blob: 89527a487ed7d5365464bd6fe537c9014cfce117 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/write.c
3 *
Trond Myklebust7c85d902006-12-13 15:23:48 -05004 * Write file data over NFS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
7 */
8
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/writeback.h>
Peter Zijlstra89a09142007-03-16 13:38:26 -080015#include <linux/swap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016
17#include <linux/sunrpc/clnt.h>
18#include <linux/nfs_fs.h>
19#include <linux/nfs_mount.h>
20#include <linux/nfs_page.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070021#include <linux/backing-dev.h>
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024
25#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050026#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050027#include "iostat.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#define NFSDBG_FACILITY NFSDBG_PAGECACHE
30
31#define MIN_POOL_WRITE (32)
32#define MIN_POOL_COMMIT (4)
33
34/*
35 * Local function declarations
36 */
37static struct nfs_page * nfs_update_request(struct nfs_open_context*,
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 struct page *,
39 unsigned int, unsigned int);
Trond Myklebustc63c7b02007-04-02 19:29:52 -040040static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
41 struct inode *inode, int ioflags);
Trond Myklebust788e7a82006-03-20 13:44:27 -050042static const struct rpc_call_ops nfs_write_partial_ops;
43static const struct rpc_call_ops nfs_write_full_ops;
44static const struct rpc_call_ops nfs_commit_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Christoph Lametere18b8902006-12-06 20:33:20 -080046static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050047static mempool_t *nfs_wdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048static mempool_t *nfs_commit_mempool;
49
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070050struct nfs_write_data *nfs_commit_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070051{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080052 struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
Chuck Lever40859d72005-11-30 18:09:02 -050053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 if (p) {
55 memset(p, 0, sizeof(*p));
56 INIT_LIST_HEAD(&p->pages);
57 }
58 return p;
59}
60
Trond Myklebust10afec92007-05-14 17:16:04 -040061static void nfs_commit_rcu_free(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062{
Trond Myklebust8aca67f2006-11-13 16:23:44 -050063 struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
Chuck Lever40859d72005-11-30 18:09:02 -050064 if (p && (p->pagevec != &p->page_array[0]))
65 kfree(p->pagevec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 mempool_free(p, nfs_commit_mempool);
67}
68
Trond Myklebust8aca67f2006-11-13 16:23:44 -050069void nfs_commit_free(struct nfs_write_data *wdata)
70{
71 call_rcu_bh(&wdata->task.u.tk_rcu, nfs_commit_rcu_free);
72}
73
Trond Myklebust8d5658c2007-04-10 09:26:35 -040074struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050075{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080076 struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050077
78 if (p) {
79 memset(p, 0, sizeof(*p));
80 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070081 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040082 if (pagecount <= ARRAY_SIZE(p->page_array))
83 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050084 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040085 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
86 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050087 mempool_free(p, nfs_wdata_mempool);
88 p = NULL;
89 }
90 }
91 }
92 return p;
93}
94
Trond Myklebust8aca67f2006-11-13 16:23:44 -050095static void nfs_writedata_rcu_free(struct rcu_head *head)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050096{
Trond Myklebust8aca67f2006-11-13 16:23:44 -050097 struct nfs_write_data *p = container_of(head, struct nfs_write_data, task.u.tk_rcu);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050098 if (p && (p->pagevec != &p->page_array[0]))
99 kfree(p->pagevec);
100 mempool_free(p, nfs_wdata_mempool);
101}
102
Trond Myklebust8aca67f2006-11-13 16:23:44 -0500103static void nfs_writedata_free(struct nfs_write_data *wdata)
104{
105 call_rcu_bh(&wdata->task.u.tk_rcu, nfs_writedata_rcu_free);
106}
107
Trond Myklebust963d8fe2006-01-03 09:55:04 +0100108void nfs_writedata_release(void *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 nfs_writedata_free(wdata);
111}
112
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400113static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
114{
115 ctx->error = error;
116 smp_wmb();
117 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
118}
119
Trond Myklebust277459d2006-12-05 00:35:35 -0500120static struct nfs_page *nfs_page_find_request_locked(struct page *page)
121{
122 struct nfs_page *req = NULL;
123
124 if (PagePrivate(page)) {
125 req = (struct nfs_page *)page_private(page);
126 if (req != NULL)
Trond Myklebustc03b4022007-06-17 13:26:38 -0400127 kref_get(&req->wb_kref);
Trond Myklebust277459d2006-12-05 00:35:35 -0500128 }
129 return req;
130}
131
132static struct nfs_page *nfs_page_find_request(struct page *page)
133{
Trond Myklebust587142f2007-07-02 09:57:54 -0400134 struct inode *inode = page->mapping->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500135 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500136
Trond Myklebust587142f2007-07-02 09:57:54 -0400137 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500138 req = nfs_page_find_request_locked(page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400139 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500140 return req;
141}
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143/* Adjust the file length if we're writing beyond the end */
144static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
145{
146 struct inode *inode = page->mapping->host;
147 loff_t end, i_size = i_size_read(inode);
Trond Myklebustca52fec2007-04-17 17:22:13 -0400148 pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150 if (i_size > 0 && page->index < end_index)
151 return;
152 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
153 if (i_size >= end)
154 return;
Chuck Lever91d5b472006-03-20 13:44:14 -0500155 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 i_size_write(inode, end);
157}
158
Trond Myklebusta301b772007-02-06 11:07:15 -0800159/* A writeback failed: mark the page as bad, and invalidate the page cache */
160static void nfs_set_pageerror(struct page *page)
161{
162 SetPageError(page);
163 nfs_zap_mapping(page->mapping->host, page->mapping);
164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166/* We can set the PG_uptodate flag if we see that a write request
167 * covers the full page.
168 */
169static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
170{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 if (PageUptodate(page))
172 return;
173 if (base != 0)
174 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500175 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500177 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
Trond Myklebuste21195a2006-12-05 00:35:39 -0500180static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 unsigned int offset, unsigned int count)
182{
183 struct nfs_page *req;
Trond Myklebuste21195a2006-12-05 00:35:39 -0500184 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Trond Myklebuste21195a2006-12-05 00:35:39 -0500186 for (;;) {
187 req = nfs_update_request(ctx, page, offset, count);
188 if (!IS_ERR(req))
189 break;
190 ret = PTR_ERR(req);
191 if (ret != -EBUSY)
192 return ret;
193 ret = nfs_wb_page(page->mapping->host, page);
194 if (ret != 0)
195 return ret;
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 /* Update file length */
198 nfs_grow_file(page, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 nfs_unlock_request(req);
Trond Myklebustabd3e642006-01-03 09:55:02 +0100200 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
203static int wb_priority(struct writeback_control *wbc)
204{
205 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400206 return FLUSH_HIGHPRI | FLUSH_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (wbc->for_kupdate)
208 return FLUSH_LOWPRI;
209 return 0;
210}
211
212/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800213 * NFS congestion control
214 */
215
216int nfs_congestion_kb;
217
218#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
219#define NFS_CONGESTION_OFF_THRESH \
220 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
221
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400222static int nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800223{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400224 int ret = test_set_page_writeback(page);
225
226 if (!ret) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800227 struct inode *inode = page->mapping->host;
228 struct nfs_server *nfss = NFS_SERVER(inode);
229
Peter Zijlstra277866a2007-05-08 00:35:12 -0700230 if (atomic_long_inc_return(&nfss->writeback) >
Peter Zijlstra89a09142007-03-16 13:38:26 -0800231 NFS_CONGESTION_ON_THRESH)
232 set_bdi_congested(&nfss->backing_dev_info, WRITE);
233 }
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400234 return ret;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800235}
236
237static void nfs_end_page_writeback(struct page *page)
238{
239 struct inode *inode = page->mapping->host;
240 struct nfs_server *nfss = NFS_SERVER(inode);
241
242 end_page_writeback(page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700243 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800244 clear_bdi_congested(&nfss->backing_dev_info, WRITE);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800245}
246
247/*
Trond Myklebuste261f512006-12-05 00:35:41 -0500248 * Find an associated nfs write request, and prepare to flush it out
Trond Myklebust9cccef92007-07-22 17:09:05 -0400249 * May return an error if the user signalled nfs_wait_on_request().
Trond Myklebuste261f512006-12-05 00:35:41 -0500250 */
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400251static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
252 struct page *page)
Trond Myklebuste261f512006-12-05 00:35:41 -0500253{
Trond Myklebust587142f2007-07-02 09:57:54 -0400254 struct inode *inode = page->mapping->host;
255 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste261f512006-12-05 00:35:41 -0500256 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500257 int ret;
258
Trond Myklebust587142f2007-07-02 09:57:54 -0400259 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500260 for(;;) {
261 req = nfs_page_find_request_locked(page);
262 if (req == NULL) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400263 spin_unlock(&inode->i_lock);
Trond Myklebust9cccef92007-07-22 17:09:05 -0400264 return 0;
Trond Myklebuste261f512006-12-05 00:35:41 -0500265 }
266 if (nfs_lock_request_dontget(req))
267 break;
268 /* Note: If we hold the page lock, as is the case in nfs_writepage,
269 * then the call to nfs_lock_request_dontget() will always
270 * succeed provided that someone hasn't already marked the
271 * request as dirty (in which case we don't care).
272 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400273 spin_unlock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500274 ret = nfs_wait_on_request(req);
275 nfs_release_request(req);
276 if (ret != 0)
277 return ret;
Trond Myklebust587142f2007-07-02 09:57:54 -0400278 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500279 }
Trond Myklebust612c9382007-04-20 16:12:45 -0400280 if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
281 /* This request is marked for commit */
Trond Myklebust587142f2007-07-02 09:57:54 -0400282 spin_unlock(&inode->i_lock);
Trond Myklebust612c9382007-04-20 16:12:45 -0400283 nfs_unlock_request(req);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400284 nfs_pageio_complete(pgio);
Trond Myklebust9cccef92007-07-22 17:09:05 -0400285 return 0;
Trond Myklebust612c9382007-04-20 16:12:45 -0400286 }
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400287 if (nfs_set_page_writeback(page) != 0) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400288 spin_unlock(&inode->i_lock);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400289 BUG();
290 }
291 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400292 NFS_PAGE_TAG_LOCKED);
Trond Myklebust587142f2007-07-02 09:57:54 -0400293 spin_unlock(&inode->i_lock);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400294 nfs_pageio_add_request(pgio, req);
Trond Myklebust9cccef92007-07-22 17:09:05 -0400295 return 0;
Trond Myklebuste261f512006-12-05 00:35:41 -0500296}
297
Trond Myklebustf758c8852007-07-22 19:27:32 -0400298static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
299{
300 struct inode *inode = page->mapping->host;
301
302 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
303 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
304
305 nfs_pageio_cond_complete(pgio, page->index);
306 return nfs_page_async_flush(pgio, page);
307}
308
Trond Myklebuste261f512006-12-05 00:35:41 -0500309/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * Write an mmapped page to the server.
311 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500312static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400314 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500315 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Trond Myklebustf758c8852007-07-22 19:27:32 -0400317 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
318 err = nfs_do_writepage(page, wbc, &pgio);
319 nfs_pageio_complete(&pgio);
320 if (err < 0)
321 return err;
322 if (pgio.pg_error < 0)
323 return pgio.pg_error;
324 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500325}
326
327int nfs_writepage(struct page *page, struct writeback_control *wbc)
328{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400329 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500330
Trond Myklebustf758c8852007-07-22 19:27:32 -0400331 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400333 return ret;
334}
335
336static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
337{
338 int ret;
339
340 ret = nfs_do_writepage(page, wbc, data);
341 unlock_page(page);
342 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343}
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
346{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 struct inode *inode = mapping->host;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400348 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 int err;
350
Chuck Lever91d5b472006-03-20 13:44:14 -0500351 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
352
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400353 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
Trond Myklebustf758c8852007-07-22 19:27:32 -0400354 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400355 nfs_pageio_complete(&pgio);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400356 if (err < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return err;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400358 if (pgio.pg_error < 0)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400359 return pgio.pg_error;
360 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
363/*
364 * Insert a write request into an inode
365 */
366static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
367{
368 struct nfs_inode *nfsi = NFS_I(inode);
369 int error;
370
371 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
372 BUG_ON(error == -EEXIST);
373 if (error)
374 return error;
375 if (!nfsi->npages) {
376 igrab(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 if (nfs_have_delegation(inode, FMODE_WRITE))
378 nfsi->change_attr++;
379 }
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500380 SetPagePrivate(req->wb_page);
Trond Myklebust277459d2006-12-05 00:35:35 -0500381 set_page_private(req->wb_page, (unsigned long)req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400383 kref_get(&req->wb_kref);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return 0;
385}
386
387/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800388 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 */
390static void nfs_inode_remove_request(struct nfs_page *req)
391{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400392 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 struct nfs_inode *nfsi = NFS_I(inode);
394
395 BUG_ON (!NFS_WBACK_BUSY(req));
396
Trond Myklebust587142f2007-07-02 09:57:54 -0400397 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500398 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500399 ClearPagePrivate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
401 nfsi->npages--;
402 if (!nfsi->npages) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400403 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 iput(inode);
405 } else
Trond Myklebust587142f2007-07-02 09:57:54 -0400406 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 nfs_clear_request(req);
408 nfs_release_request(req);
409}
410
Trond Myklebust61822ab2006-12-05 00:35:42 -0500411static void
412nfs_redirty_request(struct nfs_page *req)
413{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500414 __set_page_dirty_nobuffers(req->wb_page);
415}
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417/*
418 * Check if a request is dirty
419 */
420static inline int
421nfs_dirty_request(struct nfs_page *req)
422{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400423 struct page *page = req->wb_page;
424
Trond Myklebust612c9382007-04-20 16:12:45 -0400425 if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags))
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400426 return 0;
427 return !PageWriteback(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
429
430#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
431/*
432 * Add a request to the inode's commit list.
433 */
434static void
435nfs_mark_request_commit(struct nfs_page *req)
436{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400437 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 struct nfs_inode *nfsi = NFS_I(inode);
439
Trond Myklebust587142f2007-07-02 09:57:54 -0400440 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 nfsi->ncommit++;
Trond Myklebust612c9382007-04-20 16:12:45 -0400442 set_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Trond Myklebust5c369682007-06-17 15:27:42 -0400443 radix_tree_tag_set(&nfsi->nfs_page_tree,
444 req->wb_index,
445 NFS_PAGE_TAG_COMMIT);
Trond Myklebust587142f2007-07-02 09:57:54 -0400446 spin_unlock(&inode->i_lock);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700447 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700448 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
Trond Myklebusta1803042006-12-05 00:45:19 -0500449 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400451
452static inline
453int nfs_write_need_commit(struct nfs_write_data *data)
454{
455 return data->verf.committed != NFS_FILE_SYNC;
456}
457
458static inline
459int nfs_reschedule_unstable_write(struct nfs_page *req)
460{
Trond Myklebust612c9382007-04-20 16:12:45 -0400461 if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400462 nfs_mark_request_commit(req);
463 return 1;
464 }
465 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
466 nfs_redirty_request(req);
467 return 1;
468 }
469 return 0;
470}
471#else
472static inline void
473nfs_mark_request_commit(struct nfs_page *req)
474{
475}
476
477static inline
478int nfs_write_need_commit(struct nfs_write_data *data)
479{
480 return 0;
481}
482
483static inline
484int nfs_reschedule_unstable_write(struct nfs_page *req)
485{
486 return 0;
487}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488#endif
489
490/*
491 * Wait for a request to complete.
492 *
493 * Interruptible by signals only if mounted with intr flag.
494 */
Trond Myklebustca52fec2007-04-17 17:22:13 -0400495static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
497 struct nfs_inode *nfsi = NFS_I(inode);
498 struct nfs_page *req;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400499 pgoff_t idx_end, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 unsigned int res = 0;
501 int error;
502
503 if (npages == 0)
504 idx_end = ~0;
505 else
506 idx_end = idx_start + npages - 1;
507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 next = idx_start;
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400509 while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_LOCKED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (req->wb_index > idx_end)
511 break;
512
513 next = req->wb_index + 1;
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000514 BUG_ON(!NFS_WBACK_BUSY(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
Trond Myklebustc03b4022007-06-17 13:26:38 -0400516 kref_get(&req->wb_kref);
Trond Myklebust587142f2007-07-02 09:57:54 -0400517 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 error = nfs_wait_on_request(req);
519 nfs_release_request(req);
Trond Myklebust587142f2007-07-02 09:57:54 -0400520 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 if (error < 0)
522 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 res++;
524 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 return res;
526}
527
Trond Myklebust83715ad2006-07-05 13:17:12 -0400528static void nfs_cancel_commit_list(struct list_head *head)
529{
530 struct nfs_page *req;
531
532 while(!list_empty(head)) {
533 req = nfs_list_entry(head->next);
Trond Myklebustb6dff262006-10-19 23:28:38 -0700534 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700535 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
536 BDI_RECLAIMABLE);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400537 nfs_list_remove_request(req);
Trond Myklebust612c9382007-04-20 16:12:45 -0400538 clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400539 nfs_inode_remove_request(req);
Trond Myklebustb6dff262006-10-19 23:28:38 -0700540 nfs_unlock_request(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400541 }
542}
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
545/*
546 * nfs_scan_commit - Scan an inode for commit requests
547 * @inode: NFS inode to scan
548 * @dst: destination list
549 * @idx_start: lower bound of page->index to scan.
550 * @npages: idx_start + npages sets the upper bound to scan.
551 *
552 * Moves requests from the inode's 'commit' request list.
553 * The requests are *not* checked to ensure that they form a contiguous set.
554 */
555static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400556nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557{
558 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000559 int res = 0;
560
561 if (nfsi->ncommit != 0) {
Trond Myklebust5c369682007-06-17 15:27:42 -0400562 res = nfs_scan_list(nfsi, dst, idx_start, npages,
563 NFS_PAGE_TAG_COMMIT);
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000564 nfsi->ncommit -= res;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000565 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 return res;
567}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500568#else
Trond Myklebustca52fec2007-04-17 17:22:13 -0400569static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500570{
571 return 0;
572}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573#endif
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575/*
576 * Try to update any existing write request, or create one if there is none.
577 * In order to match, the request's credentials must match those of
578 * the calling process.
579 *
580 * Note: Should always be called with the Page Lock held!
581 */
582static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
Trond Myklebuste21195a2006-12-05 00:35:39 -0500583 struct page *page, unsigned int offset, unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
Peter Zijlstra89a09142007-03-16 13:38:26 -0800585 struct address_space *mapping = page->mapping;
586 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 struct nfs_page *req, *new = NULL;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400588 pgoff_t rqend, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 end = offset + bytes;
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 for (;;) {
593 /* Loop over all inode entries and see if we find
594 * A request for the page we wish to update
595 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400596 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500597 req = nfs_page_find_request_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 if (req) {
599 if (!nfs_lock_request_dontget(req)) {
600 int error;
Trond Myklebust277459d2006-12-05 00:35:35 -0500601
Trond Myklebust587142f2007-07-02 09:57:54 -0400602 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 error = nfs_wait_on_request(req);
604 nfs_release_request(req);
Neil Brown1dd594b2006-03-20 13:44:04 -0500605 if (error < 0) {
606 if (new)
607 nfs_release_request(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 return ERR_PTR(error);
Neil Brown1dd594b2006-03-20 13:44:04 -0500609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 continue;
611 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400612 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (new)
614 nfs_release_request(new);
615 break;
616 }
617
618 if (new) {
619 int error;
620 nfs_lock_request_dontget(new);
621 error = nfs_inode_add_request(inode, new);
622 if (error) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400623 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 nfs_unlock_request(new);
625 return ERR_PTR(error);
626 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400627 spin_unlock(&inode->i_lock);
Trond Myklebust61e930a2007-10-18 17:08:05 -0400628 req = new;
629 goto zero_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400631 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 new = nfs_create_request(ctx, inode, page, offset, bytes);
634 if (IS_ERR(new))
635 return new;
636 }
637
638 /* We have a request for our page.
639 * If the creds don't match, or the
640 * page addresses don't match,
641 * tell the caller to wait on the conflicting
642 * request.
643 */
644 rqend = req->wb_offset + req->wb_bytes;
645 if (req->wb_context != ctx
646 || req->wb_page != page
647 || !nfs_dirty_request(req)
648 || offset > rqend || end < req->wb_offset) {
649 nfs_unlock_request(req);
650 return ERR_PTR(-EBUSY);
651 }
652
653 /* Okay, the request matches. Update the region */
654 if (offset < req->wb_offset) {
655 req->wb_offset = offset;
656 req->wb_pgbase = offset;
Trond Myklebust61e930a2007-10-18 17:08:05 -0400657 req->wb_bytes = max(end, rqend) - req->wb_offset;
658 goto zero_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 }
660
661 if (end > rqend)
662 req->wb_bytes = end - req->wb_offset;
663
664 return req;
Trond Myklebust61e930a2007-10-18 17:08:05 -0400665zero_page:
666 /* If this page might potentially be marked as up to date,
667 * then we need to zero any uninitalised data. */
668 if (req->wb_pgbase == 0 && req->wb_bytes != PAGE_CACHE_SIZE
669 && !PageUptodate(req->wb_page))
670 zero_user_page(req->wb_page, req->wb_bytes,
671 PAGE_CACHE_SIZE - req->wb_bytes,
672 KM_USER0);
673 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
676int nfs_flush_incompatible(struct file *file, struct page *page)
677{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400678 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500680 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 /*
682 * Look for a request corresponding to this page. If there
683 * is one, and it belongs to another file, we flush it out
684 * before we try to copy anything into the page. Do this
685 * due to the lack of an ACCESS-type call in NFSv2.
686 * Also do the same if we find a request from an existing
687 * dropped page.
688 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500689 do {
690 req = nfs_page_find_request(page);
691 if (req == NULL)
692 return 0;
693 do_flush = req->wb_page != page || req->wb_context != ctx
Trond Myklebuste261f512006-12-05 00:35:41 -0500694 || !nfs_dirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500696 if (!do_flush)
697 return 0;
698 status = nfs_wb_page(page->mapping->host, page);
699 } while (status == 0);
700 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
703/*
704 * Update and possibly write a cached page of an NFS file.
705 *
706 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
707 * things with a page scheduled for an RPC call (e.g. invalidate it).
708 */
709int nfs_updatepage(struct file *file, struct page *page,
710 unsigned int offset, unsigned int count)
711{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400712 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 int status = 0;
715
Chuck Lever91d5b472006-03-20 13:44:14 -0500716 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800719 file->f_path.dentry->d_parent->d_name.name,
720 file->f_path.dentry->d_name.name, count,
Chuck Lever0bbacc42005-11-01 16:53:32 -0500721 (long long)(page_offset(page) +offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 /* If we're not using byte range locks, and we know the page
724 * is entirely in cache, it may be more efficient to avoid
725 * fragmenting write requests.
726 */
Trond Myklebustab0a3db2005-06-22 17:16:30 +0000727 if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500728 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731
Trond Myklebuste21195a2006-12-05 00:35:39 -0500732 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebuste261f512006-12-05 00:35:41 -0500733 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n",
736 status, (long long)i_size_read(inode));
737 if (status < 0)
Trond Myklebusta301b772007-02-06 11:07:15 -0800738 nfs_set_pageerror(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 return status;
740}
741
742static void nfs_writepage_release(struct nfs_page *req)
743{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Trond Myklebust44dd1512007-05-19 11:58:03 -0400745 if (PageError(req->wb_page)) {
746 nfs_end_page_writeback(req->wb_page);
747 nfs_inode_remove_request(req);
748 } else if (!nfs_reschedule_unstable_write(req)) {
749 /* Set the PG_uptodate flag */
750 nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400751 nfs_end_page_writeback(req->wb_page);
752 nfs_inode_remove_request(req);
753 } else
754 nfs_end_page_writeback(req->wb_page);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400755 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756}
757
758static inline int flush_task_priority(int how)
759{
760 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
761 case FLUSH_HIGHPRI:
762 return RPC_PRIORITY_HIGH;
763 case FLUSH_LOWPRI:
764 return RPC_PRIORITY_LOW;
765 }
766 return RPC_PRIORITY_NORMAL;
767}
768
769/*
770 * Set up the argument/result storage required for the RPC call.
771 */
772static void nfs_write_rpcsetup(struct nfs_page *req,
773 struct nfs_write_data *data,
Trond Myklebust788e7a82006-03-20 13:44:27 -0500774 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 unsigned int count, unsigned int offset,
776 int how)
777{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 struct inode *inode;
Trond Myklebust788e7a82006-03-20 13:44:27 -0500779 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 /* Set up the RPC argument and reply structs
782 * NB: take care not to mess about with data->commit et al. */
783
784 data->req = req;
Trond Myklebust88be9f92007-06-05 10:42:27 -0400785 data->inode = inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 data->cred = req->wb_context->cred;
787
788 data->args.fh = NFS_FH(inode);
789 data->args.offset = req_offset(req) + offset;
790 data->args.pgbase = req->wb_pgbase + offset;
791 data->args.pages = data->pagevec;
792 data->args.count = count;
793 data->args.context = req->wb_context;
794
795 data->res.fattr = &data->fattr;
796 data->res.count = count;
797 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400798 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Trond Myklebust788e7a82006-03-20 13:44:27 -0500800 /* Set up the initial task struct. */
801 flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
802 rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 NFS_PROTO(inode)->write_setup(data, how);
804
805 data->task.tk_priority = flush_task_priority(how);
806 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807
Chuck Levera3f565b2007-01-31 12:14:01 -0500808 dprintk("NFS: %5u initiated write call "
809 "(req %s/%Ld, %u bytes @ offset %Lu)\n",
Chuck Lever0bbacc42005-11-01 16:53:32 -0500810 data->task.tk_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 inode->i_sb->s_id,
812 (long long)NFS_FILEID(inode),
813 count,
814 (unsigned long long)data->args.offset);
815}
816
817static void nfs_execute_write(struct nfs_write_data *data)
818{
819 struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
820 sigset_t oldset;
821
822 rpc_clnt_sigmask(clnt, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 rpc_execute(&data->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 rpc_clnt_sigunmask(clnt, &oldset);
825}
826
827/*
828 * Generate multiple small requests to write out a single
829 * contiguous dirty area on one page.
830 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400831static int nfs_flush_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
833 struct nfs_page *req = nfs_list_entry(head->next);
834 struct page *page = req->wb_page;
835 struct nfs_write_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700836 size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
837 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 int requests = 0;
839 LIST_HEAD(list);
840
841 nfs_list_remove_request(req);
842
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400843 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700844 do {
845 size_t len = min(nbytes, wsize);
846
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400847 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 if (!data)
849 goto out_bad;
850 list_add(&data->pages, &list);
851 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700852 nbytes -= len;
853 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 atomic_set(&req->wb_complete, requests);
855
856 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400858 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 do {
860 data = list_entry(list.next, struct nfs_write_data, pages);
861 list_del_init(&data->pages);
862
863 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400865 if (nbytes < wsize)
866 wsize = nbytes;
867 nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
868 wsize, offset, how);
869 offset += wsize;
870 nbytes -= wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 nfs_execute_write(data);
872 } while (nbytes != 0);
873
874 return 0;
875
876out_bad:
877 while (!list_empty(&list)) {
878 data = list_entry(list.next, struct nfs_write_data, pages);
879 list_del(&data->pages);
Trond Myklebust8aca67f2006-11-13 16:23:44 -0500880 nfs_writedata_release(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500882 nfs_redirty_request(req);
Trond Myklebust6d677e32007-04-20 16:12:40 -0400883 nfs_end_page_writeback(req->wb_page);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400884 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 return -ENOMEM;
886}
887
888/*
889 * Create an RPC task for the given write request and kick it.
890 * The page must have been locked by the caller.
891 *
892 * It may happen that the page we're passed is not marked dirty.
893 * This is the case if nfs_updatepage detects a conflicting request
894 * that has been written but not committed.
895 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400896static int nfs_flush_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897{
898 struct nfs_page *req;
899 struct page **pages;
900 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400902 data = nfs_writedata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (!data)
904 goto out_bad;
905
906 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 while (!list_empty(head)) {
908 req = nfs_list_entry(head->next);
909 nfs_list_remove_request(req);
910 nfs_list_add_request(req, &data->pages);
911 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 }
914 req = nfs_list_entry(data->pages.next);
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 /* Set up the argument struct */
Trond Myklebust788e7a82006-03-20 13:44:27 -0500917 nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
919 nfs_execute_write(data);
920 return 0;
921 out_bad:
922 while (!list_empty(head)) {
Trond Myklebust10afec92007-05-14 17:16:04 -0400923 req = nfs_list_entry(head->next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 nfs_list_remove_request(req);
Trond Myklebust61822ab2006-12-05 00:35:42 -0500925 nfs_redirty_request(req);
Trond Myklebust6d677e32007-04-20 16:12:40 -0400926 nfs_end_page_writeback(req->wb_page);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400927 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 }
929 return -ENOMEM;
930}
931
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400932static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
933 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Trond Myklebust7d46a492006-03-20 13:44:50 -0500935 int wsize = NFS_SERVER(inode)->wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400937 if (wsize < PAGE_CACHE_SIZE)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400938 nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400939 else
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400940 nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941}
942
943/*
944 * Handle a write reply that flushed part of a page.
945 */
Trond Myklebust788e7a82006-03-20 13:44:27 -0500946static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947{
Trond Myklebust788e7a82006-03-20 13:44:27 -0500948 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 struct nfs_page *req = data->req;
950 struct page *page = req->wb_page;
951
952 dprintk("NFS: write (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400953 req->wb_context->path.dentry->d_inode->i_sb->s_id,
954 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 req->wb_bytes,
956 (long long)req_offset(req));
957
Trond Myklebust788e7a82006-03-20 13:44:27 -0500958 if (nfs_writeback_done(task, data) != 0)
959 return;
960
961 if (task->tk_status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -0800962 nfs_set_pageerror(page);
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400963 nfs_context_set_write_error(req->wb_context, task->tk_status);
Trond Myklebust788e7a82006-03-20 13:44:27 -0500964 dprintk(", error = %d\n", task->tk_status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400965 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 }
967
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400968 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400969 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400970
Trond Myklebust587142f2007-07-02 09:57:54 -0400971 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400972 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
973 /* Do nothing we need to resend the writes */
974 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
975 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
976 dprintk(" defer commit\n");
977 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
978 set_bit(PG_NEED_RESCHED, &req->wb_flags);
979 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
980 dprintk(" server reboot detected\n");
981 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400982 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400983 } else
984 dprintk(" OK\n");
985
986out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 if (atomic_dec_and_test(&req->wb_complete))
988 nfs_writepage_release(req);
989}
990
Trond Myklebust788e7a82006-03-20 13:44:27 -0500991static const struct rpc_call_ops nfs_write_partial_ops = {
992 .rpc_call_done = nfs_writeback_done_partial,
993 .rpc_release = nfs_writedata_release,
994};
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996/*
997 * Handle a write reply that flushes a whole page.
998 *
999 * FIXME: There is an inherent race with invalidate_inode_pages and
1000 * writebacks since the page->count is kept > 1 for as long
1001 * as the page has a write request pending.
1002 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001003static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001005 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 struct nfs_page *req;
1007 struct page *page;
1008
Trond Myklebust788e7a82006-03-20 13:44:27 -05001009 if (nfs_writeback_done(task, data) != 0)
1010 return;
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* Update attributes as result of writeback. */
1013 while (!list_empty(&data->pages)) {
1014 req = nfs_list_entry(data->pages.next);
1015 nfs_list_remove_request(req);
1016 page = req->wb_page;
1017
1018 dprintk("NFS: write (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001019 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1020 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 req->wb_bytes,
1022 (long long)req_offset(req));
1023
Trond Myklebust788e7a82006-03-20 13:44:27 -05001024 if (task->tk_status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001025 nfs_set_pageerror(page);
Trond Myklebust7b159fc2007-07-25 14:09:54 -04001026 nfs_context_set_write_error(req->wb_context, task->tk_status);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001027 dprintk(", error = %d\n", task->tk_status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001028 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001031 if (nfs_write_need_commit(data)) {
1032 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1033 nfs_mark_request_commit(req);
1034 nfs_end_page_writeback(page);
1035 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 goto next;
1037 }
Trond Myklebust44dd1512007-05-19 11:58:03 -04001038 /* Set the PG_uptodate flag? */
1039 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001040 dprintk(" OK\n");
1041remove_request:
1042 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001045 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
1047}
1048
Trond Myklebust788e7a82006-03-20 13:44:27 -05001049static const struct rpc_call_ops nfs_write_full_ops = {
1050 .rpc_call_done = nfs_writeback_done_full,
1051 .rpc_release = nfs_writedata_release,
1052};
1053
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055/*
1056 * This function is called when the WRITE call is complete.
1057 */
Chuck Lever462d5b32006-03-20 13:44:32 -05001058int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 struct nfs_writeargs *argp = &data->args;
1061 struct nfs_writeres *resp = &data->res;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001062 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Chuck Levera3f565b2007-01-31 12:14:01 -05001064 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 task->tk_pid, task->tk_status);
1066
Chuck Leverf551e442006-09-20 14:33:04 -04001067 /*
1068 * ->write_done will attempt to use post-op attributes to detect
1069 * conflicting writes by other clients. A strict interpretation
1070 * of close-to-open would allow us to continue caching even if
1071 * another writer had changed the file, but some applications
1072 * depend on tighter cache coherency when writing.
1073 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001074 status = NFS_PROTO(data->inode)->write_done(task, data);
1075 if (status != 0)
1076 return status;
Chuck Lever91d5b472006-03-20 13:44:14 -05001077 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1080 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1081 /* We tried a write call, but the server did not
1082 * commit data to stable storage even though we
1083 * requested it.
1084 * Note: There is a known bug in Tru64 < 5.0 in which
1085 * the server reports NFS_DATA_SYNC, but performs
1086 * NFS_FILE_SYNC. We therefore implement this checking
1087 * as a dprintk() in order to avoid filling syslog.
1088 */
1089 static unsigned long complain;
1090
1091 if (time_before(complain, jiffies)) {
1092 dprintk("NFS: faulty NFS server %s:"
1093 " (committed = %d) != (stable = %d)\n",
David Howells54ceac42006-08-22 20:06:13 -04001094 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 resp->verf->committed, argp->stable);
1096 complain = jiffies + 300 * HZ;
1097 }
1098 }
1099#endif
1100 /* Is this a short write? */
1101 if (task->tk_status >= 0 && resp->count < argp->count) {
1102 static unsigned long complain;
1103
Chuck Lever91d5b472006-03-20 13:44:14 -05001104 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 /* Has the server at least made some progress? */
1107 if (resp->count != 0) {
1108 /* Was this an NFSv2 write or an NFSv3 stable write? */
1109 if (resp->verf->committed != NFS_UNSTABLE) {
1110 /* Resend from where the server left off */
1111 argp->offset += resp->count;
1112 argp->pgbase += resp->count;
1113 argp->count -= resp->count;
1114 } else {
1115 /* Resend as a stable write in order to avoid
1116 * headaches in the case of a server crash.
1117 */
1118 argp->stable = NFS_FILE_SYNC;
1119 }
1120 rpc_restart_call(task);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001121 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
1123 if (time_before(complain, jiffies)) {
1124 printk(KERN_WARNING
1125 "NFS: Server wrote zero bytes, expected %u.\n",
1126 argp->count);
1127 complain = jiffies + 300 * HZ;
1128 }
1129 /* Can't do anything about it except throw an error. */
1130 task->tk_status = -EIO;
1131 }
Trond Myklebust788e7a82006-03-20 13:44:27 -05001132 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133}
1134
1135
1136#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001137void nfs_commit_release(void *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 nfs_commit_free(wdata);
1140}
1141
1142/*
1143 * Set up the argument/result storage required for the RPC call.
1144 */
1145static void nfs_commit_rpcsetup(struct list_head *head,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001146 struct nfs_write_data *data,
1147 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001149 struct nfs_page *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 struct inode *inode;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001151 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152
1153 /* Set up the RPC argument and reply structs
1154 * NB: take care not to mess about with data->commit et al. */
1155
1156 list_splice_init(head, &data->pages);
1157 first = nfs_list_entry(data->pages.next);
Trond Myklebust88be9f92007-06-05 10:42:27 -04001158 inode = first->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 data->inode = inode;
1161 data->cred = first->wb_context->cred;
1162
1163 data->args.fh = NFS_FH(data->inode);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001164 /* Note: we always request a commit of the entire inode */
1165 data->args.offset = 0;
1166 data->args.count = 0;
1167 data->res.count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 data->res.fattr = &data->fattr;
1169 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -04001170 nfs_fattr_init(&data->fattr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001171
1172 /* Set up the initial task struct. */
1173 flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
1174 rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 NFS_PROTO(inode)->commit_setup(data, how);
1176
1177 data->task.tk_priority = flush_task_priority(how);
1178 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179
Chuck Levera3f565b2007-01-31 12:14:01 -05001180 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
1183/*
1184 * Commit dirty pages
1185 */
1186static int
Chuck Lever40859d72005-11-30 18:09:02 -05001187nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188{
1189 struct nfs_write_data *data;
1190 struct nfs_page *req;
1191
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001192 data = nfs_commit_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194 if (!data)
1195 goto out_bad;
1196
1197 /* Set up the argument struct */
1198 nfs_commit_rpcsetup(head, data, how);
1199
1200 nfs_execute_write(data);
1201 return 0;
1202 out_bad:
1203 while (!list_empty(head)) {
1204 req = nfs_list_entry(head->next);
1205 nfs_list_remove_request(req);
1206 nfs_mark_request_commit(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001207 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001208 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1209 BDI_RECLAIMABLE);
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001210 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 }
1212 return -ENOMEM;
1213}
1214
1215/*
1216 * COMMIT call returned
1217 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001218static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001220 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 struct nfs_page *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Chuck Levera3f565b2007-01-31 12:14:01 -05001223 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 task->tk_pid, task->tk_status);
1225
Trond Myklebust788e7a82006-03-20 13:44:27 -05001226 /* Call the NFS version-specific code */
1227 if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1228 return;
1229
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 while (!list_empty(&data->pages)) {
1231 req = nfs_list_entry(data->pages.next);
1232 nfs_list_remove_request(req);
Trond Myklebust612c9382007-04-20 16:12:45 -04001233 clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001234 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001235 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1236 BDI_RECLAIMABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
1238 dprintk("NFS: commit (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001239 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1240 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 req->wb_bytes,
1242 (long long)req_offset(req));
1243 if (task->tk_status < 0) {
Trond Myklebust7b159fc2007-07-25 14:09:54 -04001244 nfs_context_set_write_error(req->wb_context, task->tk_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 nfs_inode_remove_request(req);
1246 dprintk(", error = %d\n", task->tk_status);
1247 goto next;
1248 }
1249
1250 /* Okay, COMMIT succeeded, apparently. Check the verifier
1251 * returned by the server against all stored verfs. */
1252 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1253 /* We have a match */
Trond Myklebust44dd1512007-05-19 11:58:03 -04001254 /* Set the PG_uptodate flag */
1255 nfs_mark_uptodate(req->wb_page, req->wb_pgbase,
1256 req->wb_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 nfs_inode_remove_request(req);
1258 dprintk(" OK\n");
1259 goto next;
1260 }
1261 /* We have a mismatch. Write the page again */
1262 dprintk(" mismatch\n");
Trond Myklebust61822ab2006-12-05 00:35:42 -05001263 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001265 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001268
1269static const struct rpc_call_ops nfs_commit_ops = {
1270 .rpc_call_done = nfs_commit_done,
1271 .rpc_release = nfs_commit_release,
1272};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001274int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 LIST_HEAD(head);
Trond Myklebust7d46a492006-03-20 13:44:50 -05001277 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
Trond Myklebust587142f2007-07-02 09:57:54 -04001279 spin_lock(&inode->i_lock);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001280 res = nfs_scan_commit(inode, &head, 0, 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001281 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 if (res) {
Trond Myklebust7d46a492006-03-20 13:44:50 -05001283 int error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001284 if (error < 0)
1285 return error;
1286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 return res;
1288}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001289#else
1290static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1291{
1292 return 0;
1293}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294#endif
1295
Trond Myklebust1c759502006-10-09 16:18:38 -04001296long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Trond Myklebust1c759502006-10-09 16:18:38 -04001298 struct inode *inode = mapping->host;
Trond Myklebustca52fec2007-04-17 17:22:13 -04001299 pgoff_t idx_start, idx_end;
Trond Myklebust1c759502006-10-09 16:18:38 -04001300 unsigned int npages = 0;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001301 LIST_HEAD(head);
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001302 int nocommit = how & FLUSH_NOCOMMIT;
Trond Myklebust3f442542006-09-17 14:46:44 -04001303 long pages, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Trond Myklebust1c759502006-10-09 16:18:38 -04001305 /* FIXME */
1306 if (wbc->range_cyclic)
1307 idx_start = 0;
1308 else {
1309 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
1310 idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
1311 if (idx_end > idx_start) {
Trond Myklebustca52fec2007-04-17 17:22:13 -04001312 pgoff_t l_npages = 1 + idx_end - idx_start;
Trond Myklebust1c759502006-10-09 16:18:38 -04001313 npages = l_npages;
1314 if (sizeof(npages) != sizeof(l_npages) &&
Trond Myklebustca52fec2007-04-17 17:22:13 -04001315 (pgoff_t)npages != l_npages)
Trond Myklebust1c759502006-10-09 16:18:38 -04001316 npages = 0;
1317 }
1318 }
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001319 how &= ~FLUSH_NOCOMMIT;
Trond Myklebust587142f2007-07-02 09:57:54 -04001320 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 do {
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001322 ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
1323 if (ret != 0)
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001324 continue;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001325 if (nocommit)
1326 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001327 pages = nfs_scan_commit(inode, &head, idx_start, npages);
Trond Myklebust724c4392007-04-17 17:22:13 -04001328 if (pages == 0)
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001329 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001330 if (how & FLUSH_INVALIDATE) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001331 spin_unlock(&inode->i_lock);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001332 nfs_cancel_commit_list(&head);
Trond Myklebuste8e058e2006-11-15 17:31:56 -05001333 ret = pages;
Trond Myklebust587142f2007-07-02 09:57:54 -04001334 spin_lock(&inode->i_lock);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001335 continue;
1336 }
1337 pages += nfs_scan_commit(inode, &head, 0, 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001338 spin_unlock(&inode->i_lock);
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001339 ret = nfs_commit_list(inode, &head, how);
Trond Myklebust587142f2007-07-02 09:57:54 -04001340 spin_lock(&inode->i_lock);
1341
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001342 } while (ret >= 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001343 spin_unlock(&inode->i_lock);
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001344 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345}
1346
Trond Myklebust34901f72007-07-25 14:09:54 -04001347static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
Trond Myklebust1c759502006-10-09 16:18:38 -04001348{
Trond Myklebust1c759502006-10-09 16:18:38 -04001349 int ret;
1350
Trond Myklebust34901f72007-07-25 14:09:54 -04001351 ret = nfs_writepages(mapping, wbc);
Trond Myklebust61822ab2006-12-05 00:35:42 -05001352 if (ret < 0)
1353 goto out;
Trond Myklebust34901f72007-07-25 14:09:54 -04001354 ret = nfs_sync_mapping_wait(mapping, wbc, how);
Trond Myklebusted90ef52007-07-20 13:13:28 -04001355 if (ret < 0)
1356 goto out;
1357 return 0;
Trond Myklebust61822ab2006-12-05 00:35:42 -05001358out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001359 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Trond Myklebust1c759502006-10-09 16:18:38 -04001360 return ret;
1361}
1362
Trond Myklebust34901f72007-07-25 14:09:54 -04001363/* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */
1364static int nfs_write_mapping(struct address_space *mapping, int how)
1365{
1366 struct writeback_control wbc = {
1367 .bdi = mapping->backing_dev_info,
1368 .sync_mode = WB_SYNC_NONE,
1369 .nr_to_write = LONG_MAX,
1370 .for_writepages = 1,
1371 .range_cyclic = 1,
1372 };
1373 int ret;
1374
1375 ret = __nfs_write_mapping(mapping, &wbc, how);
1376 if (ret < 0)
1377 return ret;
1378 wbc.sync_mode = WB_SYNC_ALL;
1379 return __nfs_write_mapping(mapping, &wbc, how);
1380}
1381
Trond Myklebusted90ef52007-07-20 13:13:28 -04001382/*
1383 * flush the inode to disk.
1384 */
1385int nfs_wb_all(struct inode *inode)
Trond Myklebust1c759502006-10-09 16:18:38 -04001386{
Trond Myklebusted90ef52007-07-20 13:13:28 -04001387 return nfs_write_mapping(inode->i_mapping, 0);
1388}
Trond Myklebust1c759502006-10-09 16:18:38 -04001389
Trond Myklebusted90ef52007-07-20 13:13:28 -04001390int nfs_wb_nocommit(struct inode *inode)
1391{
1392 return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT);
Trond Myklebust1c759502006-10-09 16:18:38 -04001393}
1394
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001395int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1396{
1397 struct nfs_page *req;
1398 loff_t range_start = page_offset(page);
1399 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1400 struct writeback_control wbc = {
1401 .bdi = page->mapping->backing_dev_info,
1402 .sync_mode = WB_SYNC_ALL,
1403 .nr_to_write = LONG_MAX,
1404 .range_start = range_start,
1405 .range_end = range_end,
1406 };
1407 int ret = 0;
1408
1409 BUG_ON(!PageLocked(page));
1410 for (;;) {
1411 req = nfs_page_find_request(page);
1412 if (req == NULL)
1413 goto out;
1414 if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1415 nfs_release_request(req);
1416 break;
1417 }
1418 if (nfs_lock_request_dontget(req)) {
1419 nfs_inode_remove_request(req);
1420 /*
1421 * In case nfs_inode_remove_request has marked the
1422 * page as being dirty
1423 */
1424 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1425 nfs_unlock_request(req);
1426 break;
1427 }
1428 ret = nfs_wait_on_request(req);
1429 if (ret < 0)
1430 goto out;
1431 }
1432 if (!PagePrivate(page))
1433 return 0;
1434 ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE);
1435out:
1436 return ret;
1437}
1438
Trond Myklebust61822ab2006-12-05 00:35:42 -05001439int nfs_wb_page_priority(struct inode *inode, struct page *page, int how)
Trond Myklebust1c759502006-10-09 16:18:38 -04001440{
1441 loff_t range_start = page_offset(page);
1442 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001443 struct writeback_control wbc = {
1444 .bdi = page->mapping->backing_dev_info,
1445 .sync_mode = WB_SYNC_ALL,
1446 .nr_to_write = LONG_MAX,
1447 .range_start = range_start,
1448 .range_end = range_end,
1449 };
1450 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001451
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001452 BUG_ON(!PageLocked(page));
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001453 if (clear_page_dirty_for_io(page)) {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001454 ret = nfs_writepage_locked(page, &wbc);
1455 if (ret < 0)
1456 goto out;
1457 }
Trond Myklebustf40313a2007-01-13 02:28:08 -05001458 if (!PagePrivate(page))
1459 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001460 ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
1461 if (ret >= 0)
1462 return 0;
1463out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001464 __mark_inode_dirty(inode, I_DIRTY_PAGES);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001465 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001466}
1467
1468/*
1469 * Write back all requests on one page - we do this before reading it.
1470 */
1471int nfs_wb_page(struct inode *inode, struct page* page)
1472{
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001473 return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
Trond Myklebust1c759502006-10-09 16:18:38 -04001474}
1475
David Howellsf7b422b2006-06-09 09:34:33 -04001476int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
1478 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1479 sizeof(struct nfs_write_data),
1480 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001481 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (nfs_wdata_cachep == NULL)
1483 return -ENOMEM;
1484
Matthew Dobson93d23412006-03-26 01:37:50 -08001485 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1486 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 if (nfs_wdata_mempool == NULL)
1488 return -ENOMEM;
1489
Matthew Dobson93d23412006-03-26 01:37:50 -08001490 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1491 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 if (nfs_commit_mempool == NULL)
1493 return -ENOMEM;
1494
Peter Zijlstra89a09142007-03-16 13:38:26 -08001495 /*
1496 * NFS congestion size, scale with available memory.
1497 *
1498 * 64MB: 8192k
1499 * 128MB: 11585k
1500 * 256MB: 16384k
1501 * 512MB: 23170k
1502 * 1GB: 32768k
1503 * 2GB: 46340k
1504 * 4GB: 65536k
1505 * 8GB: 92681k
1506 * 16GB: 131072k
1507 *
1508 * This allows larger machines to have larger/more transfers.
1509 * Limit the default to 256M
1510 */
1511 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1512 if (nfs_congestion_kb > 256*1024)
1513 nfs_congestion_kb = 256*1024;
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 return 0;
1516}
1517
David Brownell266bee82006-06-27 12:59:15 -07001518void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519{
1520 mempool_destroy(nfs_commit_mempool);
1521 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001522 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523}
1524