blob: 9e7c21da864f4162cea37010c84115d8a2125bbe [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 Myklebust277459d2006-12-05 00:35:35 -0500113static struct nfs_page *nfs_page_find_request_locked(struct page *page)
114{
115 struct nfs_page *req = NULL;
116
117 if (PagePrivate(page)) {
118 req = (struct nfs_page *)page_private(page);
119 if (req != NULL)
120 atomic_inc(&req->wb_count);
121 }
122 return req;
123}
124
125static struct nfs_page *nfs_page_find_request(struct page *page)
126{
127 struct nfs_page *req = NULL;
128 spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
129
130 spin_lock(req_lock);
131 req = nfs_page_find_request_locked(page);
132 spin_unlock(req_lock);
133 return req;
134}
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136/* Adjust the file length if we're writing beyond the end */
137static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
138{
139 struct inode *inode = page->mapping->host;
140 loff_t end, i_size = i_size_read(inode);
Trond Myklebustca52fec2007-04-17 17:22:13 -0400141 pgoff_t end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143 if (i_size > 0 && page->index < end_index)
144 return;
145 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
146 if (i_size >= end)
147 return;
Chuck Lever91d5b472006-03-20 13:44:14 -0500148 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 i_size_write(inode, end);
150}
151
Trond Myklebusta301b772007-02-06 11:07:15 -0800152/* A writeback failed: mark the page as bad, and invalidate the page cache */
153static void nfs_set_pageerror(struct page *page)
154{
155 SetPageError(page);
156 nfs_zap_mapping(page->mapping->host, page->mapping);
157}
158
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159/* We can set the PG_uptodate flag if we see that a write request
160 * covers the full page.
161 */
162static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
163{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 if (PageUptodate(page))
165 return;
166 if (base != 0)
167 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500168 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500170 if (count != PAGE_CACHE_SIZE)
Nate Diller60945cb2007-05-10 22:55:08 -0700171 zero_user_page(page, count, PAGE_CACHE_SIZE - count, KM_USER0);
Trond Myklebust49a70f22006-12-05 00:35:38 -0500172 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173}
174
Trond Myklebuste21195a2006-12-05 00:35:39 -0500175static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 unsigned int offset, unsigned int count)
177{
178 struct nfs_page *req;
Trond Myklebuste21195a2006-12-05 00:35:39 -0500179 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Trond Myklebuste21195a2006-12-05 00:35:39 -0500181 for (;;) {
182 req = nfs_update_request(ctx, page, offset, count);
183 if (!IS_ERR(req))
184 break;
185 ret = PTR_ERR(req);
186 if (ret != -EBUSY)
187 return ret;
188 ret = nfs_wb_page(page->mapping->host, page);
189 if (ret != 0)
190 return ret;
191 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 /* Update file length */
193 nfs_grow_file(page, offset, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 nfs_unlock_request(req);
Trond Myklebustabd3e642006-01-03 09:55:02 +0100195 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196}
197
198static int wb_priority(struct writeback_control *wbc)
199{
200 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400201 return FLUSH_HIGHPRI | FLUSH_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (wbc->for_kupdate)
203 return FLUSH_LOWPRI;
204 return 0;
205}
206
207/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800208 * NFS congestion control
209 */
210
211int nfs_congestion_kb;
212
213#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
214#define NFS_CONGESTION_OFF_THRESH \
215 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
216
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400217static int nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800218{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400219 int ret = test_set_page_writeback(page);
220
221 if (!ret) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800222 struct inode *inode = page->mapping->host;
223 struct nfs_server *nfss = NFS_SERVER(inode);
224
Peter Zijlstra277866a2007-05-08 00:35:12 -0700225 if (atomic_long_inc_return(&nfss->writeback) >
Peter Zijlstra89a09142007-03-16 13:38:26 -0800226 NFS_CONGESTION_ON_THRESH)
227 set_bdi_congested(&nfss->backing_dev_info, WRITE);
228 }
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400229 return ret;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800230}
231
232static void nfs_end_page_writeback(struct page *page)
233{
234 struct inode *inode = page->mapping->host;
235 struct nfs_server *nfss = NFS_SERVER(inode);
236
237 end_page_writeback(page);
Peter Zijlstra277866a2007-05-08 00:35:12 -0700238 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800239 clear_bdi_congested(&nfss->backing_dev_info, WRITE);
240 congestion_end(WRITE);
241 }
242}
243
244/*
Trond Myklebuste261f512006-12-05 00:35:41 -0500245 * Find an associated nfs write request, and prepare to flush it out
246 * Returns 1 if there was no write request, or if the request was
247 * already tagged by nfs_set_page_dirty.Returns 0 if the request
248 * was not tagged.
249 * May also return an error if the user signalled nfs_wait_on_request().
250 */
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{
254 struct nfs_page *req;
Trond Myklebust612c9382007-04-20 16:12:45 -0400255 struct nfs_inode *nfsi = NFS_I(page->mapping->host);
256 spinlock_t *req_lock = &nfsi->req_lock;
Trond Myklebuste261f512006-12-05 00:35:41 -0500257 int ret;
258
259 spin_lock(req_lock);
260 for(;;) {
261 req = nfs_page_find_request_locked(page);
262 if (req == NULL) {
263 spin_unlock(req_lock);
264 return 1;
265 }
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 */
273 spin_unlock(req_lock);
274 ret = nfs_wait_on_request(req);
275 nfs_release_request(req);
276 if (ret != 0)
277 return ret;
278 spin_lock(req_lock);
279 }
Trond Myklebust612c9382007-04-20 16:12:45 -0400280 if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
281 /* This request is marked for commit */
282 spin_unlock(req_lock);
283 nfs_unlock_request(req);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400284 nfs_pageio_complete(pgio);
Trond Myklebust612c9382007-04-20 16:12:45 -0400285 return 1;
286 }
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400287 if (nfs_set_page_writeback(page) != 0) {
Trond Myklebust612c9382007-04-20 16:12:45 -0400288 spin_unlock(req_lock);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400289 BUG();
290 }
291 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
292 NFS_PAGE_TAG_WRITEBACK);
Trond Myklebuste261f512006-12-05 00:35:41 -0500293 ret = test_bit(PG_NEED_FLUSH, &req->wb_flags);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400294 spin_unlock(req_lock);
295 nfs_pageio_add_request(pgio, req);
Trond Myklebuste261f512006-12-05 00:35:41 -0500296 return ret;
297}
298
299/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * Write an mmapped page to the server.
301 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500302static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303{
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400304 struct nfs_pageio_descriptor mypgio, *pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 struct nfs_open_context *ctx;
306 struct inode *inode = page->mapping->host;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500307 unsigned offset;
Trond Myklebuste261f512006-12-05 00:35:41 -0500308 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
Chuck Lever91d5b472006-03-20 13:44:14 -0500310 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
311 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
312
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400313 if (wbc->for_writepages)
314 pgio = wbc->fs_private;
315 else {
316 nfs_pageio_init_write(&mypgio, inode, wb_priority(wbc));
317 pgio = &mypgio;
318 }
319
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400320 nfs_pageio_cond_complete(pgio, page->index);
321
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400322 err = nfs_page_async_flush(pgio, page);
Trond Myklebuste261f512006-12-05 00:35:41 -0500323 if (err <= 0)
324 goto out;
325 err = 0;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500326 offset = nfs_page_length(page);
327 if (!offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 goto out;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500329
Trond Myklebust7fe7f842007-05-20 10:18:27 -0400330 nfs_pageio_cond_complete(pgio, page->index);
331
Trond Myklebustd5308382005-11-04 15:33:38 -0500332 ctx = nfs_find_open_context(inode, NULL, FMODE_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (ctx == NULL) {
334 err = -EBADF;
335 goto out;
336 }
Trond Myklebust200baa22006-12-05 00:35:40 -0500337 err = nfs_writepage_setup(ctx, page, 0, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 put_nfs_open_context(ctx);
Trond Myklebuste261f512006-12-05 00:35:41 -0500339 if (err != 0)
340 goto out;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400341 err = nfs_page_async_flush(pgio, page);
Trond Myklebuste261f512006-12-05 00:35:41 -0500342 if (err > 0)
343 err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344out:
Trond Myklebust200baa22006-12-05 00:35:40 -0500345 if (!wbc->for_writepages)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400346 nfs_pageio_complete(pgio);
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500347 return err;
348}
349
350int nfs_writepage(struct page *page, struct writeback_control *wbc)
351{
352 int err;
353
354 err = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return err;
357}
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
360{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 struct inode *inode = mapping->host;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400362 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 int err;
364
Chuck Lever91d5b472006-03-20 13:44:14 -0500365 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
366
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400367 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
368 wbc->fs_private = &pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 err = generic_writepages(mapping, wbc);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400370 nfs_pageio_complete(&pgio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 if (err)
372 return err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400373 if (pgio.pg_error)
374 return pgio.pg_error;
375 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
377
378/*
379 * Insert a write request into an inode
380 */
381static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
382{
383 struct nfs_inode *nfsi = NFS_I(inode);
384 int error;
385
386 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
387 BUG_ON(error == -EEXIST);
388 if (error)
389 return error;
390 if (!nfsi->npages) {
391 igrab(inode);
392 nfs_begin_data_update(inode);
393 if (nfs_have_delegation(inode, FMODE_WRITE))
394 nfsi->change_attr++;
395 }
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500396 SetPagePrivate(req->wb_page);
Trond Myklebust277459d2006-12-05 00:35:35 -0500397 set_page_private(req->wb_page, (unsigned long)req);
Trond Myklebust2b82f192007-04-20 16:12:50 -0400398 if (PageDirty(req->wb_page))
399 set_bit(PG_NEED_FLUSH, &req->wb_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 nfsi->npages++;
401 atomic_inc(&req->wb_count);
402 return 0;
403}
404
405/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800406 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 */
408static void nfs_inode_remove_request(struct nfs_page *req)
409{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400410 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 struct nfs_inode *nfsi = NFS_I(inode);
412
413 BUG_ON (!NFS_WBACK_BUSY(req));
414
415 spin_lock(&nfsi->req_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500416 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500417 ClearPagePrivate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
Trond Myklebust2b82f192007-04-20 16:12:50 -0400419 if (test_and_clear_bit(PG_NEED_FLUSH, &req->wb_flags))
420 __set_page_dirty_nobuffers(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 nfsi->npages--;
422 if (!nfsi->npages) {
423 spin_unlock(&nfsi->req_lock);
Trond Myklebust951a1432005-06-22 17:16:30 +0000424 nfs_end_data_update(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 iput(inode);
426 } else
427 spin_unlock(&nfsi->req_lock);
428 nfs_clear_request(req);
429 nfs_release_request(req);
430}
431
Trond Myklebust61822ab2006-12-05 00:35:42 -0500432static void
433nfs_redirty_request(struct nfs_page *req)
434{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500435 __set_page_dirty_nobuffers(req->wb_page);
436}
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438/*
439 * Check if a request is dirty
440 */
441static inline int
442nfs_dirty_request(struct nfs_page *req)
443{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400444 struct page *page = req->wb_page;
445
Trond Myklebust612c9382007-04-20 16:12:45 -0400446 if (page == NULL || test_bit(PG_NEED_COMMIT, &req->wb_flags))
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400447 return 0;
448 return !PageWriteback(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449}
450
451#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
452/*
453 * Add a request to the inode's commit list.
454 */
455static void
456nfs_mark_request_commit(struct nfs_page *req)
457{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400458 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 struct nfs_inode *nfsi = NFS_I(inode);
460
461 spin_lock(&nfsi->req_lock);
462 nfs_list_add_request(req, &nfsi->commit);
463 nfsi->ncommit++;
Trond Myklebust612c9382007-04-20 16:12:45 -0400464 set_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 spin_unlock(&nfsi->req_lock);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700466 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Trond Myklebusta1803042006-12-05 00:45:19 -0500467 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400469
470static inline
471int nfs_write_need_commit(struct nfs_write_data *data)
472{
473 return data->verf.committed != NFS_FILE_SYNC;
474}
475
476static inline
477int nfs_reschedule_unstable_write(struct nfs_page *req)
478{
Trond Myklebust612c9382007-04-20 16:12:45 -0400479 if (test_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400480 nfs_mark_request_commit(req);
481 return 1;
482 }
483 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
484 nfs_redirty_request(req);
485 return 1;
486 }
487 return 0;
488}
489#else
490static inline void
491nfs_mark_request_commit(struct nfs_page *req)
492{
493}
494
495static inline
496int nfs_write_need_commit(struct nfs_write_data *data)
497{
498 return 0;
499}
500
501static inline
502int nfs_reschedule_unstable_write(struct nfs_page *req)
503{
504 return 0;
505}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506#endif
507
508/*
509 * Wait for a request to complete.
510 *
511 * Interruptible by signals only if mounted with intr flag.
512 */
Trond Myklebustca52fec2007-04-17 17:22:13 -0400513static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
515 struct nfs_inode *nfsi = NFS_I(inode);
516 struct nfs_page *req;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400517 pgoff_t idx_end, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 unsigned int res = 0;
519 int error;
520
521 if (npages == 0)
522 idx_end = ~0;
523 else
524 idx_end = idx_start + npages - 1;
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 next = idx_start;
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000527 while (radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree, (void **)&req, next, 1, NFS_PAGE_TAG_WRITEBACK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if (req->wb_index > idx_end)
529 break;
530
531 next = req->wb_index + 1;
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000532 BUG_ON(!NFS_WBACK_BUSY(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 atomic_inc(&req->wb_count);
535 spin_unlock(&nfsi->req_lock);
536 error = nfs_wait_on_request(req);
537 nfs_release_request(req);
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500538 spin_lock(&nfsi->req_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 if (error < 0)
540 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 res++;
542 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 return res;
544}
545
Trond Myklebust83715ad2006-07-05 13:17:12 -0400546static void nfs_cancel_commit_list(struct list_head *head)
547{
548 struct nfs_page *req;
549
550 while(!list_empty(head)) {
551 req = nfs_list_entry(head->next);
Trond Myklebustb6dff262006-10-19 23:28:38 -0700552 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400553 nfs_list_remove_request(req);
Trond Myklebust612c9382007-04-20 16:12:45 -0400554 clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400555 nfs_inode_remove_request(req);
Trond Myklebustb6dff262006-10-19 23:28:38 -0700556 nfs_unlock_request(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400557 }
558}
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
561/*
562 * nfs_scan_commit - Scan an inode for commit requests
563 * @inode: NFS inode to scan
564 * @dst: destination list
565 * @idx_start: lower bound of page->index to scan.
566 * @npages: idx_start + npages sets the upper bound to scan.
567 *
568 * Moves requests from the inode's 'commit' request list.
569 * The requests are *not* checked to ensure that they form a contiguous set.
570 */
571static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400572nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573{
574 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000575 int res = 0;
576
577 if (nfsi->ncommit != 0) {
Trond Myklebustd2ccddf2006-05-31 01:13:38 -0400578 res = nfs_scan_list(nfsi, &nfsi->commit, dst, idx_start, npages);
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000579 nfsi->ncommit -= res;
580 if ((nfsi->ncommit == 0) != list_empty(&nfsi->commit))
581 printk(KERN_ERR "NFS: desynchronized value of nfs_i.ncommit.\n");
582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return res;
584}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500585#else
Trond Myklebustca52fec2007-04-17 17:22:13 -0400586static 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 -0500587{
588 return 0;
589}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590#endif
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592/*
593 * Try to update any existing write request, or create one if there is none.
594 * In order to match, the request's credentials must match those of
595 * the calling process.
596 *
597 * Note: Should always be called with the Page Lock held!
598 */
599static struct nfs_page * nfs_update_request(struct nfs_open_context* ctx,
Trond Myklebuste21195a2006-12-05 00:35:39 -0500600 struct page *page, unsigned int offset, unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601{
Peter Zijlstra89a09142007-03-16 13:38:26 -0800602 struct address_space *mapping = page->mapping;
603 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 struct nfs_inode *nfsi = NFS_I(inode);
605 struct nfs_page *req, *new = NULL;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400606 pgoff_t rqend, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 end = offset + bytes;
609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 for (;;) {
611 /* Loop over all inode entries and see if we find
612 * A request for the page we wish to update
613 */
614 spin_lock(&nfsi->req_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500615 req = nfs_page_find_request_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (req) {
617 if (!nfs_lock_request_dontget(req)) {
618 int error;
Trond Myklebust277459d2006-12-05 00:35:35 -0500619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 spin_unlock(&nfsi->req_lock);
621 error = nfs_wait_on_request(req);
622 nfs_release_request(req);
Neil Brown1dd594b2006-03-20 13:44:04 -0500623 if (error < 0) {
624 if (new)
625 nfs_release_request(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 return ERR_PTR(error);
Neil Brown1dd594b2006-03-20 13:44:04 -0500627 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 continue;
629 }
630 spin_unlock(&nfsi->req_lock);
631 if (new)
632 nfs_release_request(new);
633 break;
634 }
635
636 if (new) {
637 int error;
638 nfs_lock_request_dontget(new);
639 error = nfs_inode_add_request(inode, new);
640 if (error) {
641 spin_unlock(&nfsi->req_lock);
642 nfs_unlock_request(new);
643 return ERR_PTR(error);
644 }
645 spin_unlock(&nfsi->req_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 return new;
647 }
648 spin_unlock(&nfsi->req_lock);
649
650 new = nfs_create_request(ctx, inode, page, offset, bytes);
651 if (IS_ERR(new))
652 return new;
653 }
654
655 /* We have a request for our page.
656 * If the creds don't match, or the
657 * page addresses don't match,
658 * tell the caller to wait on the conflicting
659 * request.
660 */
661 rqend = req->wb_offset + req->wb_bytes;
662 if (req->wb_context != ctx
663 || req->wb_page != page
664 || !nfs_dirty_request(req)
665 || offset > rqend || end < req->wb_offset) {
666 nfs_unlock_request(req);
667 return ERR_PTR(-EBUSY);
668 }
669
670 /* Okay, the request matches. Update the region */
671 if (offset < req->wb_offset) {
672 req->wb_offset = offset;
673 req->wb_pgbase = offset;
674 req->wb_bytes = rqend - req->wb_offset;
675 }
676
677 if (end > rqend)
678 req->wb_bytes = end - req->wb_offset;
679
680 return req;
681}
682
683int nfs_flush_incompatible(struct file *file, struct page *page)
684{
685 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500687 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 /*
689 * Look for a request corresponding to this page. If there
690 * is one, and it belongs to another file, we flush it out
691 * before we try to copy anything into the page. Do this
692 * due to the lack of an ACCESS-type call in NFSv2.
693 * Also do the same if we find a request from an existing
694 * dropped page.
695 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500696 do {
697 req = nfs_page_find_request(page);
698 if (req == NULL)
699 return 0;
700 do_flush = req->wb_page != page || req->wb_context != ctx
Trond Myklebuste261f512006-12-05 00:35:41 -0500701 || !nfs_dirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500703 if (!do_flush)
704 return 0;
705 status = nfs_wb_page(page->mapping->host, page);
706 } while (status == 0);
707 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
710/*
711 * Update and possibly write a cached page of an NFS file.
712 *
713 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
714 * things with a page scheduled for an RPC call (e.g. invalidate it).
715 */
716int nfs_updatepage(struct file *file, struct page *page,
717 unsigned int offset, unsigned int count)
718{
719 struct nfs_open_context *ctx = (struct nfs_open_context *)file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 int status = 0;
722
Chuck Lever91d5b472006-03-20 13:44:14 -0500723 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 dprintk("NFS: nfs_updatepage(%s/%s %d@%Ld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800726 file->f_path.dentry->d_parent->d_name.name,
727 file->f_path.dentry->d_name.name, count,
Chuck Lever0bbacc42005-11-01 16:53:32 -0500728 (long long)(page_offset(page) +offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 /* If we're not using byte range locks, and we know the page
731 * is entirely in cache, it may be more efficient to avoid
732 * fragmenting write requests.
733 */
Trond Myklebustab0a3db2005-06-22 17:16:30 +0000734 if (PageUptodate(page) && inode->i_flock == NULL && !(file->f_mode & O_SYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500735 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
738
Trond Myklebuste21195a2006-12-05 00:35:39 -0500739 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebuste261f512006-12-05 00:35:41 -0500740 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 dprintk("NFS: nfs_updatepage returns %d (isize %Ld)\n",
743 status, (long long)i_size_read(inode));
744 if (status < 0)
Trond Myklebusta301b772007-02-06 11:07:15 -0800745 nfs_set_pageerror(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 return status;
747}
748
749static void nfs_writepage_release(struct nfs_page *req)
750{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
Trond Myklebust44dd1512007-05-19 11:58:03 -0400752 if (PageError(req->wb_page)) {
753 nfs_end_page_writeback(req->wb_page);
754 nfs_inode_remove_request(req);
755 } else if (!nfs_reschedule_unstable_write(req)) {
756 /* Set the PG_uptodate flag */
757 nfs_mark_uptodate(req->wb_page, req->wb_pgbase, req->wb_bytes);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400758 nfs_end_page_writeback(req->wb_page);
759 nfs_inode_remove_request(req);
760 } else
761 nfs_end_page_writeback(req->wb_page);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000762 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763}
764
765static inline int flush_task_priority(int how)
766{
767 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
768 case FLUSH_HIGHPRI:
769 return RPC_PRIORITY_HIGH;
770 case FLUSH_LOWPRI:
771 return RPC_PRIORITY_LOW;
772 }
773 return RPC_PRIORITY_NORMAL;
774}
775
776/*
777 * Set up the argument/result storage required for the RPC call.
778 */
779static void nfs_write_rpcsetup(struct nfs_page *req,
780 struct nfs_write_data *data,
Trond Myklebust788e7a82006-03-20 13:44:27 -0500781 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 unsigned int count, unsigned int offset,
783 int how)
784{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 struct inode *inode;
Trond Myklebust788e7a82006-03-20 13:44:27 -0500786 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 /* Set up the RPC argument and reply structs
789 * NB: take care not to mess about with data->commit et al. */
790
791 data->req = req;
Trond Myklebust88be9f92007-06-05 10:42:27 -0400792 data->inode = inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 data->cred = req->wb_context->cred;
794
795 data->args.fh = NFS_FH(inode);
796 data->args.offset = req_offset(req) + offset;
797 data->args.pgbase = req->wb_pgbase + offset;
798 data->args.pages = data->pagevec;
799 data->args.count = count;
800 data->args.context = req->wb_context;
801
802 data->res.fattr = &data->fattr;
803 data->res.count = count;
804 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400805 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Trond Myklebust788e7a82006-03-20 13:44:27 -0500807 /* Set up the initial task struct. */
808 flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
809 rpc_init_task(&data->task, NFS_CLIENT(inode), flags, call_ops, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 NFS_PROTO(inode)->write_setup(data, how);
811
812 data->task.tk_priority = flush_task_priority(how);
813 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Chuck Levera3f565b2007-01-31 12:14:01 -0500815 dprintk("NFS: %5u initiated write call "
816 "(req %s/%Ld, %u bytes @ offset %Lu)\n",
Chuck Lever0bbacc42005-11-01 16:53:32 -0500817 data->task.tk_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 inode->i_sb->s_id,
819 (long long)NFS_FILEID(inode),
820 count,
821 (unsigned long long)data->args.offset);
822}
823
824static void nfs_execute_write(struct nfs_write_data *data)
825{
826 struct rpc_clnt *clnt = NFS_CLIENT(data->inode);
827 sigset_t oldset;
828
829 rpc_clnt_sigmask(clnt, &oldset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 rpc_execute(&data->task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 rpc_clnt_sigunmask(clnt, &oldset);
832}
833
834/*
835 * Generate multiple small requests to write out a single
836 * contiguous dirty area on one page.
837 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400838static 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 -0700839{
840 struct nfs_page *req = nfs_list_entry(head->next);
841 struct page *page = req->wb_page;
842 struct nfs_write_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700843 size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
844 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 int requests = 0;
846 LIST_HEAD(list);
847
848 nfs_list_remove_request(req);
849
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400850 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700851 do {
852 size_t len = min(nbytes, wsize);
853
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400854 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 if (!data)
856 goto out_bad;
857 list_add(&data->pages, &list);
858 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700859 nbytes -= len;
860 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 atomic_set(&req->wb_complete, requests);
862
863 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400865 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 do {
867 data = list_entry(list.next, struct nfs_write_data, pages);
868 list_del_init(&data->pages);
869
870 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400872 if (nbytes < wsize)
873 wsize = nbytes;
874 nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
875 wsize, offset, how);
876 offset += wsize;
877 nbytes -= wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 nfs_execute_write(data);
879 } while (nbytes != 0);
880
881 return 0;
882
883out_bad:
884 while (!list_empty(&list)) {
885 data = list_entry(list.next, struct nfs_write_data, pages);
886 list_del(&data->pages);
Trond Myklebust8aca67f2006-11-13 16:23:44 -0500887 nfs_writedata_release(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500889 nfs_redirty_request(req);
Trond Myklebust6d677e32007-04-20 16:12:40 -0400890 nfs_end_page_writeback(req->wb_page);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000891 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 return -ENOMEM;
893}
894
895/*
896 * Create an RPC task for the given write request and kick it.
897 * The page must have been locked by the caller.
898 *
899 * It may happen that the page we're passed is not marked dirty.
900 * This is the case if nfs_updatepage detects a conflicting request
901 * that has been written but not committed.
902 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400903static 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 -0700904{
905 struct nfs_page *req;
906 struct page **pages;
907 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400909 data = nfs_writedata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 if (!data)
911 goto out_bad;
912
913 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 while (!list_empty(head)) {
915 req = nfs_list_entry(head->next);
916 nfs_list_remove_request(req);
917 nfs_list_add_request(req, &data->pages);
918 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 }
921 req = nfs_list_entry(data->pages.next);
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 /* Set up the argument struct */
Trond Myklebust788e7a82006-03-20 13:44:27 -0500924 nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925
926 nfs_execute_write(data);
927 return 0;
928 out_bad:
929 while (!list_empty(head)) {
Trond Myklebust10afec92007-05-14 17:16:04 -0400930 req = nfs_list_entry(head->next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 nfs_list_remove_request(req);
Trond Myklebust61822ab2006-12-05 00:35:42 -0500932 nfs_redirty_request(req);
Trond Myklebust6d677e32007-04-20 16:12:40 -0400933 nfs_end_page_writeback(req->wb_page);
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000934 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 }
936 return -ENOMEM;
937}
938
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400939static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
940 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Trond Myklebust7d46a492006-03-20 13:44:50 -0500942 int wsize = NFS_SERVER(inode)->wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400944 if (wsize < PAGE_CACHE_SIZE)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400945 nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400946 else
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400947 nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948}
949
950/*
951 * Handle a write reply that flushed part of a page.
952 */
Trond Myklebust788e7a82006-03-20 13:44:27 -0500953static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Trond Myklebust788e7a82006-03-20 13:44:27 -0500955 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 struct nfs_page *req = data->req;
957 struct page *page = req->wb_page;
958
959 dprintk("NFS: write (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -0400960 req->wb_context->path.dentry->d_inode->i_sb->s_id,
961 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 req->wb_bytes,
963 (long long)req_offset(req));
964
Trond Myklebust788e7a82006-03-20 13:44:27 -0500965 if (nfs_writeback_done(task, data) != 0)
966 return;
967
968 if (task->tk_status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -0800969 nfs_set_pageerror(page);
Trond Myklebust788e7a82006-03-20 13:44:27 -0500970 req->wb_context->error = task->tk_status;
971 dprintk(", error = %d\n", task->tk_status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400972 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 }
974
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400975 if (nfs_write_need_commit(data)) {
976 spinlock_t *req_lock = &NFS_I(page->mapping->host)->req_lock;
977
978 spin_lock(req_lock);
979 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
980 /* Do nothing we need to resend the writes */
981 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
982 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
983 dprintk(" defer commit\n");
984 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
985 set_bit(PG_NEED_RESCHED, &req->wb_flags);
986 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
987 dprintk(" server reboot detected\n");
988 }
989 spin_unlock(req_lock);
990 } else
991 dprintk(" OK\n");
992
993out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 if (atomic_dec_and_test(&req->wb_complete))
995 nfs_writepage_release(req);
996}
997
Trond Myklebust788e7a82006-03-20 13:44:27 -0500998static const struct rpc_call_ops nfs_write_partial_ops = {
999 .rpc_call_done = nfs_writeback_done_partial,
1000 .rpc_release = nfs_writedata_release,
1001};
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003/*
1004 * Handle a write reply that flushes a whole page.
1005 *
1006 * FIXME: There is an inherent race with invalidate_inode_pages and
1007 * writebacks since the page->count is kept > 1 for as long
1008 * as the page has a write request pending.
1009 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001010static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001012 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 struct nfs_page *req;
1014 struct page *page;
1015
Trond Myklebust788e7a82006-03-20 13:44:27 -05001016 if (nfs_writeback_done(task, data) != 0)
1017 return;
1018
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /* Update attributes as result of writeback. */
1020 while (!list_empty(&data->pages)) {
1021 req = nfs_list_entry(data->pages.next);
1022 nfs_list_remove_request(req);
1023 page = req->wb_page;
1024
1025 dprintk("NFS: write (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001026 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1027 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 req->wb_bytes,
1029 (long long)req_offset(req));
1030
Trond Myklebust788e7a82006-03-20 13:44:27 -05001031 if (task->tk_status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001032 nfs_set_pageerror(page);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001033 req->wb_context->error = task->tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001034 dprintk(", error = %d\n", task->tk_status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001035 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001038 if (nfs_write_need_commit(data)) {
1039 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1040 nfs_mark_request_commit(req);
1041 nfs_end_page_writeback(page);
1042 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 goto next;
1044 }
Trond Myklebust44dd1512007-05-19 11:58:03 -04001045 /* Set the PG_uptodate flag? */
1046 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001047 dprintk(" OK\n");
1048remove_request:
1049 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 next:
Trond Myklebustc6a556b2005-06-22 17:16:30 +00001052 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
1054}
1055
Trond Myklebust788e7a82006-03-20 13:44:27 -05001056static const struct rpc_call_ops nfs_write_full_ops = {
1057 .rpc_call_done = nfs_writeback_done_full,
1058 .rpc_release = nfs_writedata_release,
1059};
1060
1061
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062/*
1063 * This function is called when the WRITE call is complete.
1064 */
Chuck Lever462d5b32006-03-20 13:44:32 -05001065int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 struct nfs_writeargs *argp = &data->args;
1068 struct nfs_writeres *resp = &data->res;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001069 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Chuck Levera3f565b2007-01-31 12:14:01 -05001071 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 task->tk_pid, task->tk_status);
1073
Chuck Leverf551e442006-09-20 14:33:04 -04001074 /*
1075 * ->write_done will attempt to use post-op attributes to detect
1076 * conflicting writes by other clients. A strict interpretation
1077 * of close-to-open would allow us to continue caching even if
1078 * another writer had changed the file, but some applications
1079 * depend on tighter cache coherency when writing.
1080 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001081 status = NFS_PROTO(data->inode)->write_done(task, data);
1082 if (status != 0)
1083 return status;
Chuck Lever91d5b472006-03-20 13:44:14 -05001084 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1087 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1088 /* We tried a write call, but the server did not
1089 * commit data to stable storage even though we
1090 * requested it.
1091 * Note: There is a known bug in Tru64 < 5.0 in which
1092 * the server reports NFS_DATA_SYNC, but performs
1093 * NFS_FILE_SYNC. We therefore implement this checking
1094 * as a dprintk() in order to avoid filling syslog.
1095 */
1096 static unsigned long complain;
1097
1098 if (time_before(complain, jiffies)) {
1099 dprintk("NFS: faulty NFS server %s:"
1100 " (committed = %d) != (stable = %d)\n",
David Howells54ceac42006-08-22 20:06:13 -04001101 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 resp->verf->committed, argp->stable);
1103 complain = jiffies + 300 * HZ;
1104 }
1105 }
1106#endif
1107 /* Is this a short write? */
1108 if (task->tk_status >= 0 && resp->count < argp->count) {
1109 static unsigned long complain;
1110
Chuck Lever91d5b472006-03-20 13:44:14 -05001111 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1112
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 /* Has the server at least made some progress? */
1114 if (resp->count != 0) {
1115 /* Was this an NFSv2 write or an NFSv3 stable write? */
1116 if (resp->verf->committed != NFS_UNSTABLE) {
1117 /* Resend from where the server left off */
1118 argp->offset += resp->count;
1119 argp->pgbase += resp->count;
1120 argp->count -= resp->count;
1121 } else {
1122 /* Resend as a stable write in order to avoid
1123 * headaches in the case of a server crash.
1124 */
1125 argp->stable = NFS_FILE_SYNC;
1126 }
1127 rpc_restart_call(task);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001128 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130 if (time_before(complain, jiffies)) {
1131 printk(KERN_WARNING
1132 "NFS: Server wrote zero bytes, expected %u.\n",
1133 argp->count);
1134 complain = jiffies + 300 * HZ;
1135 }
1136 /* Can't do anything about it except throw an error. */
1137 task->tk_status = -EIO;
1138 }
Trond Myklebust788e7a82006-03-20 13:44:27 -05001139 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
1141
1142
1143#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001144void nfs_commit_release(void *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 nfs_commit_free(wdata);
1147}
1148
1149/*
1150 * Set up the argument/result storage required for the RPC call.
1151 */
1152static void nfs_commit_rpcsetup(struct list_head *head,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001153 struct nfs_write_data *data,
1154 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155{
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001156 struct nfs_page *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 struct inode *inode;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001158 int flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160 /* Set up the RPC argument and reply structs
1161 * NB: take care not to mess about with data->commit et al. */
1162
1163 list_splice_init(head, &data->pages);
1164 first = nfs_list_entry(data->pages.next);
Trond Myklebust88be9f92007-06-05 10:42:27 -04001165 inode = first->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 data->inode = inode;
1168 data->cred = first->wb_context->cred;
1169
1170 data->args.fh = NFS_FH(data->inode);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001171 /* Note: we always request a commit of the entire inode */
1172 data->args.offset = 0;
1173 data->args.count = 0;
1174 data->res.count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 data->res.fattr = &data->fattr;
1176 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -04001177 nfs_fattr_init(&data->fattr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001178
1179 /* Set up the initial task struct. */
1180 flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
1181 rpc_init_task(&data->task, NFS_CLIENT(inode), flags, &nfs_commit_ops, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 NFS_PROTO(inode)->commit_setup(data, how);
1183
1184 data->task.tk_priority = flush_task_priority(how);
1185 data->task.tk_cookie = (unsigned long)inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Chuck Levera3f565b2007-01-31 12:14:01 -05001187 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188}
1189
1190/*
1191 * Commit dirty pages
1192 */
1193static int
Chuck Lever40859d72005-11-30 18:09:02 -05001194nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195{
1196 struct nfs_write_data *data;
1197 struct nfs_page *req;
1198
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001199 data = nfs_commit_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 if (!data)
1202 goto out_bad;
1203
1204 /* Set up the argument struct */
1205 nfs_commit_rpcsetup(head, data, how);
1206
1207 nfs_execute_write(data);
1208 return 0;
1209 out_bad:
1210 while (!list_empty(head)) {
1211 req = nfs_list_entry(head->next);
1212 nfs_list_remove_request(req);
1213 nfs_mark_request_commit(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001214 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Trond Myklebust5c2d97c2006-09-18 23:20:35 -04001215 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217 return -ENOMEM;
1218}
1219
1220/*
1221 * COMMIT call returned
1222 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001223static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001225 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 struct nfs_page *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
Chuck Levera3f565b2007-01-31 12:14:01 -05001228 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 task->tk_pid, task->tk_status);
1230
Trond Myklebust788e7a82006-03-20 13:44:27 -05001231 /* Call the NFS version-specific code */
1232 if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1233 return;
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 while (!list_empty(&data->pages)) {
1236 req = nfs_list_entry(data->pages.next);
1237 nfs_list_remove_request(req);
Trond Myklebust612c9382007-04-20 16:12:45 -04001238 clear_bit(PG_NEED_COMMIT, &(req)->wb_flags);
Christoph Lameterfd39fc82006-06-30 01:55:40 -07001239 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240
1241 dprintk("NFS: commit (%s/%Ld %d@%Ld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001242 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1243 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 req->wb_bytes,
1245 (long long)req_offset(req));
1246 if (task->tk_status < 0) {
1247 req->wb_context->error = task->tk_status;
1248 nfs_inode_remove_request(req);
1249 dprintk(", error = %d\n", task->tk_status);
1250 goto next;
1251 }
1252
1253 /* Okay, COMMIT succeeded, apparently. Check the verifier
1254 * returned by the server against all stored verfs. */
1255 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1256 /* We have a match */
Trond Myklebust44dd1512007-05-19 11:58:03 -04001257 /* Set the PG_uptodate flag */
1258 nfs_mark_uptodate(req->wb_page, req->wb_pgbase,
1259 req->wb_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 nfs_inode_remove_request(req);
1261 dprintk(" OK\n");
1262 goto next;
1263 }
1264 /* We have a mismatch. Write the page again */
1265 dprintk(" mismatch\n");
Trond Myklebust61822ab2006-12-05 00:35:42 -05001266 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 next:
Trond Myklebustc6a556b2005-06-22 17:16:30 +00001268 nfs_clear_page_writeback(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001271
1272static const struct rpc_call_ops nfs_commit_ops = {
1273 .rpc_call_done = nfs_commit_done,
1274 .rpc_release = nfs_commit_release,
1275};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001277int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
1279 struct nfs_inode *nfsi = NFS_I(inode);
1280 LIST_HEAD(head);
Trond Myklebust7d46a492006-03-20 13:44:50 -05001281 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 spin_lock(&nfsi->req_lock);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001284 res = nfs_scan_commit(inode, &head, 0, 0);
1285 spin_unlock(&nfsi->req_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (res) {
Trond Myklebust7d46a492006-03-20 13:44:50 -05001287 int error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001288 if (error < 0)
1289 return error;
1290 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 return res;
1292}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001293#else
1294static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1295{
1296 return 0;
1297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298#endif
1299
Trond Myklebust1c759502006-10-09 16:18:38 -04001300long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
Trond Myklebust1c759502006-10-09 16:18:38 -04001302 struct inode *inode = mapping->host;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001303 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustca52fec2007-04-17 17:22:13 -04001304 pgoff_t idx_start, idx_end;
Trond Myklebust1c759502006-10-09 16:18:38 -04001305 unsigned int npages = 0;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001306 LIST_HEAD(head);
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001307 int nocommit = how & FLUSH_NOCOMMIT;
Trond Myklebust3f442542006-09-17 14:46:44 -04001308 long pages, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Trond Myklebust1c759502006-10-09 16:18:38 -04001310 /* FIXME */
1311 if (wbc->range_cyclic)
1312 idx_start = 0;
1313 else {
1314 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
1315 idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
1316 if (idx_end > idx_start) {
Trond Myklebustca52fec2007-04-17 17:22:13 -04001317 pgoff_t l_npages = 1 + idx_end - idx_start;
Trond Myklebust1c759502006-10-09 16:18:38 -04001318 npages = l_npages;
1319 if (sizeof(npages) != sizeof(l_npages) &&
Trond Myklebustca52fec2007-04-17 17:22:13 -04001320 (pgoff_t)npages != l_npages)
Trond Myklebust1c759502006-10-09 16:18:38 -04001321 npages = 0;
1322 }
1323 }
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001324 how &= ~FLUSH_NOCOMMIT;
1325 spin_lock(&nfsi->req_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 do {
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001327 ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
1328 if (ret != 0)
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001329 continue;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001330 if (nocommit)
1331 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001332 pages = nfs_scan_commit(inode, &head, idx_start, npages);
Trond Myklebust724c4392007-04-17 17:22:13 -04001333 if (pages == 0)
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001334 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001335 if (how & FLUSH_INVALIDATE) {
1336 spin_unlock(&nfsi->req_lock);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001337 nfs_cancel_commit_list(&head);
Trond Myklebuste8e058e2006-11-15 17:31:56 -05001338 ret = pages;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001339 spin_lock(&nfsi->req_lock);
1340 continue;
1341 }
1342 pages += nfs_scan_commit(inode, &head, 0, 0);
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001343 spin_unlock(&nfsi->req_lock);
1344 ret = nfs_commit_list(inode, &head, how);
1345 spin_lock(&nfsi->req_lock);
1346 } while (ret >= 0);
1347 spin_unlock(&nfsi->req_lock);
1348 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349}
1350
Trond Myklebust1c759502006-10-09 16:18:38 -04001351/*
1352 * flush the inode to disk.
1353 */
1354int nfs_wb_all(struct inode *inode)
1355{
1356 struct address_space *mapping = inode->i_mapping;
1357 struct writeback_control wbc = {
1358 .bdi = mapping->backing_dev_info,
1359 .sync_mode = WB_SYNC_ALL,
1360 .nr_to_write = LONG_MAX,
Trond Myklebust61822ab2006-12-05 00:35:42 -05001361 .for_writepages = 1,
Trond Myklebust1c759502006-10-09 16:18:38 -04001362 .range_cyclic = 1,
1363 };
1364 int ret;
1365
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001366 ret = nfs_writepages(mapping, &wbc);
Trond Myklebust61822ab2006-12-05 00:35:42 -05001367 if (ret < 0)
1368 goto out;
Trond Myklebust1c759502006-10-09 16:18:38 -04001369 ret = nfs_sync_mapping_wait(mapping, &wbc, 0);
1370 if (ret >= 0)
1371 return 0;
Trond Myklebust61822ab2006-12-05 00:35:42 -05001372out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001373 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Trond Myklebust1c759502006-10-09 16:18:38 -04001374 return ret;
1375}
1376
1377int nfs_sync_mapping_range(struct address_space *mapping, loff_t range_start, loff_t range_end, int how)
1378{
1379 struct writeback_control wbc = {
1380 .bdi = mapping->backing_dev_info,
1381 .sync_mode = WB_SYNC_ALL,
1382 .nr_to_write = LONG_MAX,
1383 .range_start = range_start,
1384 .range_end = range_end,
Trond Myklebust61822ab2006-12-05 00:35:42 -05001385 .for_writepages = 1,
Trond Myklebust1c759502006-10-09 16:18:38 -04001386 };
1387 int ret;
1388
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001389 ret = nfs_writepages(mapping, &wbc);
1390 if (ret < 0)
1391 goto out;
Trond Myklebust1c759502006-10-09 16:18:38 -04001392 ret = nfs_sync_mapping_wait(mapping, &wbc, how);
1393 if (ret >= 0)
1394 return 0;
Trond Myklebust61822ab2006-12-05 00:35:42 -05001395out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001396 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Trond Myklebust1c759502006-10-09 16:18:38 -04001397 return ret;
1398}
1399
Trond Myklebust61822ab2006-12-05 00:35:42 -05001400int nfs_wb_page_priority(struct inode *inode, struct page *page, int how)
Trond Myklebust1c759502006-10-09 16:18:38 -04001401{
1402 loff_t range_start = page_offset(page);
1403 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001404 struct writeback_control wbc = {
1405 .bdi = page->mapping->backing_dev_info,
1406 .sync_mode = WB_SYNC_ALL,
1407 .nr_to_write = LONG_MAX,
1408 .range_start = range_start,
1409 .range_end = range_end,
1410 };
1411 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001412
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001413 BUG_ON(!PageLocked(page));
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001414 if (clear_page_dirty_for_io(page)) {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001415 ret = nfs_writepage_locked(page, &wbc);
1416 if (ret < 0)
1417 goto out;
1418 }
Trond Myklebustf40313a2007-01-13 02:28:08 -05001419 if (!PagePrivate(page))
1420 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001421 ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
1422 if (ret >= 0)
1423 return 0;
1424out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001425 __mark_inode_dirty(inode, I_DIRTY_PAGES);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001426 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001427}
1428
1429/*
1430 * Write back all requests on one page - we do this before reading it.
1431 */
1432int nfs_wb_page(struct inode *inode, struct page* page)
1433{
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001434 return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
Trond Myklebust1c759502006-10-09 16:18:38 -04001435}
1436
Trond Myklebust1a545332006-12-05 00:35:40 -05001437int nfs_set_page_dirty(struct page *page)
1438{
Trond Myklebustd5851582007-04-30 22:17:02 -07001439 struct address_space *mapping = page->mapping;
1440 struct inode *inode;
1441 spinlock_t *req_lock;
Trond Myklebust1a545332006-12-05 00:35:40 -05001442 struct nfs_page *req;
Trond Myklebust2b82f192007-04-20 16:12:50 -04001443 int ret;
Trond Myklebust1a545332006-12-05 00:35:40 -05001444
Trond Myklebustd5851582007-04-30 22:17:02 -07001445 if (!mapping)
1446 goto out_raced;
1447 inode = mapping->host;
1448 if (!inode)
1449 goto out_raced;
1450 req_lock = &NFS_I(inode)->req_lock;
Trond Myklebust2b82f192007-04-20 16:12:50 -04001451 spin_lock(req_lock);
1452 req = nfs_page_find_request_locked(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001453 if (req != NULL) {
1454 /* Mark any existing write requests for flushing */
Trond Myklebust2b82f192007-04-20 16:12:50 -04001455 ret = !test_and_set_bit(PG_NEED_FLUSH, &req->wb_flags);
1456 spin_unlock(req_lock);
Trond Myklebust1a545332006-12-05 00:35:40 -05001457 nfs_release_request(req);
Trond Myklebust2b82f192007-04-20 16:12:50 -04001458 return ret;
Trond Myklebust1a545332006-12-05 00:35:40 -05001459 }
Trond Myklebust2b82f192007-04-20 16:12:50 -04001460 ret = __set_page_dirty_nobuffers(page);
1461 spin_unlock(req_lock);
1462 return ret;
Trond Myklebustd5851582007-04-30 22:17:02 -07001463out_raced:
1464 return !TestSetPageDirty(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001465}
1466
Trond Myklebust1c759502006-10-09 16:18:38 -04001467
David Howellsf7b422b2006-06-09 09:34:33 -04001468int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
1470 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1471 sizeof(struct nfs_write_data),
1472 0, SLAB_HWCACHE_ALIGN,
1473 NULL, NULL);
1474 if (nfs_wdata_cachep == NULL)
1475 return -ENOMEM;
1476
Matthew Dobson93d23412006-03-26 01:37:50 -08001477 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1478 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 if (nfs_wdata_mempool == NULL)
1480 return -ENOMEM;
1481
Matthew Dobson93d23412006-03-26 01:37:50 -08001482 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1483 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 if (nfs_commit_mempool == NULL)
1485 return -ENOMEM;
1486
Peter Zijlstra89a09142007-03-16 13:38:26 -08001487 /*
1488 * NFS congestion size, scale with available memory.
1489 *
1490 * 64MB: 8192k
1491 * 128MB: 11585k
1492 * 256MB: 16384k
1493 * 512MB: 23170k
1494 * 1GB: 32768k
1495 * 2GB: 46340k
1496 * 4GB: 65536k
1497 * 8GB: 92681k
1498 * 16GB: 131072k
1499 *
1500 * This allows larger machines to have larger/more transfers.
1501 * Limit the default to 256M
1502 */
1503 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1504 if (nfs_congestion_kb > 256*1024)
1505 nfs_congestion_kb = 256*1024;
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return 0;
1508}
1509
David Brownell266bee82006-06-27 12:59:15 -07001510void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
1512 mempool_destroy(nfs_commit_mempool);
1513 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001514 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516