xref: /openbmc/linux/fs/nfs/write.c (revision f40313ac)
11da177e4SLinus Torvalds /*
21da177e4SLinus Torvalds  * linux/fs/nfs/write.c
31da177e4SLinus Torvalds  *
41da177e4SLinus Torvalds  * Writing file data over NFS.
51da177e4SLinus Torvalds  *
61da177e4SLinus Torvalds  * We do it like this: When a (user) process wishes to write data to an
71da177e4SLinus Torvalds  * NFS file, a write request is allocated that contains the RPC task data
81da177e4SLinus Torvalds  * plus some info on the page to be written, and added to the inode's
91da177e4SLinus Torvalds  * write chain. If the process writes past the end of the page, an async
101da177e4SLinus Torvalds  * RPC call to write the page is scheduled immediately; otherwise, the call
111da177e4SLinus Torvalds  * is delayed for a few seconds.
121da177e4SLinus Torvalds  *
131da177e4SLinus Torvalds  * Just like readahead, no async I/O is performed if wsize < PAGE_SIZE.
141da177e4SLinus Torvalds  *
151da177e4SLinus Torvalds  * Write requests are kept on the inode's writeback list. Each entry in
161da177e4SLinus Torvalds  * that list references the page (portion) to be written. When the
171da177e4SLinus Torvalds  * cache timeout has expired, the RPC task is woken up, and tries to
181da177e4SLinus Torvalds  * lock the page. As soon as it manages to do so, the request is moved
191da177e4SLinus Torvalds  * from the writeback list to the writelock list.
201da177e4SLinus Torvalds  *
211da177e4SLinus Torvalds  * Note: we must make sure never to confuse the inode passed in the
221da177e4SLinus Torvalds  * write_page request with the one in page->inode. As far as I understand
231da177e4SLinus Torvalds  * it, these are different when doing a swap-out.
241da177e4SLinus Torvalds  *
251da177e4SLinus Torvalds  * To understand everything that goes on here and in the NFS read code,
261da177e4SLinus Torvalds  * one should be aware that a page is locked in exactly one of the following
271da177e4SLinus Torvalds  * cases:
281da177e4SLinus Torvalds  *
291da177e4SLinus Torvalds  *  -	A write request is in progress.
301da177e4SLinus Torvalds  *  -	A user process is in generic_file_write/nfs_update_page
311da177e4SLinus Torvalds  *  -	A user process is in generic_file_read
321da177e4SLinus Torvalds  *
331da177e4SLinus Torvalds  * Also note that because of the way pages are invalidated in
341da177e4SLinus Torvalds  * nfs_revalidate_inode, the following assertions hold:
351da177e4SLinus Torvalds  *
361da177e4SLinus Torvalds  *  -	If a page is dirty, there will be no read requests (a page will
371da177e4SLinus Torvalds  *	not be re-read unless invalidated by nfs_revalidate_inode).
381da177e4SLinus Torvalds  *  -	If the page is not uptodate, there will be no pending write
391da177e4SLinus Torvalds  *	requests, and no process will be in nfs_update_page.
401da177e4SLinus Torvalds  *
411da177e4SLinus Torvalds  * FIXME: Interaction with the vmscan routines is not optimal yet.
421da177e4SLinus Torvalds  * Either vmscan must be made nfs-savvy, or we need a different page
431da177e4SLinus Torvalds  * reclaim concept that supports something like FS-independent
441da177e4SLinus Torvalds  * buffer_heads with a b_ops-> field.
451da177e4SLinus Torvalds  *
461da177e4SLinus Torvalds  * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
471da177e4SLinus Torvalds  */
481da177e4SLinus Torvalds 
491da177e4SLinus Torvalds #include <linux/types.h>
501da177e4SLinus Torvalds #include <linux/slab.h>
511da177e4SLinus Torvalds #include <linux/mm.h>
521da177e4SLinus Torvalds #include <linux/pagemap.h>
531da177e4SLinus Torvalds #include <linux/file.h>
541da177e4SLinus Torvalds #include <linux/writeback.h>
551da177e4SLinus Torvalds 
561da177e4SLinus Torvalds #include <linux/sunrpc/clnt.h>
571da177e4SLinus Torvalds #include <linux/nfs_fs.h>
581da177e4SLinus Torvalds #include <linux/nfs_mount.h>
591da177e4SLinus Torvalds #include <linux/nfs_page.h>
603fcfab16SAndrew Morton #include <linux/backing-dev.h>
613fcfab16SAndrew Morton 
621da177e4SLinus Torvalds #include <asm/uaccess.h>
631da177e4SLinus Torvalds #include <linux/smp_lock.h>
641da177e4SLinus Torvalds 
651da177e4SLinus Torvalds #include "delegation.h"
6649a70f27STrond Myklebust #include "internal.h"
6791d5b470SChuck Lever #include "iostat.h"
681da177e4SLinus Torvalds 
691da177e4SLinus Torvalds #define NFSDBG_FACILITY		NFSDBG_PAGECACHE
701da177e4SLinus Torvalds 
711da177e4SLinus Torvalds #define MIN_POOL_WRITE		(32)
721da177e4SLinus Torvalds #define MIN_POOL_COMMIT		(4)
731da177e4SLinus Torvalds 
741da177e4SLinus Torvalds /*
751da177e4SLinus Torvalds  * Local function declarations
761da177e4SLinus Torvalds  */
771da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context*,
781da177e4SLinus Torvalds 					    struct page *,
791da177e4SLinus Torvalds 					    unsigned int, unsigned int);
80e261f51fSTrond Myklebust static void nfs_mark_request_dirty(struct nfs_page *req);
811da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *, int);
821da177e4SLinus Torvalds static int nfs_wait_on_requests(struct inode *, unsigned long, unsigned int);
833f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how);
84788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops;
85788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops;
86788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops;
871da177e4SLinus Torvalds 
88e18b890bSChristoph Lameter static struct kmem_cache *nfs_wdata_cachep;
893feb2d49STrond Myklebust static mempool_t *nfs_wdata_mempool;
901da177e4SLinus Torvalds static mempool_t *nfs_commit_mempool;
911da177e4SLinus Torvalds 
921da177e4SLinus Torvalds static DECLARE_WAIT_QUEUE_HEAD(nfs_write_congestion);
931da177e4SLinus Torvalds 
94e9f7bee1STrond Myklebust struct nfs_write_data *nfs_commit_alloc(void)
951da177e4SLinus Torvalds {
96e6b4f8daSChristoph Lameter 	struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
9740859d7eSChuck Lever 
981da177e4SLinus Torvalds 	if (p) {
991da177e4SLinus Torvalds 		memset(p, 0, sizeof(*p));
1001da177e4SLinus Torvalds 		INIT_LIST_HEAD(&p->pages);
1011da177e4SLinus Torvalds 	}
1021da177e4SLinus Torvalds 	return p;
1031da177e4SLinus Torvalds }
1041da177e4SLinus Torvalds 
1058aca67f0STrond Myklebust void nfs_commit_rcu_free(struct rcu_head *head)
1061da177e4SLinus Torvalds {
1078aca67f0STrond Myklebust 	struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
10840859d7eSChuck Lever 	if (p && (p->pagevec != &p->page_array[0]))
10940859d7eSChuck Lever 		kfree(p->pagevec);
1101da177e4SLinus Torvalds 	mempool_free(p, nfs_commit_mempool);
1111da177e4SLinus Torvalds }
1121da177e4SLinus Torvalds 
1138aca67f0STrond Myklebust void nfs_commit_free(struct nfs_write_data *wdata)
1148aca67f0STrond Myklebust {
1158aca67f0STrond Myklebust 	call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free);
1168aca67f0STrond Myklebust }
1178aca67f0STrond Myklebust 
118e9f7bee1STrond Myklebust struct nfs_write_data *nfs_writedata_alloc(size_t len)
1193feb2d49STrond Myklebust {
120e9f7bee1STrond Myklebust 	unsigned int pagecount = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
121e6b4f8daSChristoph Lameter 	struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
1223feb2d49STrond Myklebust 
1233feb2d49STrond Myklebust 	if (p) {
1243feb2d49STrond Myklebust 		memset(p, 0, sizeof(*p));
1253feb2d49STrond Myklebust 		INIT_LIST_HEAD(&p->pages);
126e9f7bee1STrond Myklebust 		p->npages = pagecount;
1270d0b5cb3SChuck Lever 		if (pagecount <= ARRAY_SIZE(p->page_array))
1280d0b5cb3SChuck Lever 			p->pagevec = p->page_array;
1293feb2d49STrond Myklebust 		else {
1300d0b5cb3SChuck Lever 			p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
1310d0b5cb3SChuck Lever 			if (!p->pagevec) {
1323feb2d49STrond Myklebust 				mempool_free(p, nfs_wdata_mempool);
1333feb2d49STrond Myklebust 				p = NULL;
1343feb2d49STrond Myklebust 			}
1353feb2d49STrond Myklebust 		}
1363feb2d49STrond Myklebust 	}
1373feb2d49STrond Myklebust 	return p;
1383feb2d49STrond Myklebust }
1393feb2d49STrond Myklebust 
1408aca67f0STrond Myklebust static void nfs_writedata_rcu_free(struct rcu_head *head)
1413feb2d49STrond Myklebust {
1428aca67f0STrond Myklebust 	struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
1433feb2d49STrond Myklebust 	if (p && (p->pagevec != &p->page_array[0]))
1443feb2d49STrond Myklebust 		kfree(p->pagevec);
1453feb2d49STrond Myklebust 	mempool_free(p, nfs_wdata_mempool);
1463feb2d49STrond Myklebust }
1473feb2d49STrond Myklebust 
1488aca67f0STrond Myklebust static void nfs_writedata_free(struct nfs_write_data *wdata)
1498aca67f0STrond Myklebust {
1508aca67f0STrond Myklebust 	call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free);
1518aca67f0STrond Myklebust }
1528aca67f0STrond Myklebust 
153963d8fe5STrond Myklebust void nfs_writedata_release(void *wdata)
1541da177e4SLinus Torvalds {
1551da177e4SLinus Torvalds 	nfs_writedata_free(wdata);
1561da177e4SLinus Torvalds }
1571da177e4SLinus Torvalds 
158277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request_locked(struct page *page)
159277459d2STrond Myklebust {
160277459d2STrond Myklebust 	struct nfs_page *req = NULL;
161277459d2STrond Myklebust 
162277459d2STrond Myklebust 	if (PagePrivate(page)) {
163277459d2STrond Myklebust 		req = (struct nfs_page *)page_private(page);
164277459d2STrond Myklebust 		if (req != NULL)
165277459d2STrond Myklebust 			atomic_inc(&req->wb_count);
166277459d2STrond Myklebust 	}
167277459d2STrond Myklebust 	return req;
168277459d2STrond Myklebust }
169277459d2STrond Myklebust 
170277459d2STrond Myklebust static struct nfs_page *nfs_page_find_request(struct page *page)
171277459d2STrond Myklebust {
172277459d2STrond Myklebust 	struct nfs_page *req = NULL;
173277459d2STrond Myklebust 	spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
174277459d2STrond Myklebust 
175277459d2STrond Myklebust 	spin_lock(req_lock);
176277459d2STrond Myklebust 	req = nfs_page_find_request_locked(page);
177277459d2STrond Myklebust 	spin_unlock(req_lock);
178277459d2STrond Myklebust 	return req;
179277459d2STrond Myklebust }
180277459d2STrond Myklebust 
1811da177e4SLinus Torvalds /* Adjust the file length if we're writing beyond the end */
1821da177e4SLinus Torvalds static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
1831da177e4SLinus Torvalds {
1841da177e4SLinus Torvalds 	struct inode *inode = page->mapping->host;
1851da177e4SLinus Torvalds 	loff_t end, i_size = i_size_read(inode);
1861da177e4SLinus Torvalds 	unsigned long end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
1871da177e4SLinus Torvalds 
1881da177e4SLinus Torvalds 	if (i_size > 0 && page->index < end_index)
1891da177e4SLinus Torvalds 		return;
1901da177e4SLinus Torvalds 	end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
1911da177e4SLinus Torvalds 	if (i_size >= end)
1921da177e4SLinus Torvalds 		return;
19391d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
1941da177e4SLinus Torvalds 	i_size_write(inode, end);
1951da177e4SLinus Torvalds }
1961da177e4SLinus Torvalds 
1971da177e4SLinus Torvalds /* We can set the PG_uptodate flag if we see that a write request
1981da177e4SLinus Torvalds  * covers the full page.
1991da177e4SLinus Torvalds  */
2001da177e4SLinus Torvalds static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
2011da177e4SLinus Torvalds {
2021da177e4SLinus Torvalds 	if (PageUptodate(page))
2031da177e4SLinus Torvalds 		return;
2041da177e4SLinus Torvalds 	if (base != 0)
2051da177e4SLinus Torvalds 		return;
20649a70f27STrond Myklebust 	if (count != nfs_page_length(page))
2071da177e4SLinus Torvalds 		return;
20849a70f27STrond Myklebust 	if (count != PAGE_CACHE_SIZE)
2091da177e4SLinus Torvalds 		memclear_highpage_flush(page, count, PAGE_CACHE_SIZE - count);
2101da177e4SLinus Torvalds 	SetPageUptodate(page);
2111da177e4SLinus Torvalds }
2121da177e4SLinus Torvalds 
213e21195a7STrond Myklebust static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
2141da177e4SLinus Torvalds 		unsigned int offset, unsigned int count)
2151da177e4SLinus Torvalds {
2161da177e4SLinus Torvalds 	struct nfs_page	*req;
217e21195a7STrond Myklebust 	int ret;
2181da177e4SLinus Torvalds 
219e21195a7STrond Myklebust 	for (;;) {
220e21195a7STrond Myklebust 		req = nfs_update_request(ctx, page, offset, count);
221e21195a7STrond Myklebust 		if (!IS_ERR(req))
222e21195a7STrond Myklebust 			break;
223e21195a7STrond Myklebust 		ret = PTR_ERR(req);
224e21195a7STrond Myklebust 		if (ret != -EBUSY)
225e21195a7STrond Myklebust 			return ret;
226e21195a7STrond Myklebust 		ret = nfs_wb_page(page->mapping->host, page);
227e21195a7STrond Myklebust 		if (ret != 0)
228e21195a7STrond Myklebust 			return ret;
229e21195a7STrond Myklebust 	}
2301da177e4SLinus Torvalds 	/* Update file length */
2311da177e4SLinus Torvalds 	nfs_grow_file(page, offset, count);
2321da177e4SLinus Torvalds 	/* Set the PG_uptodate flag? */
2331da177e4SLinus Torvalds 	nfs_mark_uptodate(page, offset, count);
2341da177e4SLinus Torvalds 	nfs_unlock_request(req);
235abd3e641STrond Myklebust 	return 0;
2361da177e4SLinus Torvalds }
2371da177e4SLinus Torvalds 
2381da177e4SLinus Torvalds static int wb_priority(struct writeback_control *wbc)
2391da177e4SLinus Torvalds {
2401da177e4SLinus Torvalds 	if (wbc->for_reclaim)
2411da177e4SLinus Torvalds 		return FLUSH_HIGHPRI;
2421da177e4SLinus Torvalds 	if (wbc->for_kupdate)
2431da177e4SLinus Torvalds 		return FLUSH_LOWPRI;
2441da177e4SLinus Torvalds 	return 0;
2451da177e4SLinus Torvalds }
2461da177e4SLinus Torvalds 
2471da177e4SLinus Torvalds /*
248e261f51fSTrond Myklebust  * Find an associated nfs write request, and prepare to flush it out
249e261f51fSTrond Myklebust  * Returns 1 if there was no write request, or if the request was
250e261f51fSTrond Myklebust  * already tagged by nfs_set_page_dirty.Returns 0 if the request
251e261f51fSTrond Myklebust  * was not tagged.
252e261f51fSTrond Myklebust  * May also return an error if the user signalled nfs_wait_on_request().
253e261f51fSTrond Myklebust  */
254e261f51fSTrond Myklebust static int nfs_page_mark_flush(struct page *page)
255e261f51fSTrond Myklebust {
256e261f51fSTrond Myklebust 	struct nfs_page *req;
257e261f51fSTrond Myklebust 	spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
258e261f51fSTrond Myklebust 	int ret;
259e261f51fSTrond Myklebust 
260e261f51fSTrond Myklebust 	spin_lock(req_lock);
261e261f51fSTrond Myklebust 	for(;;) {
262e261f51fSTrond Myklebust 		req = nfs_page_find_request_locked(page);
263e261f51fSTrond Myklebust 		if (req == NULL) {
264e261f51fSTrond Myklebust 			spin_unlock(req_lock);
265e261f51fSTrond Myklebust 			return 1;
266e261f51fSTrond Myklebust 		}
267e261f51fSTrond Myklebust 		if (nfs_lock_request_dontget(req))
268e261f51fSTrond Myklebust 			break;
269e261f51fSTrond Myklebust 		/* Note: If we hold the page lock, as is the case in nfs_writepage,
270e261f51fSTrond Myklebust 		 *	 then the call to nfs_lock_request_dontget() will always
271e261f51fSTrond Myklebust 		 *	 succeed provided that someone hasn't already marked the
272e261f51fSTrond Myklebust 		 *	 request as dirty (in which case we don't care).
273e261f51fSTrond Myklebust 		 */
274e261f51fSTrond Myklebust 		spin_unlock(req_lock);
275e261f51fSTrond Myklebust 		ret = nfs_wait_on_request(req);
276e261f51fSTrond Myklebust 		nfs_release_request(req);
277e261f51fSTrond Myklebust 		if (ret != 0)
278e261f51fSTrond Myklebust 			return ret;
279e261f51fSTrond Myklebust 		spin_lock(req_lock);
280e261f51fSTrond Myklebust 	}
281e261f51fSTrond Myklebust 	spin_unlock(req_lock);
28261822ab5STrond Myklebust 	if (test_and_set_bit(PG_FLUSHING, &req->wb_flags) == 0) {
283e261f51fSTrond Myklebust 		nfs_mark_request_dirty(req);
28461822ab5STrond Myklebust 		set_page_writeback(page);
28561822ab5STrond Myklebust 	}
286e261f51fSTrond Myklebust 	ret = test_bit(PG_NEED_FLUSH, &req->wb_flags);
287e261f51fSTrond Myklebust 	nfs_unlock_request(req);
288e261f51fSTrond Myklebust 	return ret;
289e261f51fSTrond Myklebust }
290e261f51fSTrond Myklebust 
291e261f51fSTrond Myklebust /*
2921da177e4SLinus Torvalds  * Write an mmapped page to the server.
2931da177e4SLinus Torvalds  */
2944d770ccfSTrond Myklebust static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
2951da177e4SLinus Torvalds {
2961da177e4SLinus Torvalds 	struct nfs_open_context *ctx;
2971da177e4SLinus Torvalds 	struct inode *inode = page->mapping->host;
29849a70f27STrond Myklebust 	unsigned offset;
299e261f51fSTrond Myklebust 	int err;
3001da177e4SLinus Torvalds 
30191d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
30291d5b470SChuck Lever 	nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
30391d5b470SChuck Lever 
304e261f51fSTrond Myklebust 	err = nfs_page_mark_flush(page);
305e261f51fSTrond Myklebust 	if (err <= 0)
3061a54533eSTrond Myklebust 		goto out;
307e261f51fSTrond Myklebust 	err = 0;
30849a70f27STrond Myklebust 	offset = nfs_page_length(page);
30949a70f27STrond Myklebust 	if (!offset)
3101da177e4SLinus Torvalds 		goto out;
31149a70f27STrond Myklebust 
312d530838bSTrond Myklebust 	ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE);
3131da177e4SLinus Torvalds 	if (ctx == NULL) {
3141da177e4SLinus Torvalds 		err = -EBADF;
3151da177e4SLinus Torvalds 		goto out;
3161da177e4SLinus Torvalds 	}
317e21195a7STrond Myklebust 	err = nfs_writepage_setup(ctx, page, 0, offset);
318200baa21STrond Myklebust 	put_nfs_open_context(ctx);
319e261f51fSTrond Myklebust 	if (err != 0)
320e261f51fSTrond Myklebust 		goto out;
321e261f51fSTrond Myklebust 	err = nfs_page_mark_flush(page);
322e261f51fSTrond Myklebust 	if (err > 0)
323e261f51fSTrond Myklebust 		err = 0;
324200baa21STrond Myklebust out:
325abd3e641STrond Myklebust 	if (!wbc->for_writepages)
32602241bc4STrond Myklebust 		nfs_flush_mapping(page->mapping, wbc, FLUSH_STABLE|wb_priority(wbc));
3274d770ccfSTrond Myklebust 	return err;
3284d770ccfSTrond Myklebust }
3294d770ccfSTrond Myklebust 
3304d770ccfSTrond Myklebust int nfs_writepage(struct page *page, struct writeback_control *wbc)
3314d770ccfSTrond Myklebust {
3324d770ccfSTrond Myklebust 	int err;
3334d770ccfSTrond Myklebust 
3344d770ccfSTrond Myklebust 	err = nfs_writepage_locked(page, wbc);
3351da177e4SLinus Torvalds 	unlock_page(page);
3361da177e4SLinus Torvalds 	return err;
3371da177e4SLinus Torvalds }
3381da177e4SLinus Torvalds 
3391da177e4SLinus Torvalds /*
3401da177e4SLinus Torvalds  * Note: causes nfs_update_request() to block on the assumption
3411da177e4SLinus Torvalds  * 	 that the writeback is generated due to memory pressure.
3421da177e4SLinus Torvalds  */
3431da177e4SLinus Torvalds int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
3441da177e4SLinus Torvalds {
3451da177e4SLinus Torvalds 	struct backing_dev_info *bdi = mapping->backing_dev_info;
3461da177e4SLinus Torvalds 	struct inode *inode = mapping->host;
3471da177e4SLinus Torvalds 	int err;
3481da177e4SLinus Torvalds 
34991d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
35091d5b470SChuck Lever 
3511da177e4SLinus Torvalds 	err = generic_writepages(mapping, wbc);
3521da177e4SLinus Torvalds 	if (err)
3531da177e4SLinus Torvalds 		return err;
3541da177e4SLinus Torvalds 	while (test_and_set_bit(BDI_write_congested, &bdi->state) != 0) {
3551da177e4SLinus Torvalds 		if (wbc->nonblocking)
3561da177e4SLinus Torvalds 			return 0;
3571da177e4SLinus Torvalds 		nfs_wait_on_write_congestion(mapping, 0);
3581da177e4SLinus Torvalds 	}
35928c6925fSTrond Myklebust 	err = nfs_flush_mapping(mapping, wbc, wb_priority(wbc));
3601da177e4SLinus Torvalds 	if (err < 0)
3611da177e4SLinus Torvalds 		goto out;
36291d5b470SChuck Lever 	nfs_add_stats(inode, NFSIOS_WRITEPAGES, err);
3631da177e4SLinus Torvalds 	if (!wbc->nonblocking && wbc->sync_mode == WB_SYNC_ALL) {
3641da177e4SLinus Torvalds 		err = nfs_wait_on_requests(inode, 0, 0);
3651da177e4SLinus Torvalds 		if (err < 0)
3661da177e4SLinus Torvalds 			goto out;
3671da177e4SLinus Torvalds 	}
3683da28eb1STrond Myklebust 	err = nfs_commit_inode(inode, wb_priority(wbc));
3693f442547STrond Myklebust 	if (err > 0)
3701da177e4SLinus Torvalds 		err = 0;
3711da177e4SLinus Torvalds out:
3721da177e4SLinus Torvalds 	clear_bit(BDI_write_congested, &bdi->state);
3731da177e4SLinus Torvalds 	wake_up_all(&nfs_write_congestion);
3743fcfab16SAndrew Morton 	congestion_end(WRITE);
3751da177e4SLinus Torvalds 	return err;
3761da177e4SLinus Torvalds }
3771da177e4SLinus Torvalds 
3781da177e4SLinus Torvalds /*
3791da177e4SLinus Torvalds  * Insert a write request into an inode
3801da177e4SLinus Torvalds  */
3811da177e4SLinus Torvalds static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
3821da177e4SLinus Torvalds {
3831da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
3841da177e4SLinus Torvalds 	int error;
3851da177e4SLinus Torvalds 
3861da177e4SLinus Torvalds 	error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
3871da177e4SLinus Torvalds 	BUG_ON(error == -EEXIST);
3881da177e4SLinus Torvalds 	if (error)
3891da177e4SLinus Torvalds 		return error;
3901da177e4SLinus Torvalds 	if (!nfsi->npages) {
3911da177e4SLinus Torvalds 		igrab(inode);
3921da177e4SLinus Torvalds 		nfs_begin_data_update(inode);
3931da177e4SLinus Torvalds 		if (nfs_have_delegation(inode, FMODE_WRITE))
3941da177e4SLinus Torvalds 			nfsi->change_attr++;
3951da177e4SLinus Torvalds 	}
396deb7d638STrond Myklebust 	SetPagePrivate(req->wb_page);
397277459d2STrond Myklebust 	set_page_private(req->wb_page, (unsigned long)req);
3981da177e4SLinus Torvalds 	nfsi->npages++;
3991da177e4SLinus Torvalds 	atomic_inc(&req->wb_count);
4001da177e4SLinus Torvalds 	return 0;
4011da177e4SLinus Torvalds }
4021da177e4SLinus Torvalds 
4031da177e4SLinus Torvalds /*
4041da177e4SLinus Torvalds  * Insert a write request into an inode
4051da177e4SLinus Torvalds  */
4061da177e4SLinus Torvalds static void nfs_inode_remove_request(struct nfs_page *req)
4071da177e4SLinus Torvalds {
4081da177e4SLinus Torvalds 	struct inode *inode = req->wb_context->dentry->d_inode;
4091da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4101da177e4SLinus Torvalds 
4111da177e4SLinus Torvalds 	BUG_ON (!NFS_WBACK_BUSY(req));
4121da177e4SLinus Torvalds 
4131da177e4SLinus Torvalds 	spin_lock(&nfsi->req_lock);
414277459d2STrond Myklebust 	set_page_private(req->wb_page, 0);
415deb7d638STrond Myklebust 	ClearPagePrivate(req->wb_page);
4161da177e4SLinus Torvalds 	radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
4171da177e4SLinus Torvalds 	nfsi->npages--;
4181da177e4SLinus Torvalds 	if (!nfsi->npages) {
4191da177e4SLinus Torvalds 		spin_unlock(&nfsi->req_lock);
420951a143bSTrond Myklebust 		nfs_end_data_update(inode);
4211da177e4SLinus Torvalds 		iput(inode);
4221da177e4SLinus Torvalds 	} else
4231da177e4SLinus Torvalds 		spin_unlock(&nfsi->req_lock);
4241da177e4SLinus Torvalds 	nfs_clear_request(req);
4251da177e4SLinus Torvalds 	nfs_release_request(req);
4261da177e4SLinus Torvalds }
4271da177e4SLinus Torvalds 
4281da177e4SLinus Torvalds /*
4291da177e4SLinus Torvalds  * Add a request to the inode's dirty list.
4301da177e4SLinus Torvalds  */
4311da177e4SLinus Torvalds static void
4321da177e4SLinus Torvalds nfs_mark_request_dirty(struct nfs_page *req)
4331da177e4SLinus Torvalds {
4341da177e4SLinus Torvalds 	struct inode *inode = req->wb_context->dentry->d_inode;
4351da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4361da177e4SLinus Torvalds 
4371da177e4SLinus Torvalds 	spin_lock(&nfsi->req_lock);
4383da28eb1STrond Myklebust 	radix_tree_tag_set(&nfsi->nfs_page_tree,
4393da28eb1STrond Myklebust 			req->wb_index, NFS_PAGE_TAG_DIRTY);
4401da177e4SLinus Torvalds 	nfs_list_add_request(req, &nfsi->dirty);
4411da177e4SLinus Torvalds 	nfsi->ndirty++;
4421da177e4SLinus Torvalds 	spin_unlock(&nfsi->req_lock);
443a1803044STrond Myklebust 	__mark_inode_dirty(inode, I_DIRTY_PAGES);
4441da177e4SLinus Torvalds }
4451da177e4SLinus Torvalds 
44661822ab5STrond Myklebust static void
44761822ab5STrond Myklebust nfs_redirty_request(struct nfs_page *req)
44861822ab5STrond Myklebust {
44961822ab5STrond Myklebust 	clear_bit(PG_FLUSHING, &req->wb_flags);
45061822ab5STrond Myklebust 	__set_page_dirty_nobuffers(req->wb_page);
45161822ab5STrond Myklebust }
45261822ab5STrond Myklebust 
4531da177e4SLinus Torvalds /*
4541da177e4SLinus Torvalds  * Check if a request is dirty
4551da177e4SLinus Torvalds  */
4561da177e4SLinus Torvalds static inline int
4571da177e4SLinus Torvalds nfs_dirty_request(struct nfs_page *req)
4581da177e4SLinus Torvalds {
459e261f51fSTrond Myklebust 	return test_bit(PG_FLUSHING, &req->wb_flags) == 0;
4601da177e4SLinus Torvalds }
4611da177e4SLinus Torvalds 
4621da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
4631da177e4SLinus Torvalds /*
4641da177e4SLinus Torvalds  * Add a request to the inode's commit list.
4651da177e4SLinus Torvalds  */
4661da177e4SLinus Torvalds static void
4671da177e4SLinus Torvalds nfs_mark_request_commit(struct nfs_page *req)
4681da177e4SLinus Torvalds {
4691da177e4SLinus Torvalds 	struct inode *inode = req->wb_context->dentry->d_inode;
4701da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4711da177e4SLinus Torvalds 
4721da177e4SLinus Torvalds 	spin_lock(&nfsi->req_lock);
4731da177e4SLinus Torvalds 	nfs_list_add_request(req, &nfsi->commit);
4741da177e4SLinus Torvalds 	nfsi->ncommit++;
4751da177e4SLinus Torvalds 	spin_unlock(&nfsi->req_lock);
476fd39fc85SChristoph Lameter 	inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
477a1803044STrond Myklebust 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
4781da177e4SLinus Torvalds }
4791da177e4SLinus Torvalds #endif
4801da177e4SLinus Torvalds 
4811da177e4SLinus Torvalds /*
4821da177e4SLinus Torvalds  * Wait for a request to complete.
4831da177e4SLinus Torvalds  *
4841da177e4SLinus Torvalds  * Interruptible by signals only if mounted with intr flag.
4851da177e4SLinus Torvalds  */
486c42de9ddSTrond Myklebust static int nfs_wait_on_requests_locked(struct inode *inode, unsigned long idx_start, unsigned int npages)
4871da177e4SLinus Torvalds {
4881da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
4891da177e4SLinus Torvalds 	struct nfs_page *req;
4901da177e4SLinus Torvalds 	unsigned long		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;
500c6a556b8STrond Myklebust 	while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) {
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 
5071da177e4SLinus Torvalds 		atomic_inc(&req->wb_count);
5081da177e4SLinus Torvalds 		spin_unlock(&nfsi->req_lock);
5091da177e4SLinus Torvalds 		error = nfs_wait_on_request(req);
5101da177e4SLinus Torvalds 		nfs_release_request(req);
511c42de9ddSTrond Myklebust 		spin_lock(&nfsi->req_lock);
5121da177e4SLinus Torvalds 		if (error < 0)
5131da177e4SLinus Torvalds 			return error;
5141da177e4SLinus Torvalds 		res++;
5151da177e4SLinus Torvalds 	}
5161da177e4SLinus Torvalds 	return res;
5171da177e4SLinus Torvalds }
5181da177e4SLinus Torvalds 
519c42de9ddSTrond Myklebust static int nfs_wait_on_requests(struct inode *inode, unsigned long idx_start, unsigned int npages)
520c42de9ddSTrond Myklebust {
521c42de9ddSTrond Myklebust 	struct nfs_inode *nfsi = NFS_I(inode);
522c42de9ddSTrond Myklebust 	int ret;
523c42de9ddSTrond Myklebust 
524c42de9ddSTrond Myklebust 	spin_lock(&nfsi->req_lock);
525c42de9ddSTrond Myklebust 	ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
526c42de9ddSTrond Myklebust 	spin_unlock(&nfsi->req_lock);
527c42de9ddSTrond Myklebust 	return ret;
528c42de9ddSTrond Myklebust }
529c42de9ddSTrond Myklebust 
53083715ad5STrond Myklebust static void nfs_cancel_dirty_list(struct list_head *head)
531d2ccddf0STrond Myklebust {
532d2ccddf0STrond Myklebust 	struct nfs_page *req;
533d2ccddf0STrond Myklebust 	while(!list_empty(head)) {
534d2ccddf0STrond Myklebust 		req = nfs_list_entry(head->next);
535d2ccddf0STrond Myklebust 		nfs_list_remove_request(req);
536d2ccddf0STrond Myklebust 		nfs_inode_remove_request(req);
537d2ccddf0STrond Myklebust 		nfs_clear_page_writeback(req);
538d2ccddf0STrond Myklebust 	}
539d2ccddf0STrond Myklebust }
540d2ccddf0STrond Myklebust 
54183715ad5STrond Myklebust static void nfs_cancel_commit_list(struct list_head *head)
54283715ad5STrond Myklebust {
54383715ad5STrond Myklebust 	struct nfs_page *req;
54483715ad5STrond Myklebust 
54583715ad5STrond Myklebust 	while(!list_empty(head)) {
54683715ad5STrond Myklebust 		req = nfs_list_entry(head->next);
547b6dff26aSTrond Myklebust 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
54883715ad5STrond Myklebust 		nfs_list_remove_request(req);
54983715ad5STrond Myklebust 		nfs_inode_remove_request(req);
550b6dff26aSTrond Myklebust 		nfs_unlock_request(req);
55183715ad5STrond Myklebust 	}
55283715ad5STrond Myklebust }
55383715ad5STrond Myklebust 
5541da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
5551da177e4SLinus Torvalds /*
5561da177e4SLinus Torvalds  * nfs_scan_commit - Scan an inode for commit requests
5571da177e4SLinus Torvalds  * @inode: NFS inode to scan
5581da177e4SLinus Torvalds  * @dst: destination list
5591da177e4SLinus Torvalds  * @idx_start: lower bound of page->index to scan.
5601da177e4SLinus Torvalds  * @npages: idx_start + npages sets the upper bound to scan.
5611da177e4SLinus Torvalds  *
5621da177e4SLinus Torvalds  * Moves requests from the inode's 'commit' request list.
5631da177e4SLinus Torvalds  * The requests are *not* checked to ensure that they form a contiguous set.
5641da177e4SLinus Torvalds  */
5651da177e4SLinus Torvalds static int
5661da177e4SLinus Torvalds nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
5671da177e4SLinus Torvalds {
5681da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
5693da28eb1STrond Myklebust 	int res = 0;
5703da28eb1STrond Myklebust 
5713da28eb1STrond Myklebust 	if (nfsi->ncommit != 0) {
572d2ccddf0STrond Myklebust 		res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages);
5731da177e4SLinus Torvalds 		nfsi->ncommit -= res;
5741da177e4SLinus Torvalds 		if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit))
5751da177e4SLinus Torvalds 			printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
5763da28eb1STrond Myklebust 	}
5771da177e4SLinus Torvalds 	return res;
5781da177e4SLinus Torvalds }
579c42de9ddSTrond Myklebust #else
580c42de9ddSTrond Myklebust static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, unsigned long idx_start, unsigned int npages)
581c42de9ddSTrond Myklebust {
582c42de9ddSTrond Myklebust 	return 0;
583c42de9ddSTrond Myklebust }
5841da177e4SLinus Torvalds #endif
5851da177e4SLinus Torvalds 
5861da177e4SLinus Torvalds static int nfs_wait_on_write_congestion(struct address_space *mapping, int intr)
5871da177e4SLinus Torvalds {
5881da177e4SLinus Torvalds 	struct backing_dev_info *bdi = mapping->backing_dev_info;
5891da177e4SLinus Torvalds 	DEFINE_WAIT(wait);
5901da177e4SLinus Torvalds 	int ret = 0;
5911da177e4SLinus Torvalds 
5921da177e4SLinus Torvalds 	might_sleep();
5931da177e4SLinus Torvalds 
5941da177e4SLinus Torvalds 	if (!bdi_write_congested(bdi))
5951da177e4SLinus Torvalds 		return 0;
59691d5b470SChuck Lever 
59791d5b470SChuck Lever 	nfs_inc_stats(mapping->host, NFSIOS_CONGESTIONWAIT);
59891d5b470SChuck Lever 
5991da177e4SLinus Torvalds 	if (intr) {
6001da177e4SLinus Torvalds 		struct rpc_clnt *clnt = NFS_CLIENT(mapping->host);
6011da177e4SLinus Torvalds 		sigset_t oldset;
6021da177e4SLinus Torvalds 
6031da177e4SLinus Torvalds 		rpc_clnt_sigmask(clnt, &oldset);
6041da177e4SLinus Torvalds 		prepare_to_wait(&nfs_write_congestion, &wait, TASK_INTERRUPTIBLE);
6051da177e4SLinus Torvalds 		if (bdi_write_congested(bdi)) {
6061da177e4SLinus Torvalds 			if (signalled())
6071da177e4SLinus Torvalds 				ret = -ERESTARTSYS;
6081da177e4SLinus Torvalds 			else
6091da177e4SLinus Torvalds 				schedule();
6101da177e4SLinus Torvalds 		}
6111da177e4SLinus Torvalds 		rpc_clnt_sigunmask(clnt, &oldset);
6121da177e4SLinus Torvalds 	} else {
6131da177e4SLinus Torvalds 		prepare_to_wait(&nfs_write_congestion, &wait, TASK_UNINTERRUPTIBLE);
6141da177e4SLinus Torvalds 		if (bdi_write_congested(bdi))
6151da177e4SLinus Torvalds 			schedule();
6161da177e4SLinus Torvalds 	}
6171da177e4SLinus Torvalds 	finish_wait(&nfs_write_congestion, &wait);
6181da177e4SLinus Torvalds 	return ret;
6191da177e4SLinus Torvalds }
6201da177e4SLinus Torvalds 
6211da177e4SLinus Torvalds 
6221da177e4SLinus Torvalds /*
6231da177e4SLinus Torvalds  * Try to update any existing write request, or create one if there is none.
6241da177e4SLinus Torvalds  * In order to match, the request's credentials must match those of
6251da177e4SLinus Torvalds  * the calling process.
6261da177e4SLinus Torvalds  *
6271da177e4SLinus Torvalds  * Note: Should always be called with the Page Lock held!
6281da177e4SLinus Torvalds  */
6291da177e4SLinus Torvalds static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
630e21195a7STrond Myklebust 		struct page *page, unsigned int offset, unsigned int bytes)
6311da177e4SLinus Torvalds {
632e21195a7STrond Myklebust 	struct inode *inode = page->mapping->host;
6331da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
6341da177e4SLinus Torvalds 	struct nfs_page		*req, *new = NULL;
6351da177e4SLinus Torvalds 	unsigned long		rqend, end;
6361da177e4SLinus Torvalds 
6371da177e4SLinus Torvalds 	end = offset + bytes;
6381da177e4SLinus Torvalds 
639e21195a7STrond Myklebust 	if (nfs_wait_on_write_congestion(page->mapping, NFS_SERVER(inode)->flags & NFS_MOUNT_INTR))
6401da177e4SLinus Torvalds 		return ERR_PTR(-ERESTARTSYS);
6411da177e4SLinus Torvalds 	for (;;) {
6421da177e4SLinus Torvalds 		/* Loop over all inode entries and see if we find
6431da177e4SLinus Torvalds 		 * A request for the page we wish to update
6441da177e4SLinus Torvalds 		 */
6451da177e4SLinus Torvalds 		spin_lock(&nfsi->req_lock);
646277459d2STrond Myklebust 		req = nfs_page_find_request_locked(page);
6471da177e4SLinus Torvalds 		if (req) {
6481da177e4SLinus Torvalds 			if (!nfs_lock_request_dontget(req)) {
6491da177e4SLinus Torvalds 				int error;
650277459d2STrond Myklebust 
6511da177e4SLinus Torvalds 				spin_unlock(&nfsi->req_lock);
6521da177e4SLinus Torvalds 				error = nfs_wait_on_request(req);
6531da177e4SLinus Torvalds 				nfs_release_request(req);
6541dd594b2SNeil Brown 				if (error < 0) {
6551dd594b2SNeil Brown 					if (new)
6561dd594b2SNeil Brown 						nfs_release_request(new);
6571da177e4SLinus Torvalds 					return ERR_PTR(error);
6581dd594b2SNeil Brown 				}
6591da177e4SLinus Torvalds 				continue;
6601da177e4SLinus Torvalds 			}
6611da177e4SLinus Torvalds 			spin_unlock(&nfsi->req_lock);
6621da177e4SLinus Torvalds 			if (new)
6631da177e4SLinus Torvalds 				nfs_release_request(new);
6641da177e4SLinus Torvalds 			break;
6651da177e4SLinus Torvalds 		}
6661da177e4SLinus Torvalds 
6671da177e4SLinus Torvalds 		if (new) {
6681da177e4SLinus Torvalds 			int error;
6691da177e4SLinus Torvalds 			nfs_lock_request_dontget(new);
6701da177e4SLinus Torvalds 			error = nfs_inode_add_request(inode, new);
6711da177e4SLinus Torvalds 			if (error) {
6721da177e4SLinus Torvalds 				spin_unlock(&nfsi->req_lock);
6731da177e4SLinus Torvalds 				nfs_unlock_request(new);
6741da177e4SLinus Torvalds 				return ERR_PTR(error);
6751da177e4SLinus Torvalds 			}
6761da177e4SLinus Torvalds 			spin_unlock(&nfsi->req_lock);
6771da177e4SLinus Torvalds 			return new;
6781da177e4SLinus Torvalds 		}
6791da177e4SLinus Torvalds 		spin_unlock(&nfsi->req_lock);
6801da177e4SLinus Torvalds 
6811da177e4SLinus Torvalds 		new = nfs_create_request(ctx, inode, page, offset, bytes);
6821da177e4SLinus Torvalds 		if (IS_ERR(new))
6831da177e4SLinus Torvalds 			return new;
6841da177e4SLinus Torvalds 	}
6851da177e4SLinus Torvalds 
6861da177e4SLinus Torvalds 	/* We have a request for our page.
6871da177e4SLinus Torvalds 	 * If the creds don't match, or the
6881da177e4SLinus Torvalds 	 * page addresses don't match,
6891da177e4SLinus Torvalds 	 * tell the caller to wait on the conflicting
6901da177e4SLinus Torvalds 	 * request.
6911da177e4SLinus Torvalds 	 */
6921da177e4SLinus Torvalds 	rqend = req->wb_offset + req->wb_bytes;
6931da177e4SLinus Torvalds 	if (req->wb_context != ctx
6941da177e4SLinus Torvalds 	    || req->wb_page != page
6951da177e4SLinus Torvalds 	    || !nfs_dirty_request(req)
6961da177e4SLinus Torvalds 	    || offset > rqend || end < req->wb_offset) {
6971da177e4SLinus Torvalds 		nfs_unlock_request(req);
6981da177e4SLinus Torvalds 		return ERR_PTR(-EBUSY);
6991da177e4SLinus Torvalds 	}
7001da177e4SLinus Torvalds 
7011da177e4SLinus Torvalds 	/* Okay, the request matches. Update the region */
7021da177e4SLinus Torvalds 	if (offset < req->wb_offset) {
7031da177e4SLinus Torvalds 		req->wb_offset = offset;
7041da177e4SLinus Torvalds 		req->wb_pgbase = offset;
7051da177e4SLinus Torvalds 		req->wb_bytes = rqend - req->wb_offset;
7061da177e4SLinus Torvalds 	}
7071da177e4SLinus Torvalds 
7081da177e4SLinus Torvalds 	if (end > rqend)
7091da177e4SLinus Torvalds 		req->wb_bytes = end - req->wb_offset;
7101da177e4SLinus Torvalds 
7111da177e4SLinus Torvalds 	return req;
7121da177e4SLinus Torvalds }
7131da177e4SLinus Torvalds 
7141da177e4SLinus Torvalds int nfs_flush_incompatible(struct file *file, struct page *page)
7151da177e4SLinus Torvalds {
7161da177e4SLinus Torvalds 	struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
7171da177e4SLinus Torvalds 	struct nfs_page	*req;
7181a54533eSTrond Myklebust 	int do_flush, status;
7191da177e4SLinus Torvalds 	/*
7201da177e4SLinus Torvalds 	 * Look for a request corresponding to this page. If there
7211da177e4SLinus Torvalds 	 * is one, and it belongs to another file, we flush it out
7221da177e4SLinus Torvalds 	 * before we try to copy anything into the page. Do this
7231da177e4SLinus Torvalds 	 * due to the lack of an ACCESS-type call in NFSv2.
7241da177e4SLinus Torvalds 	 * Also do the same if we find a request from an existing
7251da177e4SLinus Torvalds 	 * dropped page.
7261da177e4SLinus Torvalds 	 */
7271a54533eSTrond Myklebust 	do {
728277459d2STrond Myklebust 		req = nfs_page_find_request(page);
7291a54533eSTrond Myklebust 		if (req == NULL)
7301a54533eSTrond Myklebust 			return 0;
7311a54533eSTrond Myklebust 		do_flush = req->wb_page != page || req->wb_context != ctx
732e261f51fSTrond Myklebust 			|| !nfs_dirty_request(req);
7331da177e4SLinus Torvalds 		nfs_release_request(req);
7341a54533eSTrond Myklebust 		if (!do_flush)
7351a54533eSTrond Myklebust 			return 0;
736277459d2STrond Myklebust 		status = nfs_wb_page(page->mapping->host, page);
7371a54533eSTrond Myklebust 	} while (status == 0);
7381a54533eSTrond Myklebust 	return status;
7391da177e4SLinus Torvalds }
7401da177e4SLinus Torvalds 
7411da177e4SLinus Torvalds /*
7421da177e4SLinus Torvalds  * Update and possibly write a cached page of an NFS file.
7431da177e4SLinus Torvalds  *
7441da177e4SLinus Torvalds  * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
7451da177e4SLinus Torvalds  * things with a page scheduled for an RPC call (e.g. invalidate it).
7461da177e4SLinus Torvalds  */
7471da177e4SLinus Torvalds int nfs_updatepage(struct file *file, struct page *page,
7481da177e4SLinus Torvalds 		unsigned int offset, unsigned int count)
7491da177e4SLinus Torvalds {
7501da177e4SLinus Torvalds 	struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
7511da177e4SLinus Torvalds 	struct inode	*inode = page->mapping->host;
7521da177e4SLinus Torvalds 	int		status = 0;
7531da177e4SLinus Torvalds 
75491d5b470SChuck Lever 	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
75591d5b470SChuck Lever 
7561da177e4SLinus Torvalds 	dprintk("NFS:      nfs_updatepage(%s/%s %d@%Ld)\n",
75701cce933SJosef "Jeff" Sipek 		file->f_path.dentry->d_parent->d_name.name,
75801cce933SJosef "Jeff" Sipek 		file->f_path.dentry->d_name.name, count,
7590bbacc40SChuck Lever 		(long long)(page_offset(page) +offset));
7601da177e4SLinus Torvalds 
7611da177e4SLinus Torvalds 	/* If we're not using byte range locks, and we know the page
7621da177e4SLinus Torvalds 	 * is entirely in cache, it may be more efficient to avoid
7631da177e4SLinus Torvalds 	 * fragmenting write requests.
7641da177e4SLinus Torvalds 	 */
765ab0a3dbeSTrond Myklebust 	if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) {
76649a70f27STrond Myklebust 		count = max(count + offset, nfs_page_length(page));
7671da177e4SLinus Torvalds 		offset = 0;
7681da177e4SLinus Torvalds 	}
7691da177e4SLinus Torvalds 
770e21195a7STrond Myklebust 	status = nfs_writepage_setup(ctx, page, offset, count);
771e261f51fSTrond Myklebust 	__set_page_dirty_nobuffers(page);
7721da177e4SLinus Torvalds 
7731da177e4SLinus Torvalds         dprintk("NFS:      nfs_updatepage returns %d (isize %Ld)\n",
7741da177e4SLinus Torvalds 			status, (long long)i_size_read(inode));
7751da177e4SLinus Torvalds 	if (status < 0)
7761da177e4SLinus Torvalds 		ClearPageUptodate(page);
7771da177e4SLinus Torvalds 	return status;
7781da177e4SLinus Torvalds }
7791da177e4SLinus Torvalds 
7801da177e4SLinus Torvalds static void nfs_writepage_release(struct nfs_page *req)
7811da177e4SLinus Torvalds {
7821da177e4SLinus Torvalds 	end_page_writeback(req->wb_page);
7831da177e4SLinus Torvalds 
7841da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
7851da177e4SLinus Torvalds 	if (!PageError(req->wb_page)) {
7861da177e4SLinus Torvalds 		if (NFS_NEED_RESCHED(req)) {
78761822ab5STrond Myklebust 			nfs_redirty_request(req);
7881da177e4SLinus Torvalds 			goto out;
7891da177e4SLinus Torvalds 		} else if (NFS_NEED_COMMIT(req)) {
7901da177e4SLinus Torvalds 			nfs_mark_request_commit(req);
7911da177e4SLinus Torvalds 			goto out;
7921da177e4SLinus Torvalds 		}
7931da177e4SLinus Torvalds 	}
7941da177e4SLinus Torvalds 	nfs_inode_remove_request(req);
7951da177e4SLinus Torvalds 
7961da177e4SLinus Torvalds out:
7971da177e4SLinus Torvalds 	nfs_clear_commit(req);
7981da177e4SLinus Torvalds 	nfs_clear_reschedule(req);
7991da177e4SLinus Torvalds #else
8001da177e4SLinus Torvalds 	nfs_inode_remove_request(req);
8011da177e4SLinus Torvalds #endif
802c6a556b8STrond Myklebust 	nfs_clear_page_writeback(req);
8031da177e4SLinus Torvalds }
8041da177e4SLinus Torvalds 
8051da177e4SLinus Torvalds static inline int flush_task_priority(int how)
8061da177e4SLinus Torvalds {
8071da177e4SLinus Torvalds 	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
8081da177e4SLinus Torvalds 		case FLUSH_HIGHPRI:
8091da177e4SLinus Torvalds 			return RPC_PRIORITY_HIGH;
8101da177e4SLinus Torvalds 		case FLUSH_LOWPRI:
8111da177e4SLinus Torvalds 			return RPC_PRIORITY_LOW;
8121da177e4SLinus Torvalds 	}
8131da177e4SLinus Torvalds 	return RPC_PRIORITY_NORMAL;
8141da177e4SLinus Torvalds }
8151da177e4SLinus Torvalds 
8161da177e4SLinus Torvalds /*
8171da177e4SLinus Torvalds  * Set up the argument/result storage required for the RPC call.
8181da177e4SLinus Torvalds  */
8191da177e4SLinus Torvalds static void nfs_write_rpcsetup(struct nfs_page *req,
8201da177e4SLinus Torvalds 		struct nfs_write_data *data,
821788e7a89STrond Myklebust 		const struct rpc_call_ops *call_ops,
8221da177e4SLinus Torvalds 		unsigned int count, unsigned int offset,
8231da177e4SLinus Torvalds 		int how)
8241da177e4SLinus Torvalds {
8251da177e4SLinus Torvalds 	struct inode		*inode;
826788e7a89STrond Myklebust 	int flags;
8271da177e4SLinus Torvalds 
8281da177e4SLinus Torvalds 	/* Set up the RPC argument and reply structs
8291da177e4SLinus Torvalds 	 * NB: take care not to mess about with data->commit et al. */
8301da177e4SLinus Torvalds 
8311da177e4SLinus Torvalds 	data->req = req;
8321da177e4SLinus Torvalds 	data->inode = inode = req->wb_context->dentry->d_inode;
8331da177e4SLinus Torvalds 	data->cred = req->wb_context->cred;
8341da177e4SLinus Torvalds 
8351da177e4SLinus Torvalds 	data->args.fh     = NFS_FH(inode);
8361da177e4SLinus Torvalds 	data->args.offset = req_offset(req) + offset;
8371da177e4SLinus Torvalds 	data->args.pgbase = req->wb_pgbase + offset;
8381da177e4SLinus Torvalds 	data->args.pages  = data->pagevec;
8391da177e4SLinus Torvalds 	data->args.count  = count;
8401da177e4SLinus Torvalds 	data->args.context = req->wb_context;
8411da177e4SLinus Torvalds 
8421da177e4SLinus Torvalds 	data->res.fattr   = &data->fattr;
8431da177e4SLinus Torvalds 	data->res.count   = count;
8441da177e4SLinus Torvalds 	data->res.verf    = &data->verf;
8450e574af1STrond Myklebust 	nfs_fattr_init(&data->fattr);
8461da177e4SLinus Torvalds 
847788e7a89STrond Myklebust 	/* Set up the initial task struct.  */
848788e7a89STrond Myklebust 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
849788e7a89STrond Myklebust 	rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
8501da177e4SLinus Torvalds 	NFS_PROTO(inode)->write_setup(data, how);
8511da177e4SLinus Torvalds 
8521da177e4SLinus Torvalds 	data->task.tk_priority = flush_task_priority(how);
8531da177e4SLinus Torvalds 	data->task.tk_cookie = (unsigned long)inode;
8541da177e4SLinus Torvalds 
8551da177e4SLinus Torvalds 	dprintk("NFS: %4d initiated write call (req %s/%Ld, %u bytes @ offset %Lu)\n",
8560bbacc40SChuck Lever 		data->task.tk_pid,
8571da177e4SLinus Torvalds 		inode->i_sb->s_id,
8581da177e4SLinus Torvalds 		(long long)NFS_FILEID(inode),
8591da177e4SLinus Torvalds 		count,
8601da177e4SLinus Torvalds 		(unsigned long long)data->args.offset);
8611da177e4SLinus Torvalds }
8621da177e4SLinus Torvalds 
8631da177e4SLinus Torvalds static void nfs_execute_write(struct nfs_write_data *data)
8641da177e4SLinus Torvalds {
8651da177e4SLinus Torvalds 	struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
8661da177e4SLinus Torvalds 	sigset_t oldset;
8671da177e4SLinus Torvalds 
8681da177e4SLinus Torvalds 	rpc_clnt_sigmask(clnt, &oldset);
8691da177e4SLinus Torvalds 	rpc_execute(&data->task);
8701da177e4SLinus Torvalds 	rpc_clnt_sigunmask(clnt, &oldset);
8711da177e4SLinus Torvalds }
8721da177e4SLinus Torvalds 
8731da177e4SLinus Torvalds /*
8741da177e4SLinus Torvalds  * Generate multiple small requests to write out a single
8751da177e4SLinus Torvalds  * contiguous dirty area on one page.
8761da177e4SLinus Torvalds  */
8777d46a49fSTrond Myklebust static int nfs_flush_multi(struct inode *inode, struct list_head *head, int how)
8781da177e4SLinus Torvalds {
8791da177e4SLinus Torvalds 	struct nfs_page *req = nfs_list_entry(head->next);
8801da177e4SLinus Torvalds 	struct page *page = req->wb_page;
8811da177e4SLinus Torvalds 	struct nfs_write_data *data;
882e9f7bee1STrond Myklebust 	size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
883e9f7bee1STrond Myklebust 	unsigned int offset;
8841da177e4SLinus Torvalds 	int requests = 0;
8851da177e4SLinus Torvalds 	LIST_HEAD(list);
8861da177e4SLinus Torvalds 
8871da177e4SLinus Torvalds 	nfs_list_remove_request(req);
8881da177e4SLinus Torvalds 
8891da177e4SLinus Torvalds 	nbytes = req->wb_bytes;
890e9f7bee1STrond Myklebust 	do {
891e9f7bee1STrond Myklebust 		size_t len = min(nbytes, wsize);
892e9f7bee1STrond Myklebust 
893e9f7bee1STrond Myklebust 		data = nfs_writedata_alloc(len);
8941da177e4SLinus Torvalds 		if (!data)
8951da177e4SLinus Torvalds 			goto out_bad;
8961da177e4SLinus Torvalds 		list_add(&data->pages, &list);
8971da177e4SLinus Torvalds 		requests++;
898e9f7bee1STrond Myklebust 		nbytes -= len;
899e9f7bee1STrond Myklebust 	} while (nbytes != 0);
9001da177e4SLinus Torvalds 	atomic_set(&req->wb_complete, requests);
9011da177e4SLinus Torvalds 
9021da177e4SLinus Torvalds 	ClearPageError(page);
9031da177e4SLinus Torvalds 	offset = 0;
9041da177e4SLinus Torvalds 	nbytes = req->wb_bytes;
9051da177e4SLinus Torvalds 	do {
9061da177e4SLinus Torvalds 		data = list_entry(list.next, struct nfs_write_data, pages);
9071da177e4SLinus Torvalds 		list_del_init(&data->pages);
9081da177e4SLinus Torvalds 
9091da177e4SLinus Torvalds 		data->pagevec[0] = page;
9101da177e4SLinus Torvalds 
9111da177e4SLinus Torvalds 		if (nbytes > wsize) {
912788e7a89STrond Myklebust 			nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
913788e7a89STrond Myklebust 					wsize, offset, how);
9141da177e4SLinus Torvalds 			offset += wsize;
9151da177e4SLinus Torvalds 			nbytes -= wsize;
9161da177e4SLinus Torvalds 		} else {
917788e7a89STrond Myklebust 			nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
918788e7a89STrond Myklebust 					nbytes, offset, how);
9191da177e4SLinus Torvalds 			nbytes = 0;
9201da177e4SLinus Torvalds 		}
9211da177e4SLinus Torvalds 		nfs_execute_write(data);
9221da177e4SLinus Torvalds 	} while (nbytes != 0);
9231da177e4SLinus Torvalds 
9241da177e4SLinus Torvalds 	return 0;
9251da177e4SLinus Torvalds 
9261da177e4SLinus Torvalds out_bad:
9271da177e4SLinus Torvalds 	while (!list_empty(&list)) {
9281da177e4SLinus Torvalds 		data = list_entry(list.next, struct nfs_write_data, pages);
9291da177e4SLinus Torvalds 		list_del(&data->pages);
9308aca67f0STrond Myklebust 		nfs_writedata_release(data);
9311da177e4SLinus Torvalds 	}
93261822ab5STrond Myklebust 	nfs_redirty_request(req);
933c6a556b8STrond Myklebust 	nfs_clear_page_writeback(req);
9341da177e4SLinus Torvalds 	return -ENOMEM;
9351da177e4SLinus Torvalds }
9361da177e4SLinus Torvalds 
9371da177e4SLinus Torvalds /*
9381da177e4SLinus Torvalds  * Create an RPC task for the given write request and kick it.
9391da177e4SLinus Torvalds  * The page must have been locked by the caller.
9401da177e4SLinus Torvalds  *
9411da177e4SLinus Torvalds  * It may happen that the page we're passed is not marked dirty.
9421da177e4SLinus Torvalds  * This is the case if nfs_updatepage detects a conflicting request
9431da177e4SLinus Torvalds  * that has been written but not committed.
9441da177e4SLinus Torvalds  */
9457d46a49fSTrond Myklebust static int nfs_flush_one(struct inode *inode, struct list_head *head, int how)
9461da177e4SLinus Torvalds {
9471da177e4SLinus Torvalds 	struct nfs_page		*req;
9481da177e4SLinus Torvalds 	struct page		**pages;
9491da177e4SLinus Torvalds 	struct nfs_write_data	*data;
9501da177e4SLinus Torvalds 	unsigned int		count;
9511da177e4SLinus Torvalds 
952e9f7bee1STrond Myklebust 	data = nfs_writedata_alloc(NFS_SERVER(inode)->wsize);
9531da177e4SLinus Torvalds 	if (!data)
9541da177e4SLinus Torvalds 		goto out_bad;
9551da177e4SLinus Torvalds 
9561da177e4SLinus Torvalds 	pages = data->pagevec;
9571da177e4SLinus Torvalds 	count = 0;
9581da177e4SLinus Torvalds 	while (!list_empty(head)) {
9591da177e4SLinus Torvalds 		req = nfs_list_entry(head->next);
9601da177e4SLinus Torvalds 		nfs_list_remove_request(req);
9611da177e4SLinus Torvalds 		nfs_list_add_request(req, &data->pages);
9621da177e4SLinus Torvalds 		ClearPageError(req->wb_page);
9631da177e4SLinus Torvalds 		*pages++ = req->wb_page;
9641da177e4SLinus Torvalds 		count += req->wb_bytes;
9651da177e4SLinus Torvalds 	}
9661da177e4SLinus Torvalds 	req = nfs_list_entry(data->pages.next);
9671da177e4SLinus Torvalds 
9681da177e4SLinus Torvalds 	/* Set up the argument struct */
969788e7a89STrond Myklebust 	nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
9701da177e4SLinus Torvalds 
9711da177e4SLinus Torvalds 	nfs_execute_write(data);
9721da177e4SLinus Torvalds 	return 0;
9731da177e4SLinus Torvalds  out_bad:
9741da177e4SLinus Torvalds 	while (!list_empty(head)) {
9751da177e4SLinus Torvalds 		struct nfs_page *req = nfs_list_entry(head->next);
9761da177e4SLinus Torvalds 		nfs_list_remove_request(req);
97761822ab5STrond Myklebust 		nfs_redirty_request(req);
978c6a556b8STrond Myklebust 		nfs_clear_page_writeback(req);
9791da177e4SLinus Torvalds 	}
9801da177e4SLinus Torvalds 	return -ENOMEM;
9811da177e4SLinus Torvalds }
9821da177e4SLinus Torvalds 
9837d46a49fSTrond Myklebust static int nfs_flush_list(struct inode *inode, struct list_head *head, int npages, int how)
9841da177e4SLinus Torvalds {
9851da177e4SLinus Torvalds 	LIST_HEAD(one_request);
9867d46a49fSTrond Myklebust 	int (*flush_one)(struct inode *, struct list_head *, int);
9871da177e4SLinus Torvalds 	struct nfs_page	*req;
9887d46a49fSTrond Myklebust 	int wpages = NFS_SERVER(inode)->wpages;
9897d46a49fSTrond Myklebust 	int wsize = NFS_SERVER(inode)->wsize;
9907d46a49fSTrond Myklebust 	int error;
9911da177e4SLinus Torvalds 
9927d46a49fSTrond Myklebust 	flush_one = nfs_flush_one;
9937d46a49fSTrond Myklebust 	if (wsize < PAGE_CACHE_SIZE)
9947d46a49fSTrond Myklebust 		flush_one = nfs_flush_multi;
9957d46a49fSTrond Myklebust 	/* For single writes, FLUSH_STABLE is more efficient */
9967d46a49fSTrond Myklebust 	if (npages <= wpages && npages == NFS_I(inode)->npages
9977d46a49fSTrond Myklebust 			&& nfs_list_entry(head->next)->wb_bytes <= wsize)
9987d46a49fSTrond Myklebust 		how |= FLUSH_STABLE;
9997d46a49fSTrond Myklebust 
10007d46a49fSTrond Myklebust 	do {
10017d46a49fSTrond Myklebust 		nfs_coalesce_requests(head, &one_request, wpages);
10021da177e4SLinus Torvalds 		req = nfs_list_entry(one_request.next);
10037d46a49fSTrond Myklebust 		error = flush_one(inode, &one_request, how);
10041da177e4SLinus Torvalds 		if (error < 0)
10057d46a49fSTrond Myklebust 			goto out_err;
10067d46a49fSTrond Myklebust 	} while (!list_empty(head));
10077d46a49fSTrond Myklebust 	return 0;
10087d46a49fSTrond Myklebust out_err:
10091da177e4SLinus Torvalds 	while (!list_empty(head)) {
10101da177e4SLinus Torvalds 		req = nfs_list_entry(head->next);
10111da177e4SLinus Torvalds 		nfs_list_remove_request(req);
101261822ab5STrond Myklebust 		nfs_redirty_request(req);
1013c6a556b8STrond Myklebust 		nfs_clear_page_writeback(req);
10141da177e4SLinus Torvalds 	}
10151da177e4SLinus Torvalds 	return error;
10161da177e4SLinus Torvalds }
10171da177e4SLinus Torvalds 
10181da177e4SLinus Torvalds /*
10191da177e4SLinus Torvalds  * Handle a write reply that flushed part of a page.
10201da177e4SLinus Torvalds  */
1021788e7a89STrond Myklebust static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
10221da177e4SLinus Torvalds {
1023788e7a89STrond Myklebust 	struct nfs_write_data	*data = calldata;
10241da177e4SLinus Torvalds 	struct nfs_page		*req = data->req;
10251da177e4SLinus Torvalds 	struct page		*page = req->wb_page;
10261da177e4SLinus Torvalds 
10271da177e4SLinus Torvalds 	dprintk("NFS: write (%s/%Ld %d@%Ld)",
10281da177e4SLinus Torvalds 		req->wb_context->dentry->d_inode->i_sb->s_id,
10291da177e4SLinus Torvalds 		(long long)NFS_FILEID(req->wb_context->dentry->d_inode),
10301da177e4SLinus Torvalds 		req->wb_bytes,
10311da177e4SLinus Torvalds 		(long long)req_offset(req));
10321da177e4SLinus Torvalds 
1033788e7a89STrond Myklebust 	if (nfs_writeback_done(task, data) != 0)
1034788e7a89STrond Myklebust 		return;
1035788e7a89STrond Myklebust 
1036788e7a89STrond Myklebust 	if (task->tk_status < 0) {
10371da177e4SLinus Torvalds 		ClearPageUptodate(page);
10381da177e4SLinus Torvalds 		SetPageError(page);
1039788e7a89STrond Myklebust 		req->wb_context->error = task->tk_status;
1040788e7a89STrond Myklebust 		dprintk(", error = %d\n", task->tk_status);
10411da177e4SLinus Torvalds 	} else {
10421da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
10431da177e4SLinus Torvalds 		if (data->verf.committed < NFS_FILE_SYNC) {
10441da177e4SLinus Torvalds 			if (!NFS_NEED_COMMIT(req)) {
10451da177e4SLinus Torvalds 				nfs_defer_commit(req);
10461da177e4SLinus Torvalds 				memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
10471da177e4SLinus Torvalds 				dprintk(" defer commit\n");
10481da177e4SLinus Torvalds 			} else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
10491da177e4SLinus Torvalds 				nfs_defer_reschedule(req);
10501da177e4SLinus Torvalds 				dprintk(" server reboot detected\n");
10511da177e4SLinus Torvalds 			}
10521da177e4SLinus Torvalds 		} else
10531da177e4SLinus Torvalds #endif
10541da177e4SLinus Torvalds 			dprintk(" OK\n");
10551da177e4SLinus Torvalds 	}
10561da177e4SLinus Torvalds 
10571da177e4SLinus Torvalds 	if (atomic_dec_and_test(&req->wb_complete))
10581da177e4SLinus Torvalds 		nfs_writepage_release(req);
10591da177e4SLinus Torvalds }
10601da177e4SLinus Torvalds 
1061788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_partial_ops = {
1062788e7a89STrond Myklebust 	.rpc_call_done = nfs_writeback_done_partial,
1063788e7a89STrond Myklebust 	.rpc_release = nfs_writedata_release,
1064788e7a89STrond Myklebust };
1065788e7a89STrond Myklebust 
10661da177e4SLinus Torvalds /*
10671da177e4SLinus Torvalds  * Handle a write reply that flushes a whole page.
10681da177e4SLinus Torvalds  *
10691da177e4SLinus Torvalds  * FIXME: There is an inherent race with invalidate_inode_pages and
10701da177e4SLinus Torvalds  *	  writebacks since the page->count is kept > 1 for as long
10711da177e4SLinus Torvalds  *	  as the page has a write request pending.
10721da177e4SLinus Torvalds  */
1073788e7a89STrond Myklebust static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
10741da177e4SLinus Torvalds {
1075788e7a89STrond Myklebust 	struct nfs_write_data	*data = calldata;
10761da177e4SLinus Torvalds 	struct nfs_page		*req;
10771da177e4SLinus Torvalds 	struct page		*page;
10781da177e4SLinus Torvalds 
1079788e7a89STrond Myklebust 	if (nfs_writeback_done(task, data) != 0)
1080788e7a89STrond Myklebust 		return;
1081788e7a89STrond Myklebust 
10821da177e4SLinus Torvalds 	/* Update attributes as result of writeback. */
10831da177e4SLinus Torvalds 	while (!list_empty(&data->pages)) {
10841da177e4SLinus Torvalds 		req = nfs_list_entry(data->pages.next);
10851da177e4SLinus Torvalds 		nfs_list_remove_request(req);
10861da177e4SLinus Torvalds 		page = req->wb_page;
10871da177e4SLinus Torvalds 
10881da177e4SLinus Torvalds 		dprintk("NFS: write (%s/%Ld %d@%Ld)",
10891da177e4SLinus Torvalds 			req->wb_context->dentry->d_inode->i_sb->s_id,
10901da177e4SLinus Torvalds 			(long long)NFS_FILEID(req->wb_context->dentry->d_inode),
10911da177e4SLinus Torvalds 			req->wb_bytes,
10921da177e4SLinus Torvalds 			(long long)req_offset(req));
10931da177e4SLinus Torvalds 
1094788e7a89STrond Myklebust 		if (task->tk_status < 0) {
10951da177e4SLinus Torvalds 			ClearPageUptodate(page);
10961da177e4SLinus Torvalds 			SetPageError(page);
1097788e7a89STrond Myklebust 			req->wb_context->error = task->tk_status;
10981da177e4SLinus Torvalds 			end_page_writeback(page);
10991da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
1100788e7a89STrond Myklebust 			dprintk(", error = %d\n", task->tk_status);
11011da177e4SLinus Torvalds 			goto next;
11021da177e4SLinus Torvalds 		}
11031da177e4SLinus Torvalds 		end_page_writeback(page);
11041da177e4SLinus Torvalds 
11051da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
11061da177e4SLinus Torvalds 		if (data->args.stable != NFS_UNSTABLE || data->verf.committed == NFS_FILE_SYNC) {
11071da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
11081da177e4SLinus Torvalds 			dprintk(" OK\n");
11091da177e4SLinus Torvalds 			goto next;
11101da177e4SLinus Torvalds 		}
11111da177e4SLinus Torvalds 		memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
11121da177e4SLinus Torvalds 		nfs_mark_request_commit(req);
11131da177e4SLinus Torvalds 		dprintk(" marked for commit\n");
11141da177e4SLinus Torvalds #else
11151da177e4SLinus Torvalds 		nfs_inode_remove_request(req);
11161da177e4SLinus Torvalds #endif
11171da177e4SLinus Torvalds 	next:
1118c6a556b8STrond Myklebust 		nfs_clear_page_writeback(req);
11191da177e4SLinus Torvalds 	}
11201da177e4SLinus Torvalds }
11211da177e4SLinus Torvalds 
1122788e7a89STrond Myklebust static const struct rpc_call_ops nfs_write_full_ops = {
1123788e7a89STrond Myklebust 	.rpc_call_done = nfs_writeback_done_full,
1124788e7a89STrond Myklebust 	.rpc_release = nfs_writedata_release,
1125788e7a89STrond Myklebust };
1126788e7a89STrond Myklebust 
1127788e7a89STrond Myklebust 
11281da177e4SLinus Torvalds /*
11291da177e4SLinus Torvalds  * This function is called when the WRITE call is complete.
11301da177e4SLinus Torvalds  */
1131462d5b32SChuck Lever int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
11321da177e4SLinus Torvalds {
11331da177e4SLinus Torvalds 	struct nfs_writeargs	*argp = &data->args;
11341da177e4SLinus Torvalds 	struct nfs_writeres	*resp = &data->res;
1135788e7a89STrond Myklebust 	int status;
11361da177e4SLinus Torvalds 
11371da177e4SLinus Torvalds 	dprintk("NFS: %4d nfs_writeback_done (status %d)\n",
11381da177e4SLinus Torvalds 		task->tk_pid, task->tk_status);
11391da177e4SLinus Torvalds 
1140f551e44fSChuck Lever 	/*
1141f551e44fSChuck Lever 	 * ->write_done will attempt to use post-op attributes to detect
1142f551e44fSChuck Lever 	 * conflicting writes by other clients.  A strict interpretation
1143f551e44fSChuck Lever 	 * of close-to-open would allow us to continue caching even if
1144f551e44fSChuck Lever 	 * another writer had changed the file, but some applications
1145f551e44fSChuck Lever 	 * depend on tighter cache coherency when writing.
1146f551e44fSChuck Lever 	 */
1147788e7a89STrond Myklebust 	status = NFS_PROTO(data->inode)->write_done(task, data);
1148788e7a89STrond Myklebust 	if (status != 0)
1149788e7a89STrond Myklebust 		return status;
115091d5b470SChuck Lever 	nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
115191d5b470SChuck Lever 
11521da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
11531da177e4SLinus Torvalds 	if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
11541da177e4SLinus Torvalds 		/* We tried a write call, but the server did not
11551da177e4SLinus Torvalds 		 * commit data to stable storage even though we
11561da177e4SLinus Torvalds 		 * requested it.
11571da177e4SLinus Torvalds 		 * Note: There is a known bug in Tru64 < 5.0 in which
11581da177e4SLinus Torvalds 		 *	 the server reports NFS_DATA_SYNC, but performs
11591da177e4SLinus Torvalds 		 *	 NFS_FILE_SYNC. We therefore implement this checking
11601da177e4SLinus Torvalds 		 *	 as a dprintk() in order to avoid filling syslog.
11611da177e4SLinus Torvalds 		 */
11621da177e4SLinus Torvalds 		static unsigned long    complain;
11631da177e4SLinus Torvalds 
11641da177e4SLinus Torvalds 		if (time_before(complain, jiffies)) {
11651da177e4SLinus Torvalds 			dprintk("NFS: faulty NFS server %s:"
11661da177e4SLinus Torvalds 				" (committed = %d) != (stable = %d)\n",
116754ceac45SDavid Howells 				NFS_SERVER(data->inode)->nfs_client->cl_hostname,
11681da177e4SLinus Torvalds 				resp->verf->committed, argp->stable);
11691da177e4SLinus Torvalds 			complain = jiffies + 300 * HZ;
11701da177e4SLinus Torvalds 		}
11711da177e4SLinus Torvalds 	}
11721da177e4SLinus Torvalds #endif
11731da177e4SLinus Torvalds 	/* Is this a short write? */
11741da177e4SLinus Torvalds 	if (task->tk_status >= 0 && resp->count < argp->count) {
11751da177e4SLinus Torvalds 		static unsigned long    complain;
11761da177e4SLinus Torvalds 
117791d5b470SChuck Lever 		nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
117891d5b470SChuck Lever 
11791da177e4SLinus Torvalds 		/* Has the server at least made some progress? */
11801da177e4SLinus Torvalds 		if (resp->count != 0) {
11811da177e4SLinus Torvalds 			/* Was this an NFSv2 write or an NFSv3 stable write? */
11821da177e4SLinus Torvalds 			if (resp->verf->committed != NFS_UNSTABLE) {
11831da177e4SLinus Torvalds 				/* Resend from where the server left off */
11841da177e4SLinus Torvalds 				argp->offset += resp->count;
11851da177e4SLinus Torvalds 				argp->pgbase += resp->count;
11861da177e4SLinus Torvalds 				argp->count -= resp->count;
11871da177e4SLinus Torvalds 			} else {
11881da177e4SLinus Torvalds 				/* Resend as a stable write in order to avoid
11891da177e4SLinus Torvalds 				 * headaches in the case of a server crash.
11901da177e4SLinus Torvalds 				 */
11911da177e4SLinus Torvalds 				argp->stable = NFS_FILE_SYNC;
11921da177e4SLinus Torvalds 			}
11931da177e4SLinus Torvalds 			rpc_restart_call(task);
1194788e7a89STrond Myklebust 			return -EAGAIN;
11951da177e4SLinus Torvalds 		}
11961da177e4SLinus Torvalds 		if (time_before(complain, jiffies)) {
11971da177e4SLinus Torvalds 			printk(KERN_WARNING
11981da177e4SLinus Torvalds 			       "NFS: Server wrote zero bytes, expected %u.\n",
11991da177e4SLinus Torvalds 					argp->count);
12001da177e4SLinus Torvalds 			complain = jiffies + 300 * HZ;
12011da177e4SLinus Torvalds 		}
12021da177e4SLinus Torvalds 		/* Can't do anything about it except throw an error. */
12031da177e4SLinus Torvalds 		task->tk_status = -EIO;
12041da177e4SLinus Torvalds 	}
1205788e7a89STrond Myklebust 	return 0;
12061da177e4SLinus Torvalds }
12071da177e4SLinus Torvalds 
12081da177e4SLinus Torvalds 
12091da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1210963d8fe5STrond Myklebust void nfs_commit_release(void *wdata)
12111da177e4SLinus Torvalds {
12121da177e4SLinus Torvalds 	nfs_commit_free(wdata);
12131da177e4SLinus Torvalds }
12141da177e4SLinus Torvalds 
12151da177e4SLinus Torvalds /*
12161da177e4SLinus Torvalds  * Set up the argument/result storage required for the RPC call.
12171da177e4SLinus Torvalds  */
12181da177e4SLinus Torvalds static void nfs_commit_rpcsetup(struct list_head *head,
1219788e7a89STrond Myklebust 		struct nfs_write_data *data,
1220788e7a89STrond Myklebust 		int how)
12211da177e4SLinus Torvalds {
12223da28eb1STrond Myklebust 	struct nfs_page		*first;
12231da177e4SLinus Torvalds 	struct inode		*inode;
1224788e7a89STrond Myklebust 	int flags;
12251da177e4SLinus Torvalds 
12261da177e4SLinus Torvalds 	/* Set up the RPC argument and reply structs
12271da177e4SLinus Torvalds 	 * NB: take care not to mess about with data->commit et al. */
12281da177e4SLinus Torvalds 
12291da177e4SLinus Torvalds 	list_splice_init(head, &data->pages);
12301da177e4SLinus Torvalds 	first = nfs_list_entry(data->pages.next);
12311da177e4SLinus Torvalds 	inode = first->wb_context->dentry->d_inode;
12321da177e4SLinus Torvalds 
12331da177e4SLinus Torvalds 	data->inode	  = inode;
12341da177e4SLinus Torvalds 	data->cred	  = first->wb_context->cred;
12351da177e4SLinus Torvalds 
12361da177e4SLinus Torvalds 	data->args.fh     = NFS_FH(data->inode);
12373da28eb1STrond Myklebust 	/* Note: we always request a commit of the entire inode */
12383da28eb1STrond Myklebust 	data->args.offset = 0;
12393da28eb1STrond Myklebust 	data->args.count  = 0;
12403da28eb1STrond Myklebust 	data->res.count   = 0;
12411da177e4SLinus Torvalds 	data->res.fattr   = &data->fattr;
12421da177e4SLinus Torvalds 	data->res.verf    = &data->verf;
12430e574af1STrond Myklebust 	nfs_fattr_init(&data->fattr);
12441da177e4SLinus Torvalds 
1245788e7a89STrond Myklebust 	/* Set up the initial task struct.  */
1246788e7a89STrond Myklebust 	flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
1247788e7a89STrond Myklebust 	rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data);
12481da177e4SLinus Torvalds 	NFS_PROTO(inode)->commit_setup(data, how);
12491da177e4SLinus Torvalds 
12501da177e4SLinus Torvalds 	data->task.tk_priority = flush_task_priority(how);
12511da177e4SLinus Torvalds 	data->task.tk_cookie = (unsigned long)inode;
12521da177e4SLinus Torvalds 
12530bbacc40SChuck Lever 	dprintk("NFS: %4d initiated commit call\n", data->task.tk_pid);
12541da177e4SLinus Torvalds }
12551da177e4SLinus Torvalds 
12561da177e4SLinus Torvalds /*
12571da177e4SLinus Torvalds  * Commit dirty pages
12581da177e4SLinus Torvalds  */
12591da177e4SLinus Torvalds static int
126040859d7eSChuck Lever nfs_commit_list(struct inode *inode, struct list_head *head, int how)
12611da177e4SLinus Torvalds {
12621da177e4SLinus Torvalds 	struct nfs_write_data	*data;
12631da177e4SLinus Torvalds 	struct nfs_page         *req;
12641da177e4SLinus Torvalds 
1265e9f7bee1STrond Myklebust 	data = nfs_commit_alloc();
12661da177e4SLinus Torvalds 
12671da177e4SLinus Torvalds 	if (!data)
12681da177e4SLinus Torvalds 		goto out_bad;
12691da177e4SLinus Torvalds 
12701da177e4SLinus Torvalds 	/* Set up the argument struct */
12711da177e4SLinus Torvalds 	nfs_commit_rpcsetup(head, data, how);
12721da177e4SLinus Torvalds 
12731da177e4SLinus Torvalds 	nfs_execute_write(data);
12741da177e4SLinus Torvalds 	return 0;
12751da177e4SLinus Torvalds  out_bad:
12761da177e4SLinus Torvalds 	while (!list_empty(head)) {
12771da177e4SLinus Torvalds 		req = nfs_list_entry(head->next);
12781da177e4SLinus Torvalds 		nfs_list_remove_request(req);
12791da177e4SLinus Torvalds 		nfs_mark_request_commit(req);
128083715ad5STrond Myklebust 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
12815c2d97cbSTrond Myklebust 		nfs_clear_page_writeback(req);
12821da177e4SLinus Torvalds 	}
12831da177e4SLinus Torvalds 	return -ENOMEM;
12841da177e4SLinus Torvalds }
12851da177e4SLinus Torvalds 
12861da177e4SLinus Torvalds /*
12871da177e4SLinus Torvalds  * COMMIT call returned
12881da177e4SLinus Torvalds  */
1289788e7a89STrond Myklebust static void nfs_commit_done(struct rpc_task *task, void *calldata)
12901da177e4SLinus Torvalds {
1291963d8fe5STrond Myklebust 	struct nfs_write_data	*data = calldata;
12921da177e4SLinus Torvalds 	struct nfs_page		*req;
12931da177e4SLinus Torvalds 
12941da177e4SLinus Torvalds         dprintk("NFS: %4d nfs_commit_done (status %d)\n",
12951da177e4SLinus Torvalds                                 task->tk_pid, task->tk_status);
12961da177e4SLinus Torvalds 
1297788e7a89STrond Myklebust 	/* Call the NFS version-specific code */
1298788e7a89STrond Myklebust 	if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1299788e7a89STrond Myklebust 		return;
1300788e7a89STrond Myklebust 
13011da177e4SLinus Torvalds 	while (!list_empty(&data->pages)) {
13021da177e4SLinus Torvalds 		req = nfs_list_entry(data->pages.next);
13031da177e4SLinus Torvalds 		nfs_list_remove_request(req);
1304fd39fc85SChristoph Lameter 		dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
13051da177e4SLinus Torvalds 
13061da177e4SLinus Torvalds 		dprintk("NFS: commit (%s/%Ld %d@%Ld)",
13071da177e4SLinus Torvalds 			req->wb_context->dentry->d_inode->i_sb->s_id,
13081da177e4SLinus Torvalds 			(long long)NFS_FILEID(req->wb_context->dentry->d_inode),
13091da177e4SLinus Torvalds 			req->wb_bytes,
13101da177e4SLinus Torvalds 			(long long)req_offset(req));
13111da177e4SLinus Torvalds 		if (task->tk_status < 0) {
13121da177e4SLinus Torvalds 			req->wb_context->error = task->tk_status;
13131da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
13141da177e4SLinus Torvalds 			dprintk(", error = %d\n", task->tk_status);
13151da177e4SLinus Torvalds 			goto next;
13161da177e4SLinus Torvalds 		}
13171da177e4SLinus Torvalds 
13181da177e4SLinus Torvalds 		/* Okay, COMMIT succeeded, apparently. Check the verifier
13191da177e4SLinus Torvalds 		 * returned by the server against all stored verfs. */
13201da177e4SLinus Torvalds 		if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
13211da177e4SLinus Torvalds 			/* We have a match */
13221da177e4SLinus Torvalds 			nfs_inode_remove_request(req);
13231da177e4SLinus Torvalds 			dprintk(" OK\n");
13241da177e4SLinus Torvalds 			goto next;
13251da177e4SLinus Torvalds 		}
13261da177e4SLinus Torvalds 		/* We have a mismatch. Write the page again */
13271da177e4SLinus Torvalds 		dprintk(" mismatch\n");
132861822ab5STrond Myklebust 		nfs_redirty_request(req);
13291da177e4SLinus Torvalds 	next:
1330c6a556b8STrond Myklebust 		nfs_clear_page_writeback(req);
13311da177e4SLinus Torvalds 	}
13321da177e4SLinus Torvalds }
1333788e7a89STrond Myklebust 
1334788e7a89STrond Myklebust static const struct rpc_call_ops nfs_commit_ops = {
1335788e7a89STrond Myklebust 	.rpc_call_done = nfs_commit_done,
1336788e7a89STrond Myklebust 	.rpc_release = nfs_commit_release,
1337788e7a89STrond Myklebust };
1338c42de9ddSTrond Myklebust #else
1339c42de9ddSTrond Myklebust static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1340c42de9ddSTrond Myklebust {
1341c42de9ddSTrond Myklebust 	return 0;
1342c42de9ddSTrond Myklebust }
13431da177e4SLinus Torvalds #endif
13441da177e4SLinus Torvalds 
13453f442547STrond Myklebust static long nfs_flush_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
13461da177e4SLinus Torvalds {
134728c6925fSTrond Myklebust 	struct nfs_inode *nfsi = NFS_I(mapping->host);
13481da177e4SLinus Torvalds 	LIST_HEAD(head);
13493f442547STrond Myklebust 	long res;
13501da177e4SLinus Torvalds 
13511da177e4SLinus Torvalds 	spin_lock(&nfsi->req_lock);
13523f442547STrond Myklebust 	res = nfs_scan_dirty(mapping, wbc, &head);
13531da177e4SLinus Torvalds 	spin_unlock(&nfsi->req_lock);
1354ab0a3dbeSTrond Myklebust 	if (res) {
135528c6925fSTrond Myklebust 		int error = nfs_flush_list(mapping->host, &head, res, how);
13561da177e4SLinus Torvalds 		if (error < 0)
13571da177e4SLinus Torvalds 			return error;
13587d46a49fSTrond Myklebust 	}
13591da177e4SLinus Torvalds 	return res;
13601da177e4SLinus Torvalds }
13611da177e4SLinus Torvalds 
13621da177e4SLinus Torvalds #if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
13633da28eb1STrond Myklebust int nfs_commit_inode(struct inode *inode, int how)
13641da177e4SLinus Torvalds {
13651da177e4SLinus Torvalds 	struct nfs_inode *nfsi = NFS_I(inode);
13661da177e4SLinus Torvalds 	LIST_HEAD(head);
13677d46a49fSTrond Myklebust 	int res;
13681da177e4SLinus Torvalds 
13691da177e4SLinus Torvalds 	spin_lock(&nfsi->req_lock);
13703da28eb1STrond Myklebust 	res = nfs_scan_commit(inode, &head, 0, 0);
13713da28eb1STrond Myklebust 	spin_unlock(&nfsi->req_lock);
13721da177e4SLinus Torvalds 	if (res) {
13737d46a49fSTrond Myklebust 		int error = nfs_commit_list(inode, &head, how);
13741da177e4SLinus Torvalds 		if (error < 0)
13751da177e4SLinus Torvalds 			return error;
13763da28eb1STrond Myklebust 	}
13771da177e4SLinus Torvalds 	return res;
13781da177e4SLinus Torvalds }
13791da177e4SLinus Torvalds #endif
13801da177e4SLinus Torvalds 
13811c75950bSTrond Myklebust long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
13821da177e4SLinus Torvalds {
13831c75950bSTrond Myklebust 	struct inode *inode = mapping->host;
1384c42de9ddSTrond Myklebust 	struct nfs_inode *nfsi = NFS_I(inode);
13851c75950bSTrond Myklebust 	unsigned long idx_start, idx_end;
13861c75950bSTrond Myklebust 	unsigned int npages = 0;
1387c42de9ddSTrond Myklebust 	LIST_HEAD(head);
138870b9ecbdSTrond Myklebust 	int nocommit = how & FLUSH_NOCOMMIT;
13893f442547STrond Myklebust 	long pages, ret;
13901da177e4SLinus Torvalds 
13911c75950bSTrond Myklebust 	/* FIXME */
13921c75950bSTrond Myklebust 	if (wbc->range_cyclic)
13931c75950bSTrond Myklebust 		idx_start = 0;
13941c75950bSTrond Myklebust 	else {
13951c75950bSTrond Myklebust 		idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
13961c75950bSTrond Myklebust 		idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
13971c75950bSTrond Myklebust 		if (idx_end > idx_start) {
13981c75950bSTrond Myklebust 			unsigned long l_npages = 1 + idx_end - idx_start;
13991c75950bSTrond Myklebust 			npages = l_npages;
14001c75950bSTrond Myklebust 			if (sizeof(npages) != sizeof(l_npages) &&
14011c75950bSTrond Myklebust 					(unsigned long)npages != l_npages)
14021c75950bSTrond Myklebust 				npages = 0;
14031c75950bSTrond Myklebust 		}
14041c75950bSTrond Myklebust 	}
1405c42de9ddSTrond Myklebust 	how &= ~FLUSH_NOCOMMIT;
1406c42de9ddSTrond Myklebust 	spin_lock(&nfsi->req_lock);
14071da177e4SLinus Torvalds 	do {
14081c75950bSTrond Myklebust 		wbc->pages_skipped = 0;
1409c42de9ddSTrond Myklebust 		ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
1410c42de9ddSTrond Myklebust 		if (ret != 0)
1411c42de9ddSTrond Myklebust 			continue;
14121c75950bSTrond Myklebust 		pages = nfs_scan_dirty(mapping, wbc, &head);
1413c42de9ddSTrond Myklebust 		if (pages != 0) {
1414c42de9ddSTrond Myklebust 			spin_unlock(&nfsi->req_lock);
1415e8e058e8STrond Myklebust 			if (how & FLUSH_INVALIDATE) {
141683715ad5STrond Myklebust 				nfs_cancel_dirty_list(&head);
1417e8e058e8STrond Myklebust 				ret = pages;
1418e8e058e8STrond Myklebust 			} else
1419c42de9ddSTrond Myklebust 				ret = nfs_flush_list(inode, &head, pages, how);
1420c42de9ddSTrond Myklebust 			spin_lock(&nfsi->req_lock);
142170b9ecbdSTrond Myklebust 			continue;
142270b9ecbdSTrond Myklebust 		}
14231c75950bSTrond Myklebust 		if (wbc->pages_skipped != 0)
14241c75950bSTrond Myklebust 			continue;
1425c42de9ddSTrond Myklebust 		if (nocommit)
1426c42de9ddSTrond Myklebust 			break;
1427d2ccddf0STrond Myklebust 		pages = nfs_scan_commit(inode, &head, idx_start, npages);
14281c75950bSTrond Myklebust 		if (pages == 0) {
14291c75950bSTrond Myklebust 			if (wbc->pages_skipped != 0)
14301c75950bSTrond Myklebust 				continue;
1431c42de9ddSTrond Myklebust 			break;
14321c75950bSTrond Myklebust 		}
1433d2ccddf0STrond Myklebust 		if (how & FLUSH_INVALIDATE) {
1434d2ccddf0STrond Myklebust 			spin_unlock(&nfsi->req_lock);
143583715ad5STrond Myklebust 			nfs_cancel_commit_list(&head);
1436e8e058e8STrond Myklebust 			ret = pages;
1437d2ccddf0STrond Myklebust 			spin_lock(&nfsi->req_lock);
1438d2ccddf0STrond Myklebust 			continue;
1439d2ccddf0STrond Myklebust 		}
1440d2ccddf0STrond Myklebust 		pages += nfs_scan_commit(inode, &head, 0, 0);
1441c42de9ddSTrond Myklebust 		spin_unlock(&nfsi->req_lock);
1442c42de9ddSTrond Myklebust 		ret = nfs_commit_list(inode, &head, how);
1443c42de9ddSTrond Myklebust 		spin_lock(&nfsi->req_lock);
1444c42de9ddSTrond Myklebust 	} while (ret >= 0);
1445c42de9ddSTrond Myklebust 	spin_unlock(&nfsi->req_lock);
1446c42de9ddSTrond Myklebust 	return ret;
14471da177e4SLinus Torvalds }
14481da177e4SLinus Torvalds 
14491c75950bSTrond Myklebust /*
14501c75950bSTrond Myklebust  * flush the inode to disk.
14511c75950bSTrond Myklebust  */
14521c75950bSTrond Myklebust int nfs_wb_all(struct inode *inode)
14531c75950bSTrond Myklebust {
14541c75950bSTrond Myklebust 	struct address_space *mapping = inode->i_mapping;
14551c75950bSTrond Myklebust 	struct writeback_control wbc = {
14561c75950bSTrond Myklebust 		.bdi = mapping->backing_dev_info,
14571c75950bSTrond Myklebust 		.sync_mode = WB_SYNC_ALL,
14581c75950bSTrond Myklebust 		.nr_to_write = LONG_MAX,
145961822ab5STrond Myklebust 		.for_writepages = 1,
14601c75950bSTrond Myklebust 		.range_cyclic = 1,
14611c75950bSTrond Myklebust 	};
14621c75950bSTrond Myklebust 	int ret;
14631c75950bSTrond Myklebust 
146461822ab5STrond Myklebust 	ret = generic_writepages(mapping, &wbc);
146561822ab5STrond Myklebust 	if (ret < 0)
146661822ab5STrond Myklebust 		goto out;
14671c75950bSTrond Myklebust 	ret = nfs_sync_mapping_wait(mapping, &wbc, 0);
14681c75950bSTrond Myklebust 	if (ret >= 0)
14691c75950bSTrond Myklebust 		return 0;
147061822ab5STrond Myklebust out:
1471e507d9ebSTrond Myklebust 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
14721c75950bSTrond Myklebust 	return ret;
14731c75950bSTrond Myklebust }
14741c75950bSTrond Myklebust 
14751c75950bSTrond Myklebust int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how)
14761c75950bSTrond Myklebust {
14771c75950bSTrond Myklebust 	struct writeback_control wbc = {
14781c75950bSTrond Myklebust 		.bdi = mapping->backing_dev_info,
14791c75950bSTrond Myklebust 		.sync_mode = WB_SYNC_ALL,
14801c75950bSTrond Myklebust 		.nr_to_write = LONG_MAX,
14811c75950bSTrond Myklebust 		.range_start = range_start,
14821c75950bSTrond Myklebust 		.range_end = range_end,
148361822ab5STrond Myklebust 		.for_writepages = 1,
14841c75950bSTrond Myklebust 	};
14851c75950bSTrond Myklebust 	int ret;
14861c75950bSTrond Myklebust 
148761822ab5STrond Myklebust 	if (!(how & FLUSH_NOWRITEPAGE)) {
148861822ab5STrond Myklebust 		ret = generic_writepages(mapping, &wbc);
148961822ab5STrond Myklebust 		if (ret < 0)
149061822ab5STrond Myklebust 			goto out;
149161822ab5STrond Myklebust 	}
14921c75950bSTrond Myklebust 	ret = nfs_sync_mapping_wait(mapping, &wbc, how);
14931c75950bSTrond Myklebust 	if (ret >= 0)
14941c75950bSTrond Myklebust 		return 0;
149561822ab5STrond Myklebust out:
1496e507d9ebSTrond Myklebust 	__mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
14971c75950bSTrond Myklebust 	return ret;
14981c75950bSTrond Myklebust }
14991c75950bSTrond Myklebust 
150061822ab5STrond Myklebust int nfs_wb_page_priority(struct inode *inode, struct page *page, int how)
15011c75950bSTrond Myklebust {
15021c75950bSTrond Myklebust 	loff_t range_start = page_offset(page);
15031c75950bSTrond Myklebust 	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
15044d770ccfSTrond Myklebust 	struct writeback_control wbc = {
15054d770ccfSTrond Myklebust 		.bdi = page->mapping->backing_dev_info,
15064d770ccfSTrond Myklebust 		.sync_mode = WB_SYNC_ALL,
15074d770ccfSTrond Myklebust 		.nr_to_write = LONG_MAX,
15084d770ccfSTrond Myklebust 		.range_start = range_start,
15094d770ccfSTrond Myklebust 		.range_end = range_end,
15104d770ccfSTrond Myklebust 	};
15114d770ccfSTrond Myklebust 	int ret;
15121c75950bSTrond Myklebust 
15134d770ccfSTrond Myklebust 	BUG_ON(!PageLocked(page));
15144d770ccfSTrond Myklebust 	if (!(how & FLUSH_NOWRITEPAGE) && clear_page_dirty_for_io(page)) {
15154d770ccfSTrond Myklebust 		ret = nfs_writepage_locked(page, &wbc);
15164d770ccfSTrond Myklebust 		if (ret < 0)
15174d770ccfSTrond Myklebust 			goto out;
15184d770ccfSTrond Myklebust 	}
1519f40313acSTrond Myklebust 	if (!PagePrivate(page))
1520f40313acSTrond Myklebust 		return 0;
15214d770ccfSTrond Myklebust 	ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
15224d770ccfSTrond Myklebust 	if (ret >= 0)
15234d770ccfSTrond Myklebust 		return 0;
15244d770ccfSTrond Myklebust out:
1525e507d9ebSTrond Myklebust 	__mark_inode_dirty(inode, I_DIRTY_PAGES);
15264d770ccfSTrond Myklebust 	return ret;
15271c75950bSTrond Myklebust }
15281c75950bSTrond Myklebust 
15291c75950bSTrond Myklebust /*
15301c75950bSTrond Myklebust  * Write back all requests on one page - we do this before reading it.
15311c75950bSTrond Myklebust  */
15321c75950bSTrond Myklebust int nfs_wb_page(struct inode *inode, struct page* page)
15331c75950bSTrond Myklebust {
15344d770ccfSTrond Myklebust 	return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
15351c75950bSTrond Myklebust }
15361c75950bSTrond Myklebust 
15371a54533eSTrond Myklebust int nfs_set_page_dirty(struct page *page)
15381a54533eSTrond Myklebust {
15391a54533eSTrond Myklebust 	struct nfs_page *req;
15401a54533eSTrond Myklebust 
15411a54533eSTrond Myklebust 	req = nfs_page_find_request(page);
15421a54533eSTrond Myklebust 	if (req != NULL) {
15431a54533eSTrond Myklebust 		/* Mark any existing write requests for flushing */
15441a54533eSTrond Myklebust 		set_bit(PG_NEED_FLUSH, &req->wb_flags);
15451a54533eSTrond Myklebust 		nfs_release_request(req);
15461a54533eSTrond Myklebust 	}
15471a54533eSTrond Myklebust 	return __set_page_dirty_nobuffers(page);
15481a54533eSTrond Myklebust }
15491a54533eSTrond Myklebust 
15501c75950bSTrond Myklebust 
1551f7b422b1SDavid Howells int __init nfs_init_writepagecache(void)
15521da177e4SLinus Torvalds {
15531da177e4SLinus Torvalds 	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
15541da177e4SLinus Torvalds 					     sizeof(struct nfs_write_data),
15551da177e4SLinus Torvalds 					     0, SLAB_HWCACHE_ALIGN,
15561da177e4SLinus Torvalds 					     NULL, NULL);
15571da177e4SLinus Torvalds 	if (nfs_wdata_cachep == NULL)
15581da177e4SLinus Torvalds 		return -ENOMEM;
15591da177e4SLinus Torvalds 
156093d2341cSMatthew Dobson 	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
15611da177e4SLinus Torvalds 						     nfs_wdata_cachep);
15621da177e4SLinus Torvalds 	if (nfs_wdata_mempool == NULL)
15631da177e4SLinus Torvalds 		return -ENOMEM;
15641da177e4SLinus Torvalds 
156593d2341cSMatthew Dobson 	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
15661da177e4SLinus Torvalds 						      nfs_wdata_cachep);
15671da177e4SLinus Torvalds 	if (nfs_commit_mempool == NULL)
15681da177e4SLinus Torvalds 		return -ENOMEM;
15691da177e4SLinus Torvalds 
15701da177e4SLinus Torvalds 	return 0;
15711da177e4SLinus Torvalds }
15721da177e4SLinus Torvalds 
1573266bee88SDavid Brownell void nfs_destroy_writepagecache(void)
15741da177e4SLinus Torvalds {
15751da177e4SLinus Torvalds 	mempool_destroy(nfs_commit_mempool);
15761da177e4SLinus Torvalds 	mempool_destroy(nfs_wdata_mempool);
15771a1d92c1SAlexey Dobriyan 	kmem_cache_destroy(nfs_wdata_cachep);
15781da177e4SLinus Torvalds }
15791da177e4SLinus Torvalds 
1580