode = mapping->host;
340c63c7b05STrond Myklebust 	struct nfs_pageio_descriptor pgio;
3411da177e4SLinus Torvalds 	int err;
3421da177e4SLinus Torvalds 
34391d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
34491d5b470SChuck Lever 
345c63c7b05STrond Myklebust 	nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
346f758c885STrond Myklebust 	err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
347c63c7b05STrond Myklebust 	nfs_pageio_complete(&pgio);
348f758c885STrond Myklebust 	if (err < 0)
3491da177e4SLinus Torvalds 		return err;
350f758c885STrond Myklebust 	if (pgio.pg_error < 0)
351c63c7b05STrond Myklebust 		return pgio.pg_error;
352c63c7b05STrond Myklebust 	return 0;
3531da177e4SLinus Torvalds }
3541da177e4SLinus Torvalds 
3551da177e4SLinus Torvalds /*
3561da177e4SLinus Torvalds  * Insert a write request into an inode
3571da177e4SLinus Torvalds  */
35827852596SNick Piggin static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
3591da177e4SLinus Torvalds {
3601da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
3611da177e4SLinus Torvalds 	int error;
3621da177e4SLinus Torvalds 
3631da177e4SLinus Torvalds 	error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
36427852596SNick Piggin 	BUG_ON(error);
3651da177e4SLinus Torvalds 	if (!nfsi->npages) {
3661da177e4SLinus Torvalds 		igrab(inode);
3671da177e4SLinus Torvalds 		if (nfs_have_delegation(inode, FMODE_WRITE))
3681da177e4SLinus Torvalds 			nfsi->change_attr++;
3691da177e4SLinus Torvalds 	}
370deb7d638STrond Myklebust 	SetPagePrivate(req->wb_page);
371277459d2STrond Myklebust 	set_page_private(req->wb_page, (unsigned long)req);
3721da177e4SLinus Torvalds 	nfsi->npages++;
373c03b4024STrond Myklebust 	kref_get(&req->wb_kref);
37427852596SNick Piggin 	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
37527852596SNick Piggin 				NFS_PAGE_TAG_LOCKED);
3761da177e4SLinus Torvalds }
3771da177e4SLinus Torvalds 
3781da177e4SLinus Torvalds /*
37989a09141SPeter Zijlstra  * Remove a write request from an inode
3801da177e4SLinus Torvalds  */
3811da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req)
3821da177e4SLinus Torvalds {
38388be9f99STrond Myklebust 	struct inode *inode = req->wb_context->path.dentry->d_inode;
3841da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
3851da177e4SLinus Torvalds 
3861da177e4SLinus Torvalds 	BUG_ON (!NFS_WBACK_BUSY(req));
3871da177e4SLinus Torvalds 
388587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
389277459d2STrond Myklebust 	set_page_private(req->wb_page, 0);
390deb7d638STrond Myklebust 	ClearPagePrivate(req->wb_page);
3911da177e4SLinus Torvalds 	radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
3921da177e4SLinus Torvalds 	nfsi->npages--;
3931da177e4SLinus Torvalds 	if (!nfsi->npages) {
394587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
3951da177e4SLinus Torvalds 		iput(inode);
3961da177e4SLinus Torvalds 	} else
397587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
3981da177e4SLinus Torvalds 	nfs_clear_request(req);
3991da177e4SLinus Torvalds 	nfs_release_request(req);
4001da177e4SLinus Torvalds }
4011da177e4SLinus Torvalds 
40261822ab5STrond Myklebust static void
4036d884e8fSFred nfs_mark_request_dirty(struct nfs_page *req)
40461822ab5STrond Myklebust {
40561822ab5STrond Myklebust 	__set_page_dirty_nobuffers(req->wb_page);
40661822ab5STrond Myklebust }
40761822ab5STrond Myklebust 
4081da177e4SLinus Torvalds /*
4091da177e4SLinus Torvalds  * Check if a request is dirty
4101da177e4SLinus Torvalds  */
4111da177e4SLinus Torvalds static inline int
4121da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req)
4131da177e4SLinus Torvalds {
4145a6d41b3STrond Myklebust 	struct page *page = req->wb_page;
4155a6d41b3STrond Myklebust 
416612c9384STrond Myklebust 	if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags))
4175a6d41b3STrond Myklebust 		return 0;
41838def50fSFred Isaman 	return !PageWriteback(page);
4191da177e4SLinus Torvalds }
4201da177e4SLinus Torvalds 
4211da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
4221da177e4SLinus Torvalds /*
4231da177e4SLinus Torvalds  * Add a request to the inode's commit list.
4241da177e4SLinus Torvalds  */
4251da177e4SLinus Torvalds static void
4261da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req)
4271da177e4SLinus Torvalds {
42888be9f99STrond Myklebust 	struct inode *inode = req->wb_context->path.dentry->d_inode;
4291da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4301da177e4SLinus Torvalds 
431587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
4321da177e4SLinus Torvalds 	nfsi->ncommit++;
433612c9384STrond Myklebust 	set_bit(PG_NEED_COMMIT, &(req)->wb_flags);
4345c369683STrond Myklebust 	radix_tree_tag_set(&nfsi->nfs_page_tree,
4355c369683STrond Myklebust 			req->wb_index,
4365c369683STrond Myklebust 			NFS_PAGE_TAG_COMMIT);
437587142f8STrond Myklebust 	spin_unlock(&inode->i_lock);
438fd39fc85SChristoph Lameter 	inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
439c9e51e41SPeter Zijlstra 	inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
440a1803044STrond Myklebust 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
4411da177e4SLinus Torvalds }
4428e821cadSTrond Myklebust 
4438e821cadSTrond Myklebust static inline
4448e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data)
4458e821cadSTrond Myklebust {
4468e821cadSTrond Myklebust 	return data->verf.committed != NFS_FILE_SYNC;
4478e821cadSTrond Myklebust }
4488e821cadSTrond Myklebust 
4498e821cadSTrond Myklebust static inline
4508e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req)
4518e821cadSTrond Myklebust {
452612c9384STrond Myklebust 	if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
4538e821cadSTrond Myklebust 		nfs_mark_request_commit(req);
4548e821cadSTrond Myklebust 		return 1;
4558e821cadSTrond Myklebust 	}
4568e821cadSTrond Myklebust 	if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
4576d884e8fSFred 		nfs_mark_request_dirty(req);
4588e821cadSTrond Myklebust 		return 1;
4598e821cadSTrond Myklebust 	}
4608e821cadSTrond Myklebust 	return 0;
4618e821cadSTrond Myklebust }
4628e821cadSTrond Myklebust #else
4638e821cadSTrond Myklebust static inline void
4648e821cadSTrond Myklebust nfs_mark_request_commit(struct nfs_page *req)
4658e821cadSTrond Myklebust {
4668e821cadSTrond Myklebust }
4678e821cadSTrond Myklebust 
4688e821cadSTrond Myklebust static inline
4698e821cadSTrond Myklebust int nfs_write_need_commit(struct nfs_write_data *data)
4708e821cadSTrond Myklebust {
4718e821cadSTrond Myklebust 	return 0;
4728e821cadSTrond Myklebust }
4738e821cadSTrond Myklebust 
4748e821cadSTrond Myklebust static inline
4758e821cadSTrond Myklebust int nfs_reschedule_unstable_write(struct nfs_page *req)
4768e821cadSTrond Myklebust {
4778e821cadSTrond Myklebust 	return 0;
4788e821cadSTrond Myklebust }
4791da177e4SLinus Torvalds #endif
4801da177e4SLinus Torvalds 
4811da177e4SLinus Torvalds /*
4821da177e4SLinus Torvalds  * Wait for a request to complete.
4831da177e4SLinus Torvalds  *
484150030b7SMatthew Wilcox  * Interruptible by fatal signals only.
4851da177e4SLinus Torvalds  */
486ca52fec1STrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
4871da177e4SLinus Torvalds {
4881da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4891da177e4SLinus Torvalds 	struct nfs_page *req;
490ca52fec1STrond Myklebust 	pgoff_t idx_end, next;
4911da177e4SLinus Torvalds 	unsigned int		res = 0;
4921da177e4SLinus Torvalds 	int			error;
4931da177e4SLinus Torvalds 
4941da177e4SLinus Torvalds 	if (npages == 0)
4951da177e4SLinus Torvalds 		idx_end = ~0;
4961da177e4SLinus Torvalds 	else
4971da177e4SLinus Torvalds 		idx_end = idx_start + npages - 1;
4981da177e4SLinus Torvalds 
4991da177e4SLinus Torvalds 	next = idx_start;
5009fd367f0STrond Myklebust 	while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) {
5011da177e4SLinus Torvalds 		if (req->wb_index > idx_end)
5021da177e4SLinus Torvalds 			break;
5031da177e4SLinus Torvalds 
5041da177e4SLinus Torvalds 		next = req->wb_index + 1;
505c6a556b8STrond Myklebust 		BUG_ON(!NFS_WBACK_BUSY(req));
5061da177e4SLinus Torvalds 
507c03b4024STrond Myklebust 		kref_get(&req->wb_kref);
508587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
5091da177e4SLinus Torvalds 		error = nfs_wait_on_request(req);
5101da177e4SLinus Torvalds 		nfs_release_request(req);
511587142f8STrond Myklebust 		spin_lock(&inode->i_lock);
5121da177e4SLinus Torvalds 		if (error < 0)
5131da177e4SLinus Torvalds 			return error;
5141da177e4SLinus Torvalds 		res++;
5151da177e4SLinus Torvalds 	}
5161da177e4SLinus Torvalds 	return res;
5171da177e4SLinus Torvalds }
5181da177e4SLinus Torvalds 
51983715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head)
52083715ad5STrond Myklebust {
52183715ad5STrond Myklebust 	struct nfs_page *req;
52283715ad5STrond Myklebust 
52383715ad5STrond Myklebust 	while(!list_empty(head)) {
52483715ad5STrond Myklebust 		req = nfs_list_entry(head->next);
525b6dff26aSTrond Myklebust 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
526c9e51e41SPeter Zijlstra 		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
527c9e51e41SPeter Zijlstra 				BDI_RECLAIMABLE);
52883715ad5STrond Myklebust 		nfs_list_remove_request(req);
529612c9384STrond Myklebust 		clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
53083715ad5STrond Myklebust 		nfs_inode_remove_request(req);
531b6dff26aSTrond Myklebust 		nfs_unlock_request(req);
53283715ad5STrond Myklebust 	}
53383715ad5STrond Myklebust }
53483715ad5STrond Myklebust 
5351da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
5361da177e4SLinus Torvalds /*
5371da177e4SLinus Torvalds  * nfs_scan_commit - Scan an inode for commit requests
5381da177e4SLinus Torvalds  * @inode: NFS inode to scan
5391da177e4SLinus Torvalds  * @dst: destination list
5401da177e4SLinus Torvalds  * @idx_start: lower bound of page->index to scan.
5411da177e4SLinus Torvalds  * @npages: idx_start + npages sets the upper bound to scan.
5421da177e4SLinus Torvalds  *
5431da177e4SLinus Torvalds  * Moves requests from the inode's 'commit' request list.
5441da177e4SLinus Torvalds  * The requests are *not* checked to ensure that they form a contiguous set.
5451da177e4SLinus Torvalds  */
5461da177e4SLinus Torvalds static int
547ca52fec1STrond Myklebust nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
5481da177e4SLinus Torvalds {
5491da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
5503da28eb1STrond Myklebust 	int res = 0;
5513da28eb1STrond Myklebust 
5523da28eb1STrond Myklebust 	if (nfsi->ncommit != 0) {
5535c369683STrond Myklebust 		res = nfs_scan_list(nfsi, dst, idx_start, npages,
5545c369683STrond Myklebust 				NFS_PAGE_TAG_COMMIT);
5551da177e4SLinus Torvalds 		nfsi->ncommit -= res;
5563da28eb1STrond Myklebust 	}
5571da177e4SLinus Torvalds 	return res;
5581da177e4SLinus Torvalds }
559c42de9ddSTrond Myklebust #else
560ca52fec1STrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
561c42de9ddSTrond Myklebust {
562c42de9ddSTrond Myklebust 	return 0;
563c42de9ddSTrond Myklebust }
5641da177e4SLinus Torvalds #endif
5651da177e4SLinus Torvalds 
5661da177e4SLinus Torvalds /*
5671da177e4SLinus Torvalds  * Try to update any existing write request, or create one if there is none.
5681da177e4SLinus Torvalds  * In order to match, the request's credentials must match those of
5691da177e4SLinus Torvalds  * the calling process.
5701da177e4SLinus Torvalds  *
5711da177e4SLinus Torvalds  * Note: Should always be called with the Page Lock held!
5721da177e4SLinus Torvalds  */
5731da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
574e21195a7STrond Myklebust 		struct page *page, unsigned int offset, unsigned int bytes)
5751da177e4SLinus Torvalds {
57689a09141SPeter Zijlstra 	struct address_space *mapping = page->mapping;
57789a09141SPeter Zijlstra 	struct inode *inode = mapping->host;
5781da177e4SLinus Torvalds 	struct nfs_page		*req, *new = NULL;
579ca52fec1STrond Myklebust 	pgoff_t		rqend, end;
5801da177e4SLinus Torvalds 
5811da177e4SLinus Torvalds 	end = offset + bytes;
5821da177e4SLinus Torvalds 
5831da177e4SLinus Torvalds 	for (;;) {
5841da177e4SLinus Torvalds 		/* Loop over all inode entries and see if we find
5851da177e4SLinus Torvalds 		 * A request for the page we wish to update
5861da177e4SLinus Torvalds 		 */
58727852596SNick Piggin 		if (new) {
58827852596SNick Piggin 			if (radix_tree_preload(GFP_NOFS)) {
58927852596SNick Piggin 				nfs_release_request(new);
59027852596SNick Piggin 				return ERR_PTR(-ENOMEM);
59127852596SNick Piggin 			}
59227852596SNick Piggin 		}
59327852596SNick Piggin 
594587142f8STrond Myklebust 		spin_lock(&inode->i_lock);
595277459d2STrond Myklebust 		req = nfs_page_find_request_locked(page);
5961da177e4SLinus Torvalds 		if (req) {
597acee478aSTrond Myklebust 			if (!nfs_set_page_tag_locked(req)) {
5981da177e4SLinus Torvalds 				int error;
599277459d2STrond Myklebust 
600587142f8STrond Myklebust 				spin_unlock(&inode->i_lock);
6011da177e4SLinus Torvalds 				error = nfs_wait_on_request(req);
6021da177e4SLinus Torvalds 				nfs_release_request(req);
6031dd594b2SNeil Brown 				if (error < 0) {
60427852596SNick Piggin 					if (new) {
60527852596SNick Piggin 						radix_tree_preload_end();
6061dd594b2SNeil Brown 						nfs_release_request(new);
60727852596SNick Piggin 					}
6081da177e4SLinus Torvalds 					return ERR_PTR(error);
6091dd594b2SNeil Brown 				}
6101da177e4SLinus Torvalds 				continue;
6111da177e4SLinus Torvalds 			}
612587142f8STrond Myklebust 			spin_unlock(&inode->i_lock);
61327852596SNick Piggin 			if (new) {
61427852596SNick Piggin 				radix_tree_preload_end();
6151da177e4SLinus Torvalds 				nfs_release_request(new);
61627852596SNick Piggin 			}
6171da177e4SLinus Torvalds 			break;
6181da177e4SLinus Torvalds 		}
6191da177e4SLinus Torvalds 
6201da177e4SLinus Torvalds 		if (new) {
6211da177e4SLinus Torvalds 			nfs_lock_request_dontget(new);
62227852596SNick Piggin 			nfs_inode_add_request(inode, new);
623587142f8STrond Myklebust 			spin_unlock(&inode->i_lock);
62427852596SNick Piggin 			radix_tree_preload_end();
62561e930a9STrond Myklebust 			req = new;
62661e930a9STrond Myklebust 			goto zero_page;
6271da177e4SLinus Torvalds 		}
628587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
6291da177e4SLinus Torvalds 
6301da177e4SLinus Torvalds 		new = nfs_create_request(ctx, inode, page, offset, bytes);
6311da177e4SLinus Torvalds 		if (IS_ERR(new))
6321da177e4SLinus Torvalds 			return new;
6331da177e4SLinus Torvalds 	}
6341da177e4SLinus Torvalds 
6351da177e4SLinus Torvalds 	/* We have a request for our page.
6361da177e4SLinus Torvalds 	 * If the creds don't match, or the
6371da177e4SLinus Torvalds 	 * page addresses don't match,
6381da177e4SLinus Torvalds 	 * tell the caller to wait on the conflicting
6391da177e4SLinus Torvalds 	 * request.
6401da177e4SLinus Torvalds 	 */
6411da177e4SLinus Torvalds 	rqend = req->wb_offset + req->wb_bytes;
6421da177e4SLinus Torvalds 	if (req->wb_context != ctx
6431da177e4SLinus Torvalds 	    || req->wb_page != page
6441da177e4SLinus Torvalds 	    || !nfs_dirty_request(req)
6451da177e4SLinus Torvalds 	    || offset > rqend || end < req->wb_offset) {
646acee478aSTrond Myklebust 		nfs_clear_page_tag_locked(req);
6471da177e4SLinus Torvalds 		return ERR_PTR(-EBUSY);
6481da177e4SLinus Torvalds 	}
6491da177e4SLinus Torvalds 
6501da177e4SLinus Torvalds 	/* Okay, the request matches. Update the region */
6511da177e4SLinus Torvalds 	if (offset < req->wb_offset) {
6521da177e4SLinus Torvalds 		req->wb_offset = offset;
6531da177e4SLinus Torvalds 		req->wb_pgbase = offset;
65461e930a9STrond Myklebust 		req->wb_bytes = max(end, rqend) - req->wb_offset;
65561e930a9STrond Myklebust 		goto zero_page;
6561da177e4SLinus Torvalds 	}
6571da177e4SLinus Torvalds 
6581da177e4SLinus Torvalds 	if (end > rqend)
6591da177e4SLinus Torvalds 		req->wb_bytes = end - req->wb_offset;
6601da177e4SLinus Torvalds 
6611da177e4SLinus Torvalds 	return req;
66261e930a9STrond Myklebust zero_page:
66361e930a9STrond Myklebust 	/* If this page might potentially be marked as up to date,
66461e930a9STrond Myklebust 	 * then we need to zero any uninitalised data. */
66561e930a9STrond Myklebust 	if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE
66661e930a9STrond Myklebust 			&& !PageUptodate(req->wb_page))
667eebd2aa3SChristoph Lameter 		zero_user_segment(req->wb_page, req->wb_bytes, PAGE_CACHE_SIZE);
66861e930a9STrond Myklebust 	return req;
6691da177e4SLinus Torvalds }
6701da177e4SLinus Torvalds 
6711da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page)
6721da177e4SLinus Torvalds {
673cd3758e3STrond Myklebust 	struct nfs_open_context *ctx = nfs_file_open_context(file);
6741da177e4SLinus Torvalds 	struct nfs_page	*req;
6751a54533eSTrond Myklebust 	int do_flush, status;
6761da177e4SLinus Torvalds 	/*
6771da177e4SLinus Torvalds 	 * Look for a request corresponding to this page. If there
6781da177e4SLinus Torvalds 	 * is one, and it belongs to another file, we flush it out
6791da177e4SLinus Torvalds 	 * before we try to copy anything into the page. Do this
6801da177e4SLinus Torvalds 	 * due to the lack of an ACCESS-type call in NFSv2.
6811da177e4SLinus Torvalds 	 * Also do the same if we find a request from an existing
6821da177e4SLinus Torvalds 	 * dropped page.
6831da177e4SLinus Torvalds 	 */
6841a54533eSTrond Myklebust 	do {
685277459d2STrond Myklebust 		req = nfs_page_find_request(page);
6861a54533eSTrond Myklebust 		if (req == NULL)
6871a54533eSTrond Myklebust 			return 0;
6881a54533eSTrond Myklebust 		do_flush = req->wb_page != page || req->wb_context != ctx
689e261f51fSTrond Myklebust 			|| !nfs_dirty_request(req);
6901da177e4SLinus Torvalds 		nfs_release_request(req);
6911a54533eSTrond Myklebust 		if (!do_flush)
6921a54533eSTrond Myklebust 			return 0;
693277459d2STrond Myklebust 		status = nfs_wb_page(page->mapping->host, page);
6941a54533eSTrond Myklebust 	} while (status == 0);
6951a54533eSTrond Myklebust 	return status;
6961da177e4SLinus Torvalds }
6971da177e4SLinus Torvalds 
6981da177e4SLinus Torvalds /*
6995d47a356STrond Myklebust  * If the page cache is marked as unsafe or invalid, then we can't rely on
7005d47a356STrond Myklebust  * the PageUptodate() flag. In this case, we will need to turn off
7015d47a356STrond Myklebust  * write optimisations that depend on the page contents being correct.
7025d47a356STrond Myklebust  */
7035d47a356STrond Myklebust static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
7045d47a356STrond Myklebust {
7055d47a356STrond Myklebust 	return PageUptodate(page) &&
7065d47a356STrond Myklebust 		!(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
7075d47a356STrond Myklebust }
7085d47a356STrond Myklebust 
7095d47a356STrond Myklebust /*
7101da177e4SLinus Torvalds  * Update and possibly write a cached page of an NFS file.
7111da177e4SLinus Torvalds  *
7121da177e4SLinus Torvalds  * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
7131da177e4SLinus Torvalds  * things with a page scheduled for an RPC call (e.g. invalidate it).
7141da177e4SLinus Torvalds  */
7151da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page,
7161da177e4SLinus Torvalds 		unsigned int offset, unsigned int count)
7171da177e4SLinus Torvalds {
718cd3758e3STrond Myklebust 	struct nfs_open_context *ctx = nfs_file_open_context(file);
7191da177e4SLinus Torvalds 	struct inode	*inode = page->mapping->host;
7201da177e4SLinus Torvalds 	int		status = 0;
7211da177e4SLinus Torvalds 
72291d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
72391d5b470SChuck Lever 
7241da177e4SLinus Torvalds 	dprintk("NFS:      nfs_updatepage(%s/%s %d@%Ld)\n",
72501cce933SJosef "Jeff" Sipek 		file->f_path.dentry->d_parent->d_name.name,
72601cce933SJosef "Jeff" Sipek 		file->f_path.dentry->d_name.name, count,
7270bbacc40SChuck Lever 		(long long)(page_offset(page) +offset));
7281da177e4SLinus Torvalds 
7291da177e4SLinus Torvalds 	/* If we're not using byte range locks, and we know the page
7305d47a356STrond Myklebust 	 * is up to date, it may be more efficient to extend the write
7315d47a356STrond Myklebust 	 * to cover the entire page in order to avoid fragmentation
7325d47a356STrond Myklebust 	 * inefficiencies.
7331da177e4SLinus Torvalds 	 */
7345d47a356STrond Myklebust 	if (nfs_write_pageuptodate(page, inode) &&
7355d47a356STrond Myklebust 			inode->i_flock == NULL &&
7364b5621f6STrond Myklebust 			!(file->f_flags & O_SYNC)) {
73749a70f27STrond Myklebust 		count = max(count + offset, nfs_page_length(page));
7381da177e4SLinus Torvalds 		offset = 0;
7391da177e4SLinus Torvalds 	}
7401da177e4SLinus Torvalds 
741e21195a7STrond Myklebust 	status = nfs_writepage_setup(ctx, page, offset, count);
74203fa9e84STrond Myklebust 	if (status < 0)
74303fa9e84STrond Myklebust 		nfs_set_pageerror(page);
74403fa9e84STrond Myklebust 	else
745e261f51fSTrond Myklebust 		__set_page_dirty_nobuffers(page);
7461da177e4SLinus Torvalds 
7471da177e4SLinus Torvalds         dprintk("NFS:      nfs_updatepage returns %d (isize %Ld)\n",
7481da177e4SLinus Torvalds 			status, (long long)i_size_read(inode));
7491da177e4SLinus Torvalds 	return status;
7501da177e4SLinus Torvalds }
7511da177e4SLinus Torvalds 
7521da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req)
7531da177e4SLinus Torvalds {
7548e821cadSTrond Myklebust 
75544dd151dSTrond Myklebust 	if (PageError(req->wb_page)) {
75644dd151dSTrond Myklebust 		nfs_end_page_writeback(req->wb_page);
75744dd151dSTrond Myklebust 		nfs_inode_remove_request(req);
75844dd151dSTrond Myklebust 	} else if (!nfs_reschedule_unstable_write(req)) {
75944dd151dSTrond Myklebust 		/* Set the PG_uptodate flag */
76044dd151dSTrond Myklebust 		nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes);
76189a09141SPeter Zijlstra 		nfs_end_page_writeback(req->wb_page);
7621da177e4SLinus Torvalds 		nfs_inode_remove_request(req);
7638e821cadSTrond Myklebust 	} else
7648e821cadSTrond Myklebust 		nfs_end_page_writeback(req->wb_page);
7659fd367f0STrond Myklebust 	nfs_clear_page_tag_locked(req);
7661da177e4SLinus Torvalds }
7671da177e4SLinus Torvalds 
7683ff7576dSTrond Myklebust static int flush_task_priority(int how)
7691da177e4SLinus Torvalds {
7701da177e4SLinus Torvalds 	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
7711da177e4SLinus Torvalds 		case FLUSH_HIGHPRI:
7721da177e4SLinus Torvalds 			return RPC_PRIORITY_HIGH;
7731da177e4SLinus Torvalds 		case FLUSH_LOWPRI:
7741da177e4SLinus Torvalds 			return RPC_PRIORITY_LOW;
7751da177e4SLinus Torvalds 	}
7761da177e4SLinus Torvalds 	return RPC_PRIORITY_NORMAL;
7771da177e4SLinus Torvalds }
7781da177e4SLinus Torvalds 
7791da177e4SLinus Torvalds /*
7801da177e4SLinus Torvalds  * Set up the argument/result storage required for the RPC call.
7811da177e4SLinus Torvalds  */
782dbae4c73STrond Myklebust static int nfs_write_rpcsetup(struct nfs_page *req,
7831da177e4SLinus Torvalds 		struct nfs_write_data *data,
784788e7a89STrond Myklebust 		const struct rpc_call_ops *call_ops,
7851da177e4SLinus Torvalds 		unsigned int count, unsigned int offset,
7861da177e4SLinus Torvalds 		int how)
7871da177e4SLinus Torvalds {
78884115e1cSTrond Myklebust 	struct inode *inode = req->wb_context->path.dentry->d_inode;
78984115e1cSTrond Myklebust 	int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
7903ff7576dSTrond Myklebust 	int priority = flush_task_priority(how);
79107737691STrond Myklebust 	struct rpc_task *task;
792bdc7f021STrond Myklebust 	struct rpc_message msg = {
793bdc7f021STrond Myklebust 		.rpc_argp = &data->args,
794bdc7f021STrond Myklebust 		.rpc_resp = &data->res,
795bdc7f021STrond Myklebust 		.rpc_cred = req->wb_context->cred,
796bdc7f021STrond Myklebust 	};
79784115e1cSTrond Myklebust 	struct rpc_task_setup task_setup_data = {
79884115e1cSTrond Myklebust 		.rpc_client = NFS_CLIENT(inode),
79907737691STrond Myklebust 		.task = &data->task,
800bdc7f021STrond Myklebust 		.rpc_message = &msg,
80184115e1cSTrond Myklebust 		.callback_ops = call_ops,
80284115e1cSTrond Myklebust 		.callback_data = data,
803101070caSTrond Myklebust 		.workqueue = nfsiod_workqueue,
80484115e1cSTrond Myklebust 		.flags = flags,
8053ff7576dSTrond Myklebust 		.priority = priority,
80684115e1cSTrond Myklebust 	};
8071da177e4SLinus Torvalds 
8081da177e4SLinus Torvalds 	/* Set up the RPC argument and reply structs
8091da177e4SLinus Torvalds 	 * NB: take care not to mess about with data->commit et al. */
8101da177e4SLinus Torvalds 
8111da177e4SLinus Torvalds 	data->req = req;
81288be9f99STrond Myklebust 	data->inode = inode = req->wb_context->path.dentry->d_inode;
813bdc7f021STrond Myklebust 	data->cred = msg.rpc_cred;
8141da177e4SLinus Torvalds 
8151da177e4SLinus Torvalds 	data->args.fh     = NFS_FH(inode);
8161da177e4SLinus Torvalds 	data->args.offset = req_offset(req) + offset;
8171da177e4SLinus Torvalds 	data->args.pgbase = req->wb_pgbase + offset;
8181da177e4SLinus Torvalds 	data->args.pages  = data->pagevec;
8191da177e4SLinus Torvalds 	data->args.count  = count;
820383ba719STrond Myklebust 	data->args.context = get_nfs_open_context(req->wb_context);
821bdc7f021STrond Myklebust 	data->args.stable  = NFS_UNSTABLE;
822bdc7f021STrond Myklebust 	if (how & FLUSH_STABLE) {
823bdc7f021STrond Myklebust 		data->args.stable = NFS_DATA_SYNC;
824bdc7f021STrond Myklebust 		if (!NFS_I(inode)->ncommit)
825bdc7f021STrond Myklebust 			data->args.stable = NFS_FILE_SYNC;
826bdc7f021STrond Myklebust 	}
8271da177e4SLinus Torvalds 
8281da177e4SLinus Torvalds 	data->res.fattr   = &data->fattr;
8291da177e4SLinus Torvalds 	data->res.count   = count;
8301da177e4SLinus Torvalds 	data->res.verf    = &data->verf;
8310e574af1STrond Myklebust 	nfs_fattr_init(&data->fattr);
8321da177e4SLinus Torvalds 
833788e7a89STrond Myklebust 	/* Set up the initial task struct.  */
834bdc7f021STrond Myklebust 	NFS_PROTO(inode)->write_setup(data, &msg);
8351da177e4SLinus Torvalds 
836a3f565b1SChuck Lever 	dprintk("NFS: %5u initiated write call "
837a3f565b1SChuck Lever 		"(req %s/%Ld, %u bytes @ offset %Lu)\n",
8380bbacc40SChuck Lever 		data->task.tk_pid,
8391da177e4SLinus Torvalds 		inode->i_sb->s_id,
8401da177e4SLinus Torvalds 		(long long)NFS_FILEID(inode),
8411da177e4SLinus Torvalds 		count,
8421da177e4SLinus Torvalds 		(unsigned long long)data->args.offset);
8431da177e4SLinus Torvalds 
84407737691STrond Myklebust 	task = rpc_run_task(&task_setup_data);
845dbae4c73STrond Myklebust 	if (IS_ERR(task))
846dbae4c73STrond Myklebust 		return PTR_ERR(task);
84707737691STrond Myklebust 	rpc_put_task(task);
848dbae4c73STrond Myklebust 	return 0;
8491da177e4SLinus Torvalds }
8501da177e4SLinus Torvalds 
8516d884e8fSFred /* If a nfs_flush_* function fails, it should remove reqs from @head and
8526d884e8fSFred  * call this on each, which will prepare them to be retried on next
8536d884e8fSFred  * writeback using standard nfs.
8546d884e8fSFred  */
8556d884e8fSFred static void nfs_redirty_request(struct nfs_page *req)
8566d884e8fSFred {
8576d884e8fSFred 	nfs_mark_request_dirty(req);
8586d884e8fSFred 	nfs_end_page_writeback(req->wb_page);
8596d884e8fSFred 	nfs_clear_page_tag_locked(req);
8606d884e8fSFred }
8616d884e8fSFred 
8621da177e4SLinus Torvalds /*
8631da177e4SLinus Torvalds  * Generate multiple small requests to write out a single
8641da177e4SLinus Torvalds  * contiguous dirty area on one page.
8651da177e4SLinus Torvalds  */
8668d5658c9STrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
8671da177e4SLinus Torvalds {
8681da177e4SLinus Torvalds 	struct nfs_page *req = nfs_list_entry(head->next);
8691da177e4SLinus Torvalds 	struct page *page = req->wb_page;
8701da177e4SLinus Torvalds 	struct nfs_write_data *data;
871e9f7bee1STrond Myklebust 	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
872e9f7bee1STrond Myklebust 	unsigned int offset;
8731da177e4SLinus Torvalds 	int requests = 0;
874dbae4c73STrond Myklebust 	int ret = 0;
8751da177e4SLinus Torvalds 	LIST_HEAD(list);
8761da177e4SLinus Torvalds 
8771da177e4SLinus Torvalds 	nfs_list_remove_request(req);
8781da177e4SLinus Torvalds 
879bcb71bbaSTrond Myklebust 	nbytes = count;
880e9f7bee1STrond Myklebust 	do {
881e9f7bee1STrond Myklebust 		size_t len = min(nbytes, wsize);
882e9f7bee1STrond Myklebust 
8838d5658c9STrond Myklebust 		data = nfs_writedata_alloc(1);
8841da177e4SLinus Torvalds 		if (!data)
8851da177e4SLinus Torvalds 			goto out_bad;
8861da177e4SLinus Torvalds 		list_add(&data->pages, &list);
8871da177e4SLinus Torvalds 		requests++;
888e9f7bee1STrond Myklebust 		nbytes -= len;
889e9f7bee1STrond Myklebust 	} while (nbytes != 0);
8901da177e4SLinus Torvalds 	atomic_set(&req->wb_complete, requests);
8911da177e4SLinus Torvalds 
8921da177e4SLinus Torvalds 	ClearPageError(page);
8931da177e4SLinus Torvalds 	offset = 0;
894bcb71bbaSTrond Myklebust 	nbytes = count;
8951da177e4SLinus Torvalds 	do {
896dbae4c73STrond Myklebust 		int ret2;
897dbae4c73STrond Myklebust 
8981da177e4SLinus Torvalds 		data = list_entry(list.next, struct nfs_write_data, pages);
8991da177e4SLinus Torvalds 		list_del_init(&data->pages);
9001da177e4SLinus Torvalds 
9011da177e4SLinus Torvalds 		data->pagevec[0] = page;
9021da177e4SLinus Torvalds 
903bcb71bbaSTrond Myklebust 		if (nbytes < wsize)
904bcb71bbaSTrond Myklebust 			wsize = nbytes;
905dbae4c73STrond Myklebust 		ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
906788e7a89STrond Myklebust 				   wsize, offset, how);
907dbae4c73STrond Myklebust 		if (ret == 0)
908dbae4c73STrond Myklebust 			ret = ret2;
9091da177e4SLinus Torvalds 		offset += wsize;
9101da177e4SLinus Torvalds 		nbytes -= wsize;
9111da177e4SLinus Torvalds 	} while (nbytes != 0);
9121da177e4SLinus Torvalds 
913dbae4c73STrond Myklebust 	return ret;
9141da177e4SLinus Torvalds 
9151da177e4SLinus Torvalds out_bad:
9161da177e4SLinus Torvalds 	while (!list_empty(&list)) {
9171da177e4SLinus Torvalds 		data = list_entry(list.next, struct nfs_write_data, pages);
9181da177e4SLinus Torvalds 		list_del(&data->pages);
9198aca67f0STrond Myklebust 		nfs_writedata_release(data);
9201da177e4SLinus Torvalds 	}
92161822ab5STrond Myklebust 	nfs_redirty_request(req);
9221da177e4SLinus Torvalds 	return -ENOMEM;
9231da177e4SLinus Torvalds }
9241da177e4SLinus Torvalds 
9251da177e4SLinus Torvalds /*
9261da177e4SLinus Torvalds  * Create an RPC task for the given write request and kick it.
9271da177e4SLinus Torvalds  * The page must have been locked by the caller.
9281da177e4SLinus Torvalds  *
9291da177e4SLinus Torvalds  * It may happen that the page we're passed is not marked dirty.
9301da177e4SLinus Torvalds  * This is the case if nfs_updatepage detects a conflicting request
9311da177e4SLinus Torvalds  * that has been written but not committed.
9321da177e4SLinus Torvalds  */
9338d5658c9STrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
9341da177e4SLinus Torvalds {
9351da177e4SLinus Torvalds 	struct nfs_page		*req;
9361da177e4SLinus Torvalds 	struct page		**pages;
9371da177e4SLinus Torvalds 	struct nfs_write_data	*data;
9381da177e4SLinus Torvalds 
9398d5658c9STrond Myklebust 	data = nfs_writedata_alloc(npages);
9401da177e4SLinus Torvalds 	if (!data)
9411da177e4SLinus Torvalds 		goto out_bad;
9421da177e4SLinus Torvalds 
9431da177e4SLinus Torvalds 	pages = data->pagevec;
9441da177e4SLinus Torvalds 	while (!list_empty(head)) {
9451da177e4SLinus Torvalds 		req = nfs_list_entry(head->next);
9461da177e4SLinus Torvalds 		nfs_list_remove_request(req);
9471da177e4SLinus Torvalds 		nfs_list_add_request(req, &data->pages);
9481da177e4SLinus Torvalds 		ClearPageError(req->wb_page);
9491da177e4SLinus Torvalds 		*pages++ = req->wb_page;
9501da177e4SLinus Torvalds 	}
9511da177e4SLinus Torvalds 	req = nfs_list_entry(data->pages.next);
9521da177e4SLinus Torvalds 
9531da177e4SLinus Torvalds 	/* Set up the argument struct */
954dbae4c73STrond Myklebust 	return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
9551da177e4SLinus Torvalds  out_bad:
9561da177e4SLinus Torvalds 	while (!list_empty(head)) {
95710afec90STrond Myklebust 		req = nfs_list_entry(head->next);
9581da177e4SLinus Torvalds 		nfs_list_remove_request(req);
95961822ab5STrond Myklebust 		nfs_redirty_request(req);
9601da177e4SLinus Torvalds 	}
9611da177e4SLinus Torvalds 	return -ENOMEM;
9621da177e4SLinus Torvalds }
9631da177e4SLinus Torvalds 
964c63c7b05STrond Myklebust static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
965c63c7b05STrond Myklebust 				  struct inode *inode, int ioflags)
9661da177e4SLinus Torvalds {
967bf4285e7SChuck Lever 	size_t wsize = NFS_SERVER(inode)->wsize;
9681da177e4SLinus Torvalds 
969bcb71bbaSTrond Myklebust 	if (wsize < PAGE_CACHE_SIZE)
970c63c7b05STrond Myklebust 		nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
971bcb71bbaSTrond Myklebust 	else
972c63c7b05STrond Myklebust 		nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
9731da177e4SLinus Torvalds }
9741da177e4SLinus Torvalds 
9751da177e4SLinus Torvalds /*
9761da177e4SLinus Torvalds  * Handle a write reply that flushed part of a page.
9771da177e4SLinus Torvalds  */
978788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
9791da177e4SLinus Torvalds {
980788e7a89STrond Myklebust 	struct nfs_write_data	*data = calldata;
9811da177e4SLinus Torvalds 	struct nfs_page		*req = data->req;
9821da177e4SLinus Torvalds 
9831da177e4SLinus Torvalds 	dprintk("NFS: write (%s/%Ld %d@%Ld)",
98488be9f99STrond Myklebust 		req->wb_context->path.dentry->d_inode->i_sb->s_id,
98588be9f99STrond Myklebust 		(long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
9861da177e4SLinus Torvalds 		req->wb_bytes,
9871da177e4SLinus Torvalds 		(long long)req_offset(req));
9881da177e4SLinus Torvalds 
989c9d8f89dSTrond Myklebust 	nfs_writeback_done(task, data);
990c9d8f89dSTrond Myklebust }
991788e7a89STrond Myklebust 
992c9d8f89dSTrond Myklebust static void nfs_writeback_release_partial(void *calldata)
993c9d8f89dSTrond Myklebust {
994c9d8f89dSTrond Myklebust 	struct nfs_write_data	*data = calldata;
995c9d8f89dSTrond Myklebust 	struct nfs_page		*req = data->req;
996c9d8f89dSTrond Myklebust 	struct page		*page = req->wb_page;
997c9d8f89dSTrond Myklebust 	int status = data->task.tk_status;
998c9d8f89dSTrond Myklebust 
999c9d8f89dSTrond Myklebust 	if (status < 0) {
1000a301b777STrond Myklebust 		nfs_set_pageerror(page);
1001c9d8f89dSTrond Myklebust 		nfs_context_set_write_error(req->wb_context, status);
1002c9d8f89dSTrond Myklebust 		dprintk(", error = %d\n", status);
10038e821cadSTrond Myklebust 		goto out;
10048e821cadSTrond Myklebust 	}
10058e821cadSTrond Myklebust 
10068e821cadSTrond Myklebust 	if (nfs_write_need_commit(data)) {
1007587142f8STrond Myklebust 		struct inode *inode = page->mapping->host;
10088e821cadSTrond Myklebust 
1009587142f8STrond Myklebust 		spin_lock(&inode->i_lock);
10108e821cadSTrond Myklebust 		if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
10118e821cadSTrond Myklebust 			/* Do nothing we need to resend the writes */
10128e821cadSTrond Myklebust 		} else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
10131da177e4SLinus Torvalds 			memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
10141da177e4SLinus Torvalds 			dprintk(" defer commit\n");
10151da177e4SLinus Torvalds 		} else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
10168e821cadSTrond Myklebust 			set_bit(PG_NEED_RESCHED, &req->wb_flags);
10178e821cadSTrond Myklebust 			clear_bit(PG_NEED_COMMIT, &req->wb_flags);
10181da177e4SLinus Torvalds 			dprintk(" server reboot detected\n");
10191da177e4SLinus Torvalds 		}
1020587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
10211da177e4SLinus Torvalds 	} else
10221da177e4SLinus Torvalds 		dprintk(" OK\n");
10231da177e4SLinus Torvalds 
10248e821cadSTrond Myklebust out:
10251da177e4SLinus Torvalds 	if (atomic_dec_and_test(&req->wb_complete))
10261da177e4SLinus Torvalds 		nfs_writepage_release(req);
1027c9d8f89dSTrond Myklebust 	nfs_writedata_release(calldata);
10281da177e4SLinus Torvalds }
10291da177e4SLinus Torvalds 
1030788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = {
1031788e7a89STrond Myklebust 	.rpc_call_done = nfs_writeback_done_partial,
1032c9d8f89dSTrond Myklebust 	.rpc_release = nfs_writeback_release_partial,
1033788e7a89STrond Myklebust };
1034788e7a89STrond Myklebust 
10351da177e4SLinus Torvalds /*
10361da177e4SLinus Torvalds  * Handle a write reply that flushes a whole page.
10371da177e4SLinus Torvalds  *
10381da177e4SLinus Torvalds  * FIXME: There is an inherent race with invalidate_inode_pages and
10391da177e4SLinus Torvalds  *	  writebacks since the page->count is kept > 1 for as long
10401da177e4SLinus Torvalds  *	  as the page has a write request pending.
10411da177e4SLinus Torvalds  */
1042788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
10431da177e4SLinus Torvalds {
1044788e7a89STrond Myklebust 	struct nfs_write_data	*data = calldata;
10451da177e4SLinus Torvalds 
1046c9d8f89dSTrond Myklebust 	nfs_writeback_done(task, data);
1047c9d8f89dSTrond Myklebust }
1048c9d8f89dSTrond Myklebust 
1049c9d8f89dSTrond Myklebust static void nfs_writeback_release_full(void *calldata)
1050c9d8f89dSTrond Myklebust {
1051c9d8f89dSTrond Myklebust 	struct nfs_write_data	*data = calldata;
1052c9d8f89dSTrond Myklebust 	int status = data->task.tk_status;
1053788e7a89STrond Myklebust 
10541da177e4SLinus Torvalds 	/* Update attributes as result of writeback. */
10551da177e4SLinus Torvalds 	while (!list_empty(&data->pages)) {
1056c9d8f89dSTrond Myklebust 		struct nfs_page *req = nfs_list_entry(data->pages.next);
1057c9d8f89dSTrond Myklebust 		struct page *page = req->wb_page;
1058c9d8f89dSTrond Myklebust 
10591da177e4SLinus Torvalds 		nfs_list_remove_request(req);
10601da177e4SLinus Torvalds 
10611da177e4SLinus Torvalds 		dprintk("NFS: write (%s/%Ld %d@%Ld)",
106288be9f99STrond Myklebust 			req->wb_context->path.dentry->d_inode->i_sb->s_id,
106388be9f99STrond Myklebust 			(long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
10641da177e4SLinus Torvalds 			req->wb_bytes,
10651da177e4SLinus Torvalds 			(long long)req_offset(req));
10661da177e4SLinus Torvalds 
1067c9d8f89dSTrond Myklebust 		if (status < 0) {
1068a301b777STrond Myklebust 			nfs_set_pageerror(page);
1069c9d8f89dSTrond Myklebust 			nfs_context_set_write_error(req->wb_context, status);
1070c9d8f89dSTrond Myklebust 			dprintk(", error = %d\n", status);
10718e821cadSTrond Myklebust 			goto remove_request;
10721da177e4SLinus Torvalds 		}
10731da177e4SLinus Torvalds 
10748e821cadSTrond Myklebust 		if (nfs_write_need_commit(data)) {
10751da177e4SLinus Torvalds 			memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
10761da177e4SLinus Torvalds 			nfs_mark_request_commit(req);
10778e821cadSTrond Myklebust 			nfs_end_page_writeback(page);
10781da177e4SLinus Torvalds 			dprintk(" marked for commit\n");
10798e821cadSTrond Myklebust 			goto next;
10808e821cadSTrond Myklebust 		}
108144dd151dSTrond Myklebust 		/* Set the PG_uptodate flag? */
108244dd151dSTrond Myklebust 		nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
10838e821cadSTrond Myklebust 		dprintk(" OK\n");
10848e821cadSTrond Myklebust remove_request:
10858e821cadSTrond Myklebust 		nfs_end_page_writeback(page);
10861da177e4SLinus Torvalds 		nfs_inode_remove_request(req);
10871da177e4SLinus Torvalds 	next:
10889fd367f0STrond Myklebust 		nfs_clear_page_tag_locked(req);
10891da177e4SLinus Torvalds 	}
1090c9d8f89dSTrond Myklebust 	nfs_writedata_release(calldata);
10911da177e4SLinus Torvalds }
10921da177e4SLinus Torvalds 
1093788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = {
1094788e7a89STrond Myklebust 	.rpc_call_done = nfs_writeback_done_full,
1095c9d8f89dSTrond Myklebust 	.rpc_release = nfs_writeback_release_full,
1096788e7a89STrond Myklebust };
1097788e7a89STrond Myklebust 
1098788e7a89STrond Myklebust 
10991da177e4SLinus Torvalds /*
11001da177e4SLinus Torvalds  * This function is called when the WRITE call is complete.
11011da177e4SLinus Torvalds  */
1102462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
11031da177e4SLinus Torvalds {
11041da177e4SLinus Torvalds 	struct nfs_writeargs	*argp = &data->args;
11051da177e4SLinus Torvalds 	struct nfs_writeres	*resp = &data->res;
1106788e7a89STrond Myklebust 	int status;
11071da177e4SLinus Torvalds 
1108a3f565b1SChuck Lever 	dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
11091da177e4SLinus Torvalds 		task->tk_pid, task->tk_status);
11101da177e4SLinus Torvalds 
1111f551e44fSChuck Lever 	/*
1112f551e44fSChuck Lever 	 * ->write_done will attempt to use post-op attributes to detect
1113f551e44fSChuck Lever 	 * conflicting writes by other clients.  A strict interpretation
1114f551e44fSChuck Lever 	 * of close-to-open would allow us to continue caching even if
1115f551e44fSChuck Lever 	 * another writer had changed the file, but some applications
1116f551e44fSChuck Lever 	 * depend on tighter cache coherency when writing.
1117f551e44fSChuck Lever 	 */
1118788e7a89STrond Myklebust 	status = NFS_PROTO(data->inode)->write_done(task, data);
1119788e7a89STrond Myklebust 	if (status != 0)
1120788e7a89STrond Myklebust 		return status;
112191d5b470SChuck Lever 	nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
112291d5b470SChuck Lever 
11231da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
11241da177e4SLinus Torvalds 	if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
11251da177e4SLinus Torvalds 		/* We tried a write call, but the server did not
11261da177e4SLinus Torvalds 		 * commit data to stable storage even though we
11271da177e4SLinus Torvalds 		 * requested it.
11281da177e4SLinus Torvalds 		 * Note: There is a known bug in Tru64 < 5.0 in which
11291da177e4SLinus Torvalds 		 *	 the server reports NFS_DATA_SYNC, but performs
11301da177e4SLinus Torvalds 		 *	 NFS_FILE_SYNC. We therefore implement this checking
11311da177e4SLinus Torvalds 		 *	 as a dprintk() in order to avoid filling syslog.
11321da177e4SLinus Torvalds 		 */
11331da177e4SLinus Torvalds 		static unsigned long    complain;
11341da177e4SLinus Torvalds 
11351da177e4SLinus Torvalds 		if (time_before(complain, jiffies)) {
11361da177e4SLinus Torvalds 			dprintk("NFS: faulty NFS server %s:"
11371da177e4SLinus Torvalds 				" (committed = %d) != (stable = %d)\n",
113854ceac45SDavid Howells 				NFS_SERVER(data->inode)->nfs_client->cl_hostname,
11391da177e4SLinus Torvalds 				resp->verf->committed, argp->stable);
11401da177e4SLinus Torvalds 			complain = jiffies + 300 * HZ;
11411da177e4SLinus Torvalds 		}
11421da177e4SLinus Torvalds 	}
11431da177e4SLinus Torvalds #endif
11441da177e4SLinus Torvalds 	/* Is this a short write? */
11451da177e4SLinus Torvalds 	if (task->tk_status >= 0 && resp->count < argp->count) {
11461da177e4SLinus Torvalds 		static unsigned long    complain;
11471da177e4SLinus Torvalds 
114891d5b470SChuck Lever 		nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
114991d5b470SChuck Lever 
11501da177e4SLinus Torvalds 		/* Has the server at least made some progress? */
11511da177e4SLinus Torvalds 		if (resp->count != 0) {
11521da177e4SLinus Torvalds 			/* Was this an NFSv2 write or an NFSv3 stable write? */
11531da177e4SLinus Torvalds 			if (resp->verf->committed != NFS_UNSTABLE) {
11541da177e4SLinus Torvalds 				/* Resend from where the server left off */
11551da177e4SLinus Torvalds 				argp->offset += resp->count;
11561da177e4SLinus Torvalds 				argp->pgbase += resp->count;
11571da177e4SLinus Torvalds 				argp->count -= resp->count;
11581da177e4SLinus Torvalds 			} else {
11591da177e4SLinus Torvalds 				/* Resend as a stable write in order to avoid
11601da177e4SLinus Torvalds 				 * headaches in the case of a server crash.
11611da177e4SLinus Torvalds 				 */
11621da177e4SLinus Torvalds 				argp->stable = NFS_FILE_SYNC;
11631da177e4SLinus Torvalds 			}
11641da177e4SLinus Torvalds 			rpc_restart_call(task);
1165788e7a89STrond Myklebust 			return -EAGAIN;
11661da177e4SLinus Torvalds 		}
11671da177e4SLinus Torvalds 		if (time_before(complain, jiffies)) {
11681da177e4SLinus Torvalds 			printk(KERN_WARNING
11691da177e4SLinus Torvalds 			       "NFS: Server wrote zero bytes, expected %u.\n",
11701da177e4SLinus Torvalds 					argp->count);
11711da177e4SLinus Torvalds 			complain = jiffies + 300 * HZ;
11721da177e4SLinus Torvalds 		}
11731da177e4SLinus Torvalds 		/* Can't do anything about it except throw an error. */
11741da177e4SLinus Torvalds 		task->tk_status = -EIO;
11751da177e4SLinus Torvalds 	}
1176788e7a89STrond Myklebust 	return 0;
11771da177e4SLinus Torvalds }
11781da177e4SLinus Torvalds 
11791da177e4SLinus Torvalds 
11801da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1181c9d8f89dSTrond Myklebust void nfs_commitdata_release(void *data)
11821da177e4SLinus Torvalds {
1183383ba719STrond Myklebust 	struct nfs_write_data *wdata = data;
1184383ba719STrond Myklebust 
1185383ba719STrond Myklebust 	put_nfs_open_context(wdata->args.context);
11861da177e4SLinus Torvalds 	nfs_commit_free(wdata);
11871da177e4SLinus Torvalds }
11881da177e4SLinus Torvalds 
11891da177e4SLinus Torvalds /*
11901da177e4SLinus Torvalds  * Set up the argument/result storage required for the RPC call.
11911da177e4SLinus Torvalds  */
1192dbae4c73STrond Myklebust static int nfs_commit_rpcsetup(struct list_head *head,
1193788e7a89STrond Myklebust 		struct nfs_write_data *data,
1194788e7a89STrond Myklebust 		int how)
11951da177e4SLinus Torvalds {
119684115e1cSTrond Myklebust 	struct nfs_page *first = nfs_list_entry(head->next);
119784115e1cSTrond Myklebust 	struct inode *inode = first->wb_context->path.dentry->d_inode;
119884115e1cSTrond Myklebust 	int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
11993ff7576dSTrond Myklebust 	int priority = flush_task_priority(how);
120007737691STrond Myklebust 	struct rpc_task *task;
1201bdc7f021STrond Myklebust 	struct rpc_message msg = {
1202bdc7f021STrond Myklebust 		.rpc_argp = &data->args,
1203bdc7f021STrond Myklebust 		.rpc_resp = &data->res,
1204bdc7f021STrond Myklebust 		.rpc_cred = first->wb_context->cred,
1205bdc7f021STrond Myklebust 	};
120684115e1cSTrond Myklebust 	struct rpc_task_setup task_setup_data = {
120707737691STrond Myklebust 		.task = &data->task,
120884115e1cSTrond Myklebust 		.rpc_client = NFS_CLIENT(inode),
1209bdc7f021STrond Myklebust 		.rpc_message = &msg,
121084115e1cSTrond Myklebust 		.callback_ops = &nfs_commit_ops,
121184115e1cSTrond Myklebust 		.callback_data = data,
1212101070caSTrond Myklebust 		.workqueue = nfsiod_workqueue,
121384115e1cSTrond Myklebust 		.flags = flags,
12143ff7576dSTrond Myklebust 		.priority = priority,
121584115e1cSTrond Myklebust 	};
12161da177e4SLinus Torvalds 
12171da177e4SLinus Torvalds 	/* Set up the RPC argument and reply structs
12181da177e4SLinus Torvalds 	 * NB: take care not to mess about with data->commit et al. */
12191da177e4SLinus Torvalds 
12201da177e4SLinus Torvalds 	list_splice_init(head, &data->pages);
12211da177e4SLinus Torvalds 
12221da177e4SLinus Torvalds 	data->inode	  = inode;
1223bdc7f021STrond Myklebust 	data->cred	  = msg.rpc_cred;
12241da177e4SLinus Torvalds 
12251da177e4SLinus Torvalds 	data->args.fh     = NFS_FH(data->inode);
12263da28eb1STrond Myklebust 	/* Note: we always request a commit of the entire inode */
12273da28eb1STrond Myklebust 	data->args.offset = 0;
12283da28eb1STrond Myklebust 	data->args.count  = 0;
1229383ba719STrond Myklebust 	data->args.context = get_nfs_open_context(first->wb_context);
12303da28eb1STrond Myklebust 	data->res.count   = 0;
12311da177e4SLinus Torvalds 	data->res.fattr   = &data->fattr;
12321da177e4SLinus Torvalds 	data->res.verf    = &data->verf;
12330e574af1STrond Myklebust 	nfs_fattr_init(&data->fattr);
12341da177e4SLinus Torvalds 
1235788e7a89STrond Myklebust 	/* Set up the initial task struct.  */
1236bdc7f021STrond Myklebust 	NFS_PROTO(inode)->commit_setup(data, &msg);
12371da177e4SLinus Torvalds 
1238a3f565b1SChuck Lever 	dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
1239bdc7f021STrond Myklebust 
124007737691STrond Myklebust 	task = rpc_run_task(&task_setup_data);
1241dbae4c73STrond Myklebust 	if (IS_ERR(task))
1242dbae4c73STrond Myklebust 		return PTR_ERR(task);
124307737691STrond Myklebust 	rpc_put_task(task);
1244dbae4c73STrond Myklebust 	return 0;
12451da177e4SLinus Torvalds }
12461da177e4SLinus Torvalds 
12471da177e4SLinus Torvalds /*
12481da177e4SLinus Torvalds  * Commit dirty pages
12491da177e4SLinus Torvalds  */
12501da177e4SLinus Torvalds static int
125140859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how)
12521da177e4SLinus Torvalds {
12531da177e4SLinus Torvalds 	struct nfs_write_data	*data;
12541da177e4SLinus Torvalds 	struct nfs_page         *req;
12551da177e4SLinus Torvalds 
1256c9d8f89dSTrond Myklebust 	data = nfs_commitdata_alloc();
12571da177e4SLinus Torvalds 
12581da177e4SLinus Torvalds 	if (!data)
12591da177e4SLinus Torvalds 		goto out_bad;
12601da177e4SLinus Torvalds 
12611da177e4SLinus Torvalds 	/* Set up the argument struct */
1262dbae4c73STrond Myklebust 	return nfs_commit_rpcsetup(head, data, how);
12631da177e4SLinus Torvalds  out_bad:
12641da177e4SLinus Torvalds 	while (!list_empty(head)) {
12651da177e4SLinus Torvalds 		req = nfs_list_entry(head->next);
12661da177e4SLinus Torvalds 		nfs_list_remove_request(req);
12671da177e4SLinus Torvalds 		nfs_mark_request_commit(req);
126883715ad5STrond Myklebust 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1269c9e51e41SPeter Zijlstra 		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1270c9e51e41SPeter Zijlstra 				BDI_RECLAIMABLE);
12719fd367f0STrond Myklebust 		nfs_clear_page_tag_locked(req);
12721da177e4SLinus Torvalds 	}
12731da177e4SLinus Torvalds 	return -ENOMEM;
12741da177e4SLinus Torvalds }
12751da177e4SLinus Torvalds 
12761da177e4SLinus Torvalds /*
12771da177e4SLinus Torvalds  * COMMIT call returned
12781da177e4SLinus Torvalds  */
1279788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata)
12801da177e4SLinus Torvalds {
1281963d8fe5STrond Myklebust 	struct nfs_write_data	*data = calldata;
12821da177e4SLinus Torvalds 
1283a3f565b1SChuck Lever         dprintk("NFS: %5u nfs_commit_done (status %d)\n",
12841da177e4SLinus Torvalds                                 task->tk_pid, task->tk_status);
12851da177e4SLinus Torvalds 
1286788e7a89STrond Myklebust 	/* Call the NFS version-specific code */
1287788e7a89STrond Myklebust 	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1288788e7a89STrond Myklebust 		return;
1289c9d8f89dSTrond Myklebust }
1290c9d8f89dSTrond Myklebust 
1291c9d8f89dSTrond Myklebust static void nfs_commit_release(void *calldata)
1292c9d8f89dSTrond Myklebust {
1293c9d8f89dSTrond Myklebust 	struct nfs_write_data	*data = calldata;
1294c9d8f89dSTrond Myklebust 	struct nfs_page		*req;
1295c9d8f89dSTrond Myklebust 	int status = data->task.tk_status;
1296788e7a89STrond Myklebust 
12971da177e4SLinus Torvalds 	while (!list_empty(&data->pages)) {
12981da177e4SLinus Torvalds 		req = nfs_list_entry(data->pages.next);
12991da177e4SLinus Torvalds 		nfs_list_remove_request(req);
1300612c9384STrond Myklebust 		clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
1301fd39fc85SChristoph Lameter 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1302c9e51e41SPeter Zijlstra 		dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1303c9e51e41SPeter Zijlstra 				BDI_RECLAIMABLE);
13041da177e4SLinus Torvalds 
13051da177e4SLinus Torvalds 		dprintk("NFS: commit (%s/%Ld %d@%Ld)",
130688be9f99STrond Myklebust 			req->wb_context->path.dentry->d_inode->i_sb->s_id,
130788be9f99STrond Myklebust 			(long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
13081da177e4SLinus Torvalds 			req->wb_bytes,
13091da177e4SLinus Torvalds 			(long long)req_offset(req));
1310c9d8f89dSTrond Myklebust 		if (status < 0) {
1311c9d8f89dSTrond Myklebust 			nfs_context_set_write_error(req->wb_context, status);
13121da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
1313c9d8f89dSTrond Myklebust 			dprintk(", error = %d\n", status);
13141da177e4SLinus Torvalds 			goto next;
13151da177e4SLinus Torvalds 		}
13161da177e4SLinus Torvalds 
13171da177e4SLinus Torvalds 		/* Okay, COMMIT succeeded, apparently. Check the verifier
13181da177e4SLinus Torvalds 		 * returned by the server against all stored verfs. */
13191da177e4SLinus Torvalds 		if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
13201da177e4SLinus Torvalds 			/* We have a match */
132144dd151dSTrond Myklebust 			/* Set the PG_uptodate flag */
132244dd151dSTrond Myklebust 			nfs_mark_uptodate(req->wb_page, req->wb_pgbase,
132344dd151dSTrond Myklebust 					req->wb_bytes);
13241da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
13251da177e4SLinus Torvalds 			dprintk(" OK\n");
13261da177e4SLinus Torvalds 			goto next;
13271da177e4SLinus Torvalds 		}
13281da177e4SLinus Torvalds 		/* We have a mismatch. Write the page again */
13291da177e4SLinus Torvalds 		dprintk(" mismatch\n");
13306d884e8fSFred 		nfs_mark_request_dirty(req);
13311da177e4SLinus Torvalds 	next:
13329fd367f0STrond Myklebust 		nfs_clear_page_tag_locked(req);
13331da177e4SLinus Torvalds 	}
1334c9d8f89dSTrond Myklebust 	nfs_commitdata_release(calldata);
13351da177e4SLinus Torvalds }
1336788e7a89STrond Myklebust 
1337788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = {
1338788e7a89STrond Myklebust 	.rpc_call_done = nfs_commit_done,
1339788e7a89STrond Myklebust 	.rpc_release = nfs_commit_release,
1340788e7a89STrond Myklebust };
13411da177e4SLinus Torvalds 
13423da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how)
13431da177e4SLinus Torvalds {
13441da177e4SLinus Torvalds 	LIST_HEAD(head);
13457d46a49fSTrond Myklebust 	int res;
13461da177e4SLinus Torvalds 
1347587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
13483da28eb1STrond Myklebust 	res = nfs_scan_commit(inode, &head, 0, 0);
1349587142f8STrond Myklebust 	spin_unlock(&inode->i_lock);
13501da177e4SLinus Torvalds 	if (res) {
13517d46a49fSTrond Myklebust 		int error = nfs_commit_list(inode, &head, how);
13521da177e4SLinus Torvalds 		if (error < 0)
13531da177e4SLinus Torvalds 			return error;
13543da28eb1STrond Myklebust 	}
13551da177e4SLinus Torvalds 	return res;
13561da177e4SLinus Torvalds }
1357c63c7b05STrond Myklebust #else
1358c63c7b05STrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1359c63c7b05STrond Myklebust {
1360c63c7b05STrond Myklebust 	return 0;
1361c63c7b05STrond Myklebust }
13621da177e4SLinus Torvalds #endif
13631da177e4SLinus Torvalds 
13641c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
13651da177e4SLinus Torvalds {
13661c75950bSTrond Myklebust 	struct inode *inode = mapping->host;
1367ca52fec1STrond Myklebust 	pgoff_t idx_start, idx_end;
13681c75950bSTrond Myklebust 	unsigned int npages = 0;
1369c42de9ddSTrond Myklebust 	LIST_HEAD(head);
137070b9ecbdSTrond Myklebust 	int nocommit = how & FLUSH_NOCOMMIT;
13713f442547STrond Myklebust 	long pages, ret;
13721da177e4SLinus Torvalds 
13731c75950bSTrond Myklebust 	/* FIXME */
13741c75950bSTrond Myklebust 	if (wbc->range_cyclic)
13751c75950bSTrond Myklebust 		idx_start = 0;
13761c75950bSTrond Myklebust 	else {
13771c75950bSTrond Myklebust 		idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
13781c75950bSTrond Myklebust 		idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
13791c75950bSTrond Myklebust 		if (idx_end > idx_start) {
1380ca52fec1STrond Myklebust 			pgoff_t l_npages = 1 + idx_end - idx_start;
13811c75950bSTrond Myklebust 			npages = l_npages;
13821c75950bSTrond Myklebust 			if (sizeof(npages) != sizeof(l_npages) &&
1383ca52fec1STrond Myklebust 					(pgoff_t)npages != l_npages)
13841c75950bSTrond Myklebust 				npages = 0;
13851c75950bSTrond Myklebust 		}
13861c75950bSTrond Myklebust 	}
1387c42de9ddSTrond Myklebust 	how &= ~FLUSH_NOCOMMIT;
1388587142f8STrond Myklebust 	spin_lock(&inode->i_lock);
13891da177e4SLinus Torvalds 	do {
1390c42de9ddSTrond Myklebust 		ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
1391c42de9ddSTrond Myklebust 		if (ret != 0)
1392c42de9ddSTrond Myklebust 			continue;
1393c42de9ddSTrond Myklebust 		if (nocommit)
1394c42de9ddSTrond Myklebust 			break;
1395d2ccddf0STrond Myklebust 		pages = nfs_scan_commit(inode, &head, idx_start, npages);
1396724c439cSTrond Myklebust 		if (pages == 0)
1397c42de9ddSTrond Myklebust 			break;
1398d2ccddf0STrond Myklebust 		if (how & FLUSH_INVALIDATE) {
1399587142f8STrond Myklebust 			spin_unlock(&inode->i_lock);
140083715ad5STrond Myklebust 			nfs_cancel_commit_list(&head);
1401e8e058e8STrond Myklebust 			ret = pages;
1402587142f8STrond Myklebust 			spin_lock(&inode->i_lock);
1403d2ccddf0STrond Myklebust 			continue;
1404d2ccddf0STrond Myklebust 		}
1405d2ccddf0STrond Myklebust 		pages += nfs_scan_commit(inode, &head, 0, 0);
1406587142f8STrond Myklebust 		spin_unlock(&inode->i_lock);
1407c42de9ddSTrond Myklebust 		ret = nfs_commit_list(inode, &head, how);
1408587142f8STrond Myklebust 		spin_lock(&inode->i_lock);
1409587142f8STrond Myklebust 
1410c42de9ddSTrond Myklebust 	} while (ret >= 0);
1411587142f8STrond Myklebust 	spin_unlock(&inode->i_lock);
1412c42de9ddSTrond Myklebust 	return ret;
14131da177e4SLinus Torvalds }
14141da177e4SLinus Torvalds 
141534901f70STrond Myklebust static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
14161c75950bSTrond Myklebust {
14171c75950bSTrond Myklebust 	int ret;
14181c75950bSTrond Myklebust 
141934901f70STrond Myklebust 	ret = nfs_writepages(mapping, wbc);
142061822ab5STrond Myklebust 	if (ret < 0)
142161822ab5STrond Myklebust 		goto out;
142234901f70STrond Myklebust 	ret = nfs_sync_mapping_wait(mapping, wbc, how);
1423ed90ef51STrond Myklebust 	if (ret < 0)
1424ed90ef51STrond Myklebust 		goto out;
14251c75950bSTrond Myklebust 	return 0;
142661822ab5STrond Myklebust out:
1427e507d9ebSTrond Myklebust 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
14281c75950bSTrond Myklebust 	return ret;
14291c75950bSTrond Myklebust }
14301c75950bSTrond Myklebust 
143134901f70STrond Myklebust /* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */
143234901f70STrond Myklebust static int nfs_write_mapping(struct address_space *mapping, int how)
143334901f70STrond Myklebust {
143434901f70STrond Myklebust 	struct writeback_control wbc = {
143534901f70STrond Myklebust 		.bdi = mapping->backing_dev_info,
143634901f70STrond Myklebust 		.sync_mode = WB_SYNC_NONE,
143734901f70STrond Myklebust 		.nr_to_write = LONG_MAX,
143834901f70STrond Myklebust 		.for_writepages = 1,
143934901f70STrond Myklebust 		.range_cyclic = 1,
144034901f70STrond Myklebust 	};
144134901f70STrond Myklebust 	int ret;
144234901f70STrond Myklebust 
144334901f70STrond Myklebust 	ret = __nfs_write_mapping(mapping, &wbc, how);
144434901f70STrond Myklebust 	if (ret < 0)
144534901f70STrond Myklebust 		return ret;
144634901f70STrond Myklebust 	wbc.sync_mode = WB_SYNC_ALL;
144734901f70STrond Myklebust 	return __nfs_write_mapping(mapping, &wbc, how);
144834901f70STrond Myklebust }
144934901f70STrond Myklebust 
1450ed90ef51STrond Myklebust /*
1451ed90ef51STrond Myklebust  * flush the inode to disk.
1452ed90ef51STrond Myklebust  */
1453ed90ef51STrond Myklebust int nfs_wb_all(struct inode *inode)
14541c75950bSTrond Myklebust {
1455ed90ef51STrond Myklebust 	return nfs_write_mapping(inode->i_mapping, 0);
1456ed90ef51STrond Myklebust }
14571c75950bSTrond Myklebust 
1458ed90ef51STrond Myklebust int nfs_wb_nocommit(struct inode *inode)
1459ed90ef51STrond Myklebust {
1460ed90ef51STrond Myklebust 	return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT);
14611c75950bSTrond Myklebust }
14621c75950bSTrond Myklebust 
14631b3b4a1aSTrond Myklebust int nfs_wb_page_cancel(struct inode *inode, struct page *page)
14641b3b4a1aSTrond Myklebust {
14651b3b4a1aSTrond Myklebust 	struct nfs_page *req;
14661b3b4a1aSTrond Myklebust 	loff_t range_start = page_offset(page);
14671b3b4a1aSTrond Myklebust 	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
14681b3b4a1aSTrond Myklebust 	struct writeback_control wbc = {
14691b3b4a1aSTrond Myklebust 		.bdi = page->mapping->backing_dev_info,
14701b3b4a1aSTrond Myklebust 		.sync_mode = WB_SYNC_ALL,
14711b3b4a1aSTrond Myklebust 		.nr_to_write = LONG_MAX,
14721b3b4a1aSTrond Myklebust 		.range_start = range_start,
14731b3b4a1aSTrond Myklebust 		.range_end = range_end,
14741b3b4a1aSTrond Myklebust 	};
14751b3b4a1aSTrond Myklebust 	int ret = 0;
14761b3b4a1aSTrond Myklebust 
14771b3b4a1aSTrond Myklebust 	BUG_ON(!PageLocked(page));
14781b3b4a1aSTrond Myklebust 	for (;;) {
14791b3b4a1aSTrond Myklebust 		req = nfs_page_find_request(page);
14801b3b4a1aSTrond Myklebust 		if (req == NULL)
14811b3b4a1aSTrond Myklebust 			goto out;
14821b3b4a1aSTrond Myklebust 		if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
14831b3b4a1aSTrond Myklebust 			nfs_release_request(req);
14841b3b4a1aSTrond Myklebust 			break;
14851b3b4a1aSTrond Myklebust 		}
14861b3b4a1aSTrond Myklebust 		if (nfs_lock_request_dontget(req)) {
14871b3b4a1aSTrond Myklebust 			nfs_inode_remove_request(req);
14881b3b4a1aSTrond Myklebust 			/*
14891b3b4a1aSTrond Myklebust 			 * In case nfs_inode_remove_request has marked the
14901b3b4a1aSTrond Myklebust 			 * page as being dirty
14911b3b4a1aSTrond Myklebust 			 */
14921b3b4a1aSTrond Myklebust 			cancel_dirty_page(page, PAGE_CACHE_SIZE);
14931b3b4a1aSTrond Myklebust 			nfs_unlock_request(req);
14941b3b4a1aSTrond Myklebust 			break;
14951b3b4a1aSTrond Myklebust 		}
14961b3b4a1aSTrond Myklebust 		ret = nfs_wait_on_request(req);
14971b3b4a1aSTrond Myklebust 		if (ret < 0)
14981b3b4a1aSTrond Myklebust 			goto out;
14991b3b4a1aSTrond Myklebust 	}
15001b3b4a1aSTrond Myklebust 	if (!PagePrivate(page))
15011b3b4a1aSTrond Myklebust 		return 0;
15021b3b4a1aSTrond Myklebust 	ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE);
15031b3b4a1aSTrond Myklebust out:
15041b3b4a1aSTrond Myklebust 	return ret;
15051b3b4a1aSTrond Myklebust }
15061b3b4a1aSTrond Myklebust 
15075334eb13SAdrian Bunk static int nfs_wb_page_priority(struct inode *inode, struct page *page,
15085334eb13SAdrian Bunk 				int how)
15091c75950bSTrond Myklebust {
15101c75950bSTrond Myklebust 	loff_t range_start = page_offset(page);
15111c75950bSTrond Myklebust 	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
15124d770ccfSTrond Myklebust 	struct writeback_control wbc = {
15134d770ccfSTrond Myklebust 		.bdi = page->mapping->backing_dev_info,
15144d770ccfSTrond Myklebust 		.sync_mode = WB_SYNC_ALL,
15154d770ccfSTrond Myklebust 		.nr_to_write = LONG_MAX,
15164d770ccfSTrond Myklebust 		.range_start = range_start,
15174d770ccfSTrond Myklebust 		.range_end = range_end,
15184d770ccfSTrond Myklebust 	};
15194d770ccfSTrond Myklebust 	int ret;
15201c75950bSTrond Myklebust 
152173e3302fSTrond Myklebust 	do {
1522c63c7b05STrond Myklebust 		if (clear_page_dirty_for_io(page)) {
15234d770ccfSTrond Myklebust 			ret = nfs_writepage_locked(page, &wbc);
15244d770ccfSTrond Myklebust 			if (ret < 0)
152573e3302fSTrond Myklebust 				goto out_error;
152673e3302fSTrond Myklebust 		} else if (!PagePrivate(page))
152773e3302fSTrond Myklebust 			break;
15284d770ccfSTrond Myklebust 		ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
152973e3302fSTrond Myklebust 		if (ret < 0)
153073e3302fSTrond Myklebust 			goto out_error;
153173e3302fSTrond Myklebust 	} while (PagePrivate(page));
15324d770ccfSTrond Myklebust 	return 0;
153373e3302fSTrond Myklebust out_error:
1534e507d9ebSTrond Myklebust 	__mark_inode_dirty(inode, I_DIRTY_PAGES);
15354d770ccfSTrond Myklebust 	return ret;
15361c75950bSTrond Myklebust }
15371c75950bSTrond Myklebust 
15381c75950bSTrond Myklebust /*
15391c75950bSTrond Myklebust  * Write back all requests on one page - we do this before reading it.
15401c75950bSTrond Myklebust  */
15411c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page)
15421c75950bSTrond Myklebust {
15434d770ccfSTrond Myklebust 	return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
15441c75950bSTrond Myklebust }
15451c75950bSTrond Myklebust 
1546f7b422b1SDavid Howells int __init nfs_init_writepagecache(void)
15471da177e4SLinus Torvalds {
15481da177e4SLinus Torvalds 	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
15491da177e4SLinus Torvalds 					     sizeof(struct nfs_write_data),
15501da177e4SLinus Torvalds 					     0, SLAB_HWCACHE_ALIGN,
155120c2df83SPaul Mundt 					     NULL);
15521da177e4SLinus Torvalds 	if (nfs_wdata_cachep == NULL)
15531da177e4SLinus Torvalds 		return -ENOMEM;
15541da177e4SLinus Torvalds 
155593d2341cSMatthew Dobson 	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
15561da177e4SLinus Torvalds 						     nfs_wdata_cachep);
15571da177e4SLinus Torvalds 	if (nfs_wdata_mempool == NULL)
15581da177e4SLinus Torvalds 		return -ENOMEM;
15591da177e4SLinus Torvalds 
156093d2341cSMatthew Dobson 	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
15611da177e4SLinus Torvalds 						      nfs_wdata_cachep);
15621da177e4SLinus Torvalds 	if (nfs_commit_mempool == NULL)
15631da177e4SLinus Torvalds 		return -ENOMEM;
15641da177e4SLinus Torvalds 
156589a09141SPeter Zijlstra 	/*
156689a09141SPeter Zijlstra 	 * NFS congestion size, scale with available memory.
156789a09141SPeter Zijlstra 	 *
156889a09141SPeter Zijlstra 	 *  64MB:    8192k
156989a09141SPeter Zijlstra 	 * 128MB:   11585k
157089a09141SPeter Zijlstra 	 * 256MB:   16384k
157189a09141SPeter Zijlstra 	 * 512MB:   23170k
157289a09141SPeter Zijlstra 	 *   1GB:   32768k
157389a09141SPeter Zijlstra 	 *   2GB:   46340k
157489a09141SPeter Zijlstra 	 *   4GB:   65536k
157589a09141SPeter Zijlstra 	 *   8GB:   92681k
157689a09141SPeter Zijlstra 	 *  16GB:  131072k
157789a09141SPeter Zijlstra 	 *
157889a09141SPeter Zijlstra 	 * This allows larger machines to have larger/more transfers.
157989a09141SPeter Zijlstra 	 * Limit the default to 256M
158089a09141SPeter Zijlstra 	 */
158189a09141SPeter Zijlstra 	nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
158289a09141SPeter Zijlstra 	if (nfs_congestion_kb > 256*1024)
158389a09141SPeter Zijlstra 		nfs_congestion_kb = 256*1024;
158489a09141SPeter Zijlstra 
15851da177e4SLinus Torvalds 	return 0;
15861da177e4SLinus Torvalds }
15871da177e4SLinus Torvalds 
1588266bee88SDavid Brownell void nfs_destroy_writepagecache(void)
15891da177e4SLinus Torvalds {
15901da177e4SLinus Torvalds 	mempool_destroy(nfs_commit_mempool);
15911da177e4SLinus Torvalds 	mempool_destroy(nfs_wdata_mempool);
15921a1d92c1SAlexey Dobriyan 	kmem_cache_destroy(nfs_wdata_cachep);
15931da177e4SLinus Torvalds }
15941da177e4SLinus Torvalds 
1595