blob: e40e949598fd29e7eb2fa6dcae0bf2e8fa8d9a5f [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>
Trond Myklebust074cc1d2009-08-10 08:54:13 -040016#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070022#include <linux/backing-dev.h>
23
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050027#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050028#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040029#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040030#include "fscache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32#define NFSDBG_FACILITY NFSDBG_PAGECACHE
33
34#define MIN_POOL_WRITE (32)
35#define MIN_POOL_COMMIT (4)
36
37/*
38 * Local function declarations
39 */
Trond Myklebustc63c7b02007-04-02 19:29:52 -040040static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
41 struct inode *inode, int ioflags);
Fred Isamanf8512ad2008-03-19 11:24:39 -040042static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050043static const struct rpc_call_ops nfs_write_partial_ops;
44static const struct rpc_call_ops nfs_write_full_ops;
45static const struct rpc_call_ops nfs_commit_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Christoph Lametere18b8902006-12-06 20:33:20 -080047static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050048static mempool_t *nfs_wdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049static mempool_t *nfs_commit_mempool;
50
Trond Myklebustc9d8f892008-04-15 16:56:39 -040051struct nfs_write_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070052{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080053 struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
Chuck Lever40859d72005-11-30 18:09:02 -050054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 if (p) {
56 memset(p, 0, sizeof(*p));
57 INIT_LIST_HEAD(&p->pages);
Andy Adamson5f7dbd52009-04-01 09:22:05 -040058 p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 }
60 return p;
61}
62
Trond Myklebust5e4424a2008-02-25 21:53:49 -080063void nfs_commit_free(struct nfs_write_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064{
Chuck Lever40859d72005-11-30 18:09:02 -050065 if (p && (p->pagevec != &p->page_array[0]))
66 kfree(p->pagevec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 mempool_free(p, nfs_commit_mempool);
68}
69
Trond Myklebust8d5658c2007-04-10 09:26:35 -040070struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050071{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080072 struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050073
74 if (p) {
75 memset(p, 0, sizeof(*p));
76 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070077 p->npages = pagecount;
Andy Adamson5f7dbd52009-04-01 09:22:05 -040078 p->res.seq_res.sr_slotid = NFS4_MAX_SLOT_TABLE;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040079 if (pagecount <= ARRAY_SIZE(p->page_array))
80 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050081 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040082 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
83 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050084 mempool_free(p, nfs_wdata_mempool);
85 p = NULL;
86 }
87 }
88 }
89 return p;
90}
91
Trond Myklebust1ae88b22009-08-12 09:12:30 -040092void nfs_writedata_free(struct nfs_write_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050093{
94 if (p && (p->pagevec != &p->page_array[0]))
95 kfree(p->pagevec);
96 mempool_free(p, nfs_wdata_mempool);
97}
98
Trond Myklebust1ae88b22009-08-12 09:12:30 -040099static void nfs_writedata_release(struct nfs_write_data *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
Trond Myklebust383ba712008-02-19 20:04:20 -0500101 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 nfs_writedata_free(wdata);
103}
104
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400105static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
106{
107 ctx->error = error;
108 smp_wmb();
109 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
110}
111
Trond Myklebust277459d2006-12-05 00:35:35 -0500112static struct nfs_page *nfs_page_find_request_locked(struct page *page)
113{
114 struct nfs_page *req = NULL;
115
116 if (PagePrivate(page)) {
117 req = (struct nfs_page *)page_private(page);
118 if (req != NULL)
Trond Myklebustc03b4022007-06-17 13:26:38 -0400119 kref_get(&req->wb_kref);
Trond Myklebust277459d2006-12-05 00:35:35 -0500120 }
121 return req;
122}
123
124static struct nfs_page *nfs_page_find_request(struct page *page)
125{
Trond Myklebust587142f2007-07-02 09:57:54 -0400126 struct inode *inode = page->mapping->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500127 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500128
Trond Myklebust587142f2007-07-02 09:57:54 -0400129 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500130 req = nfs_page_find_request_locked(page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400131 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500132 return req;
133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/* Adjust the file length if we're writing beyond the end */
136static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
137{
138 struct inode *inode = page->mapping->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400139 loff_t end, i_size;
140 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Trond Myklebusta3d01452008-06-11 12:21:19 -0400142 spin_lock(&inode->i_lock);
143 i_size = i_size_read(inode);
144 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (i_size > 0 && page->index < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400146 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
148 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400149 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400151 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
152out:
153 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154}
155
Trond Myklebusta301b772007-02-06 11:07:15 -0800156/* A writeback failed: mark the page as bad, and invalidate the page cache */
157static void nfs_set_pageerror(struct page *page)
158{
159 SetPageError(page);
160 nfs_zap_mapping(page->mapping->host, page->mapping);
161}
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163/* We can set the PG_uptodate flag if we see that a write request
164 * covers the full page.
165 */
166static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
167{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 if (PageUptodate(page))
169 return;
170 if (base != 0)
171 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500172 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500174 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175}
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static int wb_priority(struct writeback_control *wbc)
178{
179 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400180 return FLUSH_HIGHPRI | FLUSH_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100181 if (wbc->for_kupdate || wbc->for_background)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 return FLUSH_LOWPRI;
183 return 0;
184}
185
186/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800187 * NFS congestion control
188 */
189
190int nfs_congestion_kb;
191
192#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
193#define NFS_CONGESTION_OFF_THRESH \
194 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
195
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400196static int nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800197{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400198 int ret = test_set_page_writeback(page);
199
200 if (!ret) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800201 struct inode *inode = page->mapping->host;
202 struct nfs_server *nfss = NFS_SERVER(inode);
203
Peter Zijlstra277866a2007-05-08 00:35:12 -0700204 if (atomic_long_inc_return(&nfss->writeback) >
Jens Axboe8aa7e842009-07-09 14:52:32 +0200205 NFS_CONGESTION_ON_THRESH) {
206 set_bdi_congested(&nfss->backing_dev_info,
207 BLK_RW_ASYNC);
208 }
Peter Zijlstra89a09142007-03-16 13:38:26 -0800209 }
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400210 return ret;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800211}
212
213static void nfs_end_page_writeback(struct page *page)
214{
215 struct inode *inode = page->mapping->host;
216 struct nfs_server *nfss = NFS_SERVER(inode);
217
218 end_page_writeback(page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700219 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200220 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800221}
222
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400223static struct nfs_page *nfs_find_and_lock_request(struct page *page)
Trond Myklebuste261f512006-12-05 00:35:41 -0500224{
Trond Myklebust587142f2007-07-02 09:57:54 -0400225 struct inode *inode = page->mapping->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500226 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500227 int ret;
228
Trond Myklebust587142f2007-07-02 09:57:54 -0400229 spin_lock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400230 for (;;) {
Trond Myklebuste261f512006-12-05 00:35:41 -0500231 req = nfs_page_find_request_locked(page);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400232 if (req == NULL)
233 break;
Trond Myklebustacee4782008-01-22 17:13:07 -0500234 if (nfs_set_page_tag_locked(req))
Trond Myklebuste261f512006-12-05 00:35:41 -0500235 break;
236 /* Note: If we hold the page lock, as is the case in nfs_writepage,
Trond Myklebustacee4782008-01-22 17:13:07 -0500237 * then the call to nfs_set_page_tag_locked() will always
Trond Myklebuste261f512006-12-05 00:35:41 -0500238 * succeed provided that someone hasn't already marked the
239 * request as dirty (in which case we don't care).
240 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400241 spin_unlock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500242 ret = nfs_wait_on_request(req);
243 nfs_release_request(req);
244 if (ret != 0)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400245 return ERR_PTR(ret);
Trond Myklebust587142f2007-07-02 09:57:54 -0400246 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500247 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400248 spin_unlock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400249 return req;
250}
251
252/*
253 * Find an associated nfs write request, and prepare to flush it out
254 * May return an error if the user signalled nfs_wait_on_request().
255 */
256static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
257 struct page *page)
258{
259 struct nfs_page *req;
260 int ret = 0;
261
262 req = nfs_find_and_lock_request(page);
263 if (!req)
264 goto out;
265 ret = PTR_ERR(req);
266 if (IS_ERR(req))
267 goto out;
268
269 ret = nfs_set_page_writeback(page);
270 BUG_ON(ret != 0);
271 BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));
272
Fred Isamanf8512ad2008-03-19 11:24:39 -0400273 if (!nfs_pageio_add_request(pgio, req)) {
274 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400275 ret = pgio->pg_error;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400276 }
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400277out:
278 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500279}
280
Trond Myklebustf758c8852007-07-22 19:27:32 -0400281static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
282{
283 struct inode *inode = page->mapping->host;
284
285 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
286 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
287
288 nfs_pageio_cond_complete(pgio, page->index);
289 return nfs_page_async_flush(pgio, page);
290}
291
Trond Myklebuste261f512006-12-05 00:35:41 -0500292/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 * Write an mmapped page to the server.
294 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500295static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400297 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500298 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Trond Myklebustf758c8852007-07-22 19:27:32 -0400300 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
301 err = nfs_do_writepage(page, wbc, &pgio);
302 nfs_pageio_complete(&pgio);
303 if (err < 0)
304 return err;
305 if (pgio.pg_error < 0)
306 return pgio.pg_error;
307 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500308}
309
310int nfs_writepage(struct page *page, struct writeback_control *wbc)
311{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400312 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500313
Trond Myklebustf758c8852007-07-22 19:27:32 -0400314 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400316 return ret;
317}
318
319static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
320{
321 int ret;
322
323 ret = nfs_do_writepage(page, wbc, data);
324 unlock_page(page);
325 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
329{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400331 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400332 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 int err;
334
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400335 /* Stop dirtying of new pages while we sync */
336 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
337 nfs_wait_bit_killable, TASK_KILLABLE);
338 if (err)
339 goto out_err;
340
Chuck Lever91d5b472006-03-20 13:44:14 -0500341 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
342
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400343 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
Trond Myklebustf758c8852007-07-22 19:27:32 -0400344 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400345 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400346
347 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
348 smp_mb__after_clear_bit();
349 wake_up_bit(bitlock, NFS_INO_FLUSHING);
350
Trond Myklebustf758c8852007-07-22 19:27:32 -0400351 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400352 goto out_err;
353 err = pgio.pg_error;
354 if (err < 0)
355 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400356 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400357out_err:
358 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359}
360
361/*
362 * Insert a write request into an inode
363 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400364static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
366 struct nfs_inode *nfsi = NFS_I(inode);
367 int error;
368
Trond Myklebuste7d39062008-06-13 12:12:32 -0400369 error = radix_tree_preload(GFP_NOFS);
370 if (error != 0)
371 goto out;
372
373 /* Lock the request! */
374 nfs_lock_request_dontget(req);
375
376 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
Nick Piggin27852592008-02-04 23:48:37 -0800378 BUG_ON(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (!nfsi->npages) {
380 igrab(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 if (nfs_have_delegation(inode, FMODE_WRITE))
382 nfsi->change_attr++;
383 }
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500384 SetPagePrivate(req->wb_page);
Trond Myklebust277459d2006-12-05 00:35:35 -0500385 set_page_private(req->wb_page, (unsigned long)req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400387 kref_get(&req->wb_kref);
Nick Piggin27852592008-02-04 23:48:37 -0800388 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
389 NFS_PAGE_TAG_LOCKED);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400390 spin_unlock(&inode->i_lock);
391 radix_tree_preload_end();
392out:
393 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
396/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800397 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 */
399static void nfs_inode_remove_request(struct nfs_page *req)
400{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400401 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 struct nfs_inode *nfsi = NFS_I(inode);
403
404 BUG_ON (!NFS_WBACK_BUSY(req));
405
Trond Myklebust587142f2007-07-02 09:57:54 -0400406 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500407 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500408 ClearPagePrivate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
410 nfsi->npages--;
411 if (!nfsi->npages) {
Trond Myklebust587142f2007-07-02 09:57:54 -0400412 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 iput(inode);
414 } else
Trond Myklebust587142f2007-07-02 09:57:54 -0400415 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 nfs_clear_request(req);
417 nfs_release_request(req);
418}
419
Trond Myklebust61822ab2006-12-05 00:35:42 -0500420static void
Fred6d884e82008-03-19 11:24:38 -0400421nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500422{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500423 __set_page_dirty_nobuffers(req->wb_page);
424}
425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
427/*
428 * Add a request to the inode's commit list.
429 */
430static void
431nfs_mark_request_commit(struct nfs_page *req)
432{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400433 struct inode *inode = req->wb_context->path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 struct nfs_inode *nfsi = NFS_I(inode);
435
Trond Myklebust587142f2007-07-02 09:57:54 -0400436 spin_lock(&inode->i_lock);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400437 set_bit(PG_CLEAN, &(req)->wb_flags);
Trond Myklebust5c369682007-06-17 15:27:42 -0400438 radix_tree_tag_set(&nfsi->nfs_page_tree,
439 req->wb_index,
440 NFS_PAGE_TAG_COMMIT);
Trond Myklebustff778d02010-02-19 16:53:39 -0800441 nfsi->ncommit++;
Trond Myklebust587142f2007-07-02 09:57:54 -0400442 spin_unlock(&inode->i_lock);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700443 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700444 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
Trond Myklebusta1803042006-12-05 00:45:19 -0500445 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400447
Trond Myklebuste468bae2008-06-13 13:25:22 -0400448static int
449nfs_clear_request_commit(struct nfs_page *req)
450{
451 struct page *page = req->wb_page;
452
453 if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
454 dec_zone_page_state(page, NR_UNSTABLE_NFS);
455 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
456 return 1;
457 }
458 return 0;
459}
460
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400461static inline
462int nfs_write_need_commit(struct nfs_write_data *data)
463{
464 return data->verf.committed != NFS_FILE_SYNC;
465}
466
467static inline
468int nfs_reschedule_unstable_write(struct nfs_page *req)
469{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400470 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400471 nfs_mark_request_commit(req);
472 return 1;
473 }
474 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
Fred6d884e82008-03-19 11:24:38 -0400475 nfs_mark_request_dirty(req);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400476 return 1;
477 }
478 return 0;
479}
480#else
481static inline void
482nfs_mark_request_commit(struct nfs_page *req)
483{
484}
485
Trond Myklebuste468bae2008-06-13 13:25:22 -0400486static inline int
487nfs_clear_request_commit(struct nfs_page *req)
488{
489 return 0;
490}
491
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400492static inline
493int nfs_write_need_commit(struct nfs_write_data *data)
494{
495 return 0;
496}
497
498static inline
499int nfs_reschedule_unstable_write(struct nfs_page *req)
500{
501 return 0;
502}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503#endif
504
505/*
506 * Wait for a request to complete.
507 *
Matthew Wilcox150030b2007-12-06 16:24:39 -0500508 * Interruptible by fatal signals only.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 */
Trond Myklebustca52fec2007-04-17 17:22:13 -0400510static int nfs_wait_on_requests_locked(struct inode *inode, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
512 struct nfs_inode *nfsi = NFS_I(inode);
513 struct nfs_page *req;
Trond Myklebustca52fec2007-04-17 17:22:13 -0400514 pgoff_t idx_end, next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 unsigned int res = 0;
516 int error;
517
518 if (npages == 0)
519 idx_end = ~0;
520 else
521 idx_end = idx_start + npages - 1;
522
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 next = idx_start;
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400524 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 -0700525 if (req->wb_index > idx_end)
526 break;
527
528 next = req->wb_index + 1;
Trond Myklebustc6a556b2005-06-22 17:16:30 +0000529 BUG_ON(!NFS_WBACK_BUSY(req));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
Trond Myklebustc03b4022007-06-17 13:26:38 -0400531 kref_get(&req->wb_kref);
Trond Myklebust587142f2007-07-02 09:57:54 -0400532 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 error = nfs_wait_on_request(req);
534 nfs_release_request(req);
Trond Myklebust587142f2007-07-02 09:57:54 -0400535 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 if (error < 0)
537 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 res++;
539 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 return res;
541}
542
Trond Myklebust83715ad2006-07-05 13:17:12 -0400543static void nfs_cancel_commit_list(struct list_head *head)
544{
545 struct nfs_page *req;
546
547 while(!list_empty(head)) {
548 req = nfs_list_entry(head->next);
549 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400550 nfs_clear_request_commit(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400551 nfs_inode_remove_request(req);
Trond Myklebustb6dff262006-10-19 23:28:38 -0700552 nfs_unlock_request(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -0400553 }
554}
555
Trond Myklebust47c62562009-03-16 08:13:41 -0400556#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400557static int
558nfs_need_commit(struct nfs_inode *nfsi)
559{
560 return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * nfs_scan_commit - Scan an inode for commit requests
565 * @inode: NFS inode to scan
566 * @dst: destination list
567 * @idx_start: lower bound of page->index to scan.
568 * @npages: idx_start + npages sets the upper bound to scan.
569 *
570 * Moves requests from the inode's 'commit' request list.
571 * The requests are *not* checked to ensure that they form a contiguous set.
572 */
573static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400574nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575{
576 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustff778d02010-02-19 16:53:39 -0800577 int ret;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000578
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400579 if (!nfs_need_commit(nfsi))
580 return 0;
581
Trond Myklebustff778d02010-02-19 16:53:39 -0800582 ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
583 if (ret > 0)
584 nfsi->ncommit -= ret;
Trond Myklebust2928db12010-02-19 17:03:18 -0800585 if (nfs_need_commit(NFS_I(inode)))
586 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Trond Myklebustff778d02010-02-19 16:53:39 -0800587 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500589#else
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400590static inline int nfs_need_commit(struct nfs_inode *nfsi)
591{
592 return 0;
593}
594
Trond Myklebustca52fec2007-04-17 17:22:13 -0400595static 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 -0500596{
597 return 0;
598}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599#endif
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400602 * Search for an existing write request, and attempt to update
603 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400605 * If the attempt fails, then the existing request is flushed out
606 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400608static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
609 struct page *page,
610 unsigned int offset,
611 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400613 struct nfs_page *req;
614 unsigned int rqend;
615 unsigned int end;
616 int error;
617
618 if (!PagePrivate(page))
619 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400622 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 for (;;) {
Trond Myklebust277459d2006-12-05 00:35:35 -0500625 req = nfs_page_find_request_locked(page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400626 if (req == NULL)
627 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500628
Trond Myklebuste7d39062008-06-13 12:12:32 -0400629 rqend = req->wb_offset + req->wb_bytes;
630 /*
631 * Tell the caller to flush out the request if
632 * the offsets are non-contiguous.
633 * Note: nfs_flush_incompatible() will already
634 * have flushed out requests having wrong owners.
635 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400636 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400637 || end < req->wb_offset)
638 goto out_flushme;
639
640 if (nfs_set_page_tag_locked(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Trond Myklebuste7d39062008-06-13 12:12:32 -0400643 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400644 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400645 error = nfs_wait_on_request(req);
646 nfs_release_request(req);
647 if (error != 0)
648 goto out_err;
649 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 }
651
Trond Myklebustff778d02010-02-19 16:53:39 -0800652 if (nfs_clear_request_commit(req) &&
653 radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
654 req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL)
655 NFS_I(inode)->ncommit--;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 /* Okay, the request matches. Update the region */
658 if (offset < req->wb_offset) {
659 req->wb_offset = offset;
660 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 if (end > rqend)
663 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400664 else
665 req->wb_bytes = rqend - req->wb_offset;
666out_unlock:
667 spin_unlock(&inode->i_lock);
668 return req;
669out_flushme:
670 spin_unlock(&inode->i_lock);
671 nfs_release_request(req);
672 error = nfs_wb_page(inode, page);
673out_err:
674 return ERR_PTR(error);
675}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Trond Myklebuste7d39062008-06-13 12:12:32 -0400677/*
678 * Try to update an existing write request, or create one if there is none.
679 *
680 * Note: Should always be called with the Page Lock held to prevent races
681 * if we have to add a new request. Also assumes that the caller has
682 * already called nfs_flush_incompatible() if necessary.
683 */
684static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
685 struct page *page, unsigned int offset, unsigned int bytes)
686{
687 struct inode *inode = page->mapping->host;
688 struct nfs_page *req;
689 int error;
690
691 req = nfs_try_to_update_request(inode, page, offset, bytes);
692 if (req != NULL)
693 goto out;
694 req = nfs_create_request(ctx, inode, page, offset, bytes);
695 if (IS_ERR(req))
696 goto out;
697 error = nfs_inode_add_request(inode, req);
698 if (error != 0) {
699 nfs_release_request(req);
700 req = ERR_PTR(error);
701 }
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400702out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400703 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
Trond Myklebuste7d39062008-06-13 12:12:32 -0400706static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
707 unsigned int offset, unsigned int count)
708{
709 struct nfs_page *req;
710
711 req = nfs_setup_write_request(ctx, page, offset, count);
712 if (IS_ERR(req))
713 return PTR_ERR(req);
714 /* Update file length */
715 nfs_grow_file(page, offset, count);
716 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
717 nfs_clear_page_tag_locked(req);
718 return 0;
719}
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721int nfs_flush_incompatible(struct file *file, struct page *page)
722{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400723 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500725 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 /*
727 * Look for a request corresponding to this page. If there
728 * is one, and it belongs to another file, we flush it out
729 * before we try to copy anything into the page. Do this
730 * due to the lack of an ACCESS-type call in NFSv2.
731 * Also do the same if we find a request from an existing
732 * dropped page.
733 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500734 do {
735 req = nfs_page_find_request(page);
736 if (req == NULL)
737 return 0;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400738 do_flush = req->wb_page != page || req->wb_context != ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500740 if (!do_flush)
741 return 0;
742 status = nfs_wb_page(page->mapping->host, page);
743 } while (status == 0);
744 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
747/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500748 * If the page cache is marked as unsafe or invalid, then we can't rely on
749 * the PageUptodate() flag. In this case, we will need to turn off
750 * write optimisations that depend on the page contents being correct.
751 */
752static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
753{
754 return PageUptodate(page) &&
755 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
756}
757
758/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 * Update and possibly write a cached page of an NFS file.
760 *
761 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
762 * things with a page scheduled for an RPC call (e.g. invalidate it).
763 */
764int nfs_updatepage(struct file *file, struct page *page,
765 unsigned int offset, unsigned int count)
766{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400767 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 int status = 0;
770
Chuck Lever91d5b472006-03-20 13:44:14 -0500771 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
772
Chuck Lever48186c72008-06-11 17:56:05 -0400773 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800774 file->f_path.dentry->d_parent->d_name.name,
775 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400776 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500779 * is up to date, it may be more efficient to extend the write
780 * to cover the entire page in order to avoid fragmentation
781 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500783 if (nfs_write_pageuptodate(page, inode) &&
784 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100785 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500786 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789
Trond Myklebuste21195a2006-12-05 00:35:39 -0500790 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400791 if (status < 0)
792 nfs_set_pageerror(page);
793 else
794 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Chuck Lever48186c72008-06-11 17:56:05 -0400796 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 return status;
799}
800
801static void nfs_writepage_release(struct nfs_page *req)
802{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Trond Myklebust7e5f6142008-05-25 12:59:19 -0400804 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req)) {
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400805 nfs_end_page_writeback(req->wb_page);
806 nfs_inode_remove_request(req);
807 } else
808 nfs_end_page_writeback(req->wb_page);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400809 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Trond Myklebust3ff75762007-07-14 15:40:00 -0400812static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
814 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
815 case FLUSH_HIGHPRI:
816 return RPC_PRIORITY_HIGH;
817 case FLUSH_LOWPRI:
818 return RPC_PRIORITY_LOW;
819 }
820 return RPC_PRIORITY_NORMAL;
821}
822
823/*
824 * Set up the argument/result storage required for the RPC call.
825 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400826static int nfs_write_rpcsetup(struct nfs_page *req,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 struct nfs_write_data *data,
Trond Myklebust788e7a82006-03-20 13:44:27 -0500828 const struct rpc_call_ops *call_ops,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 unsigned int count, unsigned int offset,
830 int how)
831{
Trond Myklebust84115e12007-07-14 15:39:59 -0400832 struct inode *inode = req->wb_context->path.dentry->d_inode;
833 int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400834 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400835 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400836 struct rpc_message msg = {
837 .rpc_argp = &data->args,
838 .rpc_resp = &data->res,
839 .rpc_cred = req->wb_context->cred,
840 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400841 struct rpc_task_setup task_setup_data = {
842 .rpc_client = NFS_CLIENT(inode),
Trond Myklebust07737692007-10-25 18:42:54 -0400843 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400844 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400845 .callback_ops = call_ops,
846 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500847 .workqueue = nfsiod_workqueue,
Trond Myklebust84115e12007-07-14 15:39:59 -0400848 .flags = flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400849 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400850 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851
852 /* Set up the RPC argument and reply structs
853 * NB: take care not to mess about with data->commit et al. */
854
855 data->req = req;
Trond Myklebust88be9f92007-06-05 10:42:27 -0400856 data->inode = inode = req->wb_context->path.dentry->d_inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400857 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 data->args.fh = NFS_FH(inode);
860 data->args.offset = req_offset(req) + offset;
861 data->args.pgbase = req->wb_pgbase + offset;
862 data->args.pages = data->pagevec;
863 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500864 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400865 data->args.stable = NFS_UNSTABLE;
866 if (how & FLUSH_STABLE) {
867 data->args.stable = NFS_DATA_SYNC;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400868 if (!nfs_need_commit(NFS_I(inode)))
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400869 data->args.stable = NFS_FILE_SYNC;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872 data->res.fattr = &data->fattr;
873 data->res.count = count;
874 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400875 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Trond Myklebust788e7a82006-03-20 13:44:27 -0500877 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400878 NFS_PROTO(inode)->write_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Chuck Levera3f565b2007-01-31 12:14:01 -0500880 dprintk("NFS: %5u initiated write call "
Chuck Lever48186c72008-06-11 17:56:05 -0400881 "(req %s/%lld, %u bytes @ offset %llu)\n",
Chuck Lever0bbacc42005-11-01 16:53:32 -0500882 data->task.tk_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 inode->i_sb->s_id,
884 (long long)NFS_FILEID(inode),
885 count,
886 (unsigned long long)data->args.offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Trond Myklebust07737692007-10-25 18:42:54 -0400888 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400889 if (IS_ERR(task))
890 return PTR_ERR(task);
891 rpc_put_task(task);
892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893}
894
Fred6d884e82008-03-19 11:24:38 -0400895/* If a nfs_flush_* function fails, it should remove reqs from @head and
896 * call this on each, which will prepare them to be retried on next
897 * writeback using standard nfs.
898 */
899static void nfs_redirty_request(struct nfs_page *req)
900{
901 nfs_mark_request_dirty(req);
902 nfs_end_page_writeback(req->wb_page);
903 nfs_clear_page_tag_locked(req);
904}
905
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906/*
907 * Generate multiple small requests to write out a single
908 * contiguous dirty area on one page.
909 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400910static 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 -0700911{
912 struct nfs_page *req = nfs_list_entry(head->next);
913 struct page *page = req->wb_page;
914 struct nfs_write_data *data;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700915 size_t wsize = NFS_SERVER(inode)->wsize, nbytes;
916 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400918 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 LIST_HEAD(list);
920
921 nfs_list_remove_request(req);
922
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400923 nbytes = count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700924 do {
925 size_t len = min(nbytes, wsize);
926
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400927 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 if (!data)
929 goto out_bad;
930 list_add(&data->pages, &list);
931 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700932 nbytes -= len;
933 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 atomic_set(&req->wb_complete, requests);
935
936 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 offset = 0;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400938 nbytes = count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 do {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400940 int ret2;
941
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 data = list_entry(list.next, struct nfs_write_data, pages);
943 list_del_init(&data->pages);
944
945 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400947 if (nbytes < wsize)
948 wsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400949 ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400950 wsize, offset, how);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400951 if (ret == 0)
952 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400953 offset += wsize;
954 nbytes -= wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 } while (nbytes != 0);
956
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400957 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
959out_bad:
960 while (!list_empty(&list)) {
961 data = list_entry(list.next, struct nfs_write_data, pages);
962 list_del(&data->pages);
Trond Myklebust8aca67f2006-11-13 16:23:44 -0500963 nfs_writedata_release(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500965 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 return -ENOMEM;
967}
968
969/*
970 * Create an RPC task for the given write request and kick it.
971 * The page must have been locked by the caller.
972 *
973 * It may happen that the page we're passed is not marked dirty.
974 * This is the case if nfs_updatepage detects a conflicting request
975 * that has been written but not committed.
976 */
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400977static 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 -0700978{
979 struct nfs_page *req;
980 struct page **pages;
981 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400983 data = nfs_writedata_alloc(npages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 if (!data)
985 goto out_bad;
986
987 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 while (!list_empty(head)) {
989 req = nfs_list_entry(head->next);
990 nfs_list_remove_request(req);
991 nfs_list_add_request(req, &data->pages);
992 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 }
995 req = nfs_list_entry(data->pages.next);
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 /* Set up the argument struct */
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400998 return nfs_write_rpcsetup(req, data, &nfs_write_full_ops, count, 0, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 out_bad:
1000 while (!list_empty(head)) {
Trond Myklebust10afec92007-05-14 17:16:04 -04001001 req = nfs_list_entry(head->next);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 nfs_list_remove_request(req);
Trond Myklebust61822ab2006-12-05 00:35:42 -05001003 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 return -ENOMEM;
1006}
1007
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001008static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1009 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010{
Chuck Leverbf4285e2007-12-20 14:54:57 -05001011 size_t wsize = NFS_SERVER(inode)->wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001013 if (wsize < PAGE_CACHE_SIZE)
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001014 nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001015 else
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001016 nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017}
1018
1019/*
1020 * Handle a write reply that flushed part of a page.
1021 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001022static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001024 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
Chuck Lever48186c72008-06-11 17:56:05 -04001026 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1027 task->tk_pid,
1028 data->req->wb_context->path.dentry->d_inode->i_sb->s_id,
1029 (long long)
1030 NFS_FILEID(data->req->wb_context->path.dentry->d_inode),
1031 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001033 nfs_writeback_done(task, data);
1034}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001035
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001036static void nfs_writeback_release_partial(void *calldata)
1037{
1038 struct nfs_write_data *data = calldata;
1039 struct nfs_page *req = data->req;
1040 struct page *page = req->wb_page;
1041 int status = data->task.tk_status;
1042
1043 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001044 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001045 nfs_context_set_write_error(req->wb_context, status);
1046 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001047 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
1049
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001050 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001051 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001052
Trond Myklebust587142f2007-07-02 09:57:54 -04001053 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001054 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1055 /* Do nothing we need to resend the writes */
1056 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1057 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1058 dprintk(" defer commit\n");
1059 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1060 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1061 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1062 dprintk(" server reboot detected\n");
1063 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001064 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001065 } else
1066 dprintk(" OK\n");
1067
1068out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 if (atomic_dec_and_test(&req->wb_complete))
1070 nfs_writepage_release(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001071 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072}
1073
Andy Adamsondef6ed72009-04-01 09:22:26 -04001074#if defined(CONFIG_NFS_V4_1)
1075void nfs_write_prepare(struct rpc_task *task, void *calldata)
1076{
1077 struct nfs_write_data *data = calldata;
1078 struct nfs_client *clp = (NFS_SERVER(data->inode))->nfs_client;
1079
1080 if (nfs4_setup_sequence(clp, &data->args.seq_args,
1081 &data->res.seq_res, 1, task))
1082 return;
1083 rpc_call_start(task);
1084}
1085#endif /* CONFIG_NFS_V4_1 */
1086
Trond Myklebust788e7a82006-03-20 13:44:27 -05001087static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001088#if defined(CONFIG_NFS_V4_1)
1089 .rpc_call_prepare = nfs_write_prepare,
1090#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001091 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001092 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001093};
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095/*
1096 * Handle a write reply that flushes a whole page.
1097 *
1098 * FIXME: There is an inherent race with invalidate_inode_pages and
1099 * writebacks since the page->count is kept > 1 for as long
1100 * as the page has a write request pending.
1101 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001102static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001104 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001106 nfs_writeback_done(task, data);
1107}
1108
1109static void nfs_writeback_release_full(void *calldata)
1110{
1111 struct nfs_write_data *data = calldata;
1112 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 /* Update attributes as result of writeback. */
1115 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001116 struct nfs_page *req = nfs_list_entry(data->pages.next);
1117 struct page *page = req->wb_page;
1118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Chuck Lever48186c72008-06-11 17:56:05 -04001121 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1122 data->task.tk_pid,
Trond Myklebust88be9f92007-06-05 10:42:27 -04001123 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1124 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 req->wb_bytes,
1126 (long long)req_offset(req));
1127
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001128 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001129 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001130 nfs_context_set_write_error(req->wb_context, status);
1131 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001132 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001135 if (nfs_write_need_commit(data)) {
1136 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1137 nfs_mark_request_commit(req);
1138 nfs_end_page_writeback(page);
1139 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 goto next;
1141 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001142 dprintk(" OK\n");
1143remove_request:
1144 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001147 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001149 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150}
1151
Trond Myklebust788e7a82006-03-20 13:44:27 -05001152static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001153#if defined(CONFIG_NFS_V4_1)
1154 .rpc_call_prepare = nfs_write_prepare,
1155#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001156 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001157 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001158};
1159
1160
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161/*
1162 * This function is called when the WRITE call is complete.
1163 */
Chuck Lever462d5b32006-03-20 13:44:32 -05001164int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 struct nfs_writeargs *argp = &data->args;
1167 struct nfs_writeres *resp = &data->res;
Andy Adamsoneedc0202009-04-01 09:22:41 -04001168 struct nfs_server *server = NFS_SERVER(data->inode);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001169 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170
Chuck Levera3f565b2007-01-31 12:14:01 -05001171 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 task->tk_pid, task->tk_status);
1173
Chuck Leverf551e442006-09-20 14:33:04 -04001174 /*
1175 * ->write_done will attempt to use post-op attributes to detect
1176 * conflicting writes by other clients. A strict interpretation
1177 * of close-to-open would allow us to continue caching even if
1178 * another writer had changed the file, but some applications
1179 * depend on tighter cache coherency when writing.
1180 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001181 status = NFS_PROTO(data->inode)->write_done(task, data);
1182 if (status != 0)
1183 return status;
Chuck Lever91d5b472006-03-20 13:44:14 -05001184 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1187 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1188 /* We tried a write call, but the server did not
1189 * commit data to stable storage even though we
1190 * requested it.
1191 * Note: There is a known bug in Tru64 < 5.0 in which
1192 * the server reports NFS_DATA_SYNC, but performs
1193 * NFS_FILE_SYNC. We therefore implement this checking
1194 * as a dprintk() in order to avoid filling syslog.
1195 */
1196 static unsigned long complain;
1197
1198 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001199 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 " (committed = %d) != (stable = %d)\n",
Andy Adamsoneedc0202009-04-01 09:22:41 -04001201 server->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 resp->verf->committed, argp->stable);
1203 complain = jiffies + 300 * HZ;
1204 }
1205 }
1206#endif
1207 /* Is this a short write? */
1208 if (task->tk_status >= 0 && resp->count < argp->count) {
1209 static unsigned long complain;
1210
Chuck Lever91d5b472006-03-20 13:44:14 -05001211 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 /* Has the server at least made some progress? */
1214 if (resp->count != 0) {
1215 /* Was this an NFSv2 write or an NFSv3 stable write? */
1216 if (resp->verf->committed != NFS_UNSTABLE) {
1217 /* Resend from where the server left off */
1218 argp->offset += resp->count;
1219 argp->pgbase += resp->count;
1220 argp->count -= resp->count;
1221 } else {
1222 /* Resend as a stable write in order to avoid
1223 * headaches in the case of a server crash.
1224 */
1225 argp->stable = NFS_FILE_SYNC;
1226 }
Trond Myklebust0110ee12009-12-07 09:00:24 -05001227 nfs_restart_rpc(task, server->nfs_client);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001228 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
1230 if (time_before(complain, jiffies)) {
1231 printk(KERN_WARNING
1232 "NFS: Server wrote zero bytes, expected %u.\n",
1233 argp->count);
1234 complain = jiffies + 300 * HZ;
1235 }
1236 /* Can't do anything about it except throw an error. */
1237 task->tk_status = -EIO;
1238 }
Trond Myklebust788e7a82006-03-20 13:44:27 -05001239 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240}
1241
1242
1243#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
H Hartley Sweeten0aa05882010-01-26 15:42:03 -05001244static void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
Trond Myklebust383ba712008-02-19 20:04:20 -05001246 struct nfs_write_data *wdata = data;
1247
1248 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 nfs_commit_free(wdata);
1250}
1251
1252/*
1253 * Set up the argument/result storage required for the RPC call.
1254 */
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001255static int nfs_commit_rpcsetup(struct list_head *head,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001256 struct nfs_write_data *data,
1257 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258{
Trond Myklebust84115e12007-07-14 15:39:59 -04001259 struct nfs_page *first = nfs_list_entry(head->next);
1260 struct inode *inode = first->wb_context->path.dentry->d_inode;
1261 int flags = (how & FLUSH_SYNC) ? 0 : RPC_TASK_ASYNC;
Trond Myklebust3ff75762007-07-14 15:40:00 -04001262 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -04001263 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001264 struct rpc_message msg = {
1265 .rpc_argp = &data->args,
1266 .rpc_resp = &data->res,
1267 .rpc_cred = first->wb_context->cred,
1268 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001269 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001270 .task = &data->task,
Trond Myklebust84115e12007-07-14 15:39:59 -04001271 .rpc_client = NFS_CLIENT(inode),
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001272 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -04001273 .callback_ops = &nfs_commit_ops,
1274 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001275 .workqueue = nfsiod_workqueue,
Trond Myklebust84115e12007-07-14 15:39:59 -04001276 .flags = flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001277 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001278 };
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 /* Set up the RPC argument and reply structs
1281 * NB: take care not to mess about with data->commit et al. */
1282
1283 list_splice_init(head, &data->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 data->inode = inode;
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001286 data->cred = msg.rpc_cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 data->args.fh = NFS_FH(data->inode);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001289 /* Note: we always request a commit of the entire inode */
1290 data->args.offset = 0;
1291 data->args.count = 0;
Trond Myklebust383ba712008-02-19 20:04:20 -05001292 data->args.context = get_nfs_open_context(first->wb_context);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001293 data->res.count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 data->res.fattr = &data->fattr;
1295 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -04001296 nfs_fattr_init(&data->fattr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001297
1298 /* Set up the initial task struct. */
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001299 NFS_PROTO(inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Chuck Levera3f565b2007-01-31 12:14:01 -05001301 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001302
Trond Myklebust07737692007-10-25 18:42:54 -04001303 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001304 if (IS_ERR(task))
1305 return PTR_ERR(task);
1306 rpc_put_task(task);
1307 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308}
1309
1310/*
1311 * Commit dirty pages
1312 */
1313static int
Chuck Lever40859d72005-11-30 18:09:02 -05001314nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
1316 struct nfs_write_data *data;
1317 struct nfs_page *req;
1318
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001319 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
1321 if (!data)
1322 goto out_bad;
1323
1324 /* Set up the argument struct */
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001325 return nfs_commit_rpcsetup(head, data, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 out_bad:
1327 while (!list_empty(head)) {
1328 req = nfs_list_entry(head->next);
1329 nfs_list_remove_request(req);
1330 nfs_mark_request_commit(req);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001331 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -07001332 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1333 BDI_RECLAIMABLE);
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001334 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 }
1336 return -ENOMEM;
1337}
1338
1339/*
1340 * COMMIT call returned
1341 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001342static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001344 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Chuck Levera3f565b2007-01-31 12:14:01 -05001346 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 task->tk_pid, task->tk_status);
1348
Trond Myklebust788e7a82006-03-20 13:44:27 -05001349 /* Call the NFS version-specific code */
1350 if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1351 return;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001352}
1353
1354static void nfs_commit_release(void *calldata)
1355{
1356 struct nfs_write_data *data = calldata;
1357 struct nfs_page *req;
1358 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 while (!list_empty(&data->pages)) {
1361 req = nfs_list_entry(data->pages.next);
1362 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -04001363 nfs_clear_request_commit(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
Chuck Lever48186c72008-06-11 17:56:05 -04001365 dprintk("NFS: commit (%s/%lld %d@%lld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001366 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1367 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 req->wb_bytes,
1369 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001370 if (status < 0) {
1371 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001373 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 goto next;
1375 }
1376
1377 /* Okay, COMMIT succeeded, apparently. Check the verifier
1378 * returned by the server against all stored verfs. */
1379 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1380 /* We have a match */
1381 nfs_inode_remove_request(req);
1382 dprintk(" OK\n");
1383 goto next;
1384 }
1385 /* We have a mismatch. Write the page again */
1386 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001387 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001389 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001391 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001393
1394static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001395#if defined(CONFIG_NFS_V4_1)
1396 .rpc_call_prepare = nfs_write_prepare,
1397#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001398 .rpc_call_done = nfs_commit_done,
1399 .rpc_release = nfs_commit_release,
1400};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001402static int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 LIST_HEAD(head);
Trond Myklebust7d46a492006-03-20 13:44:50 -05001405 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406
Trond Myklebust587142f2007-07-02 09:57:54 -04001407 spin_lock(&inode->i_lock);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001408 res = nfs_scan_commit(inode, &head, 0, 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001409 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 if (res) {
Trond Myklebust7d46a492006-03-20 13:44:50 -05001411 int error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001412 if (error < 0)
1413 return error;
1414 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 return res;
1416}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001417
1418static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1419{
Trond Myklebust420e3642010-02-19 17:00:02 -08001420 struct nfs_inode *nfsi = NFS_I(inode);
1421 int flags = FLUSH_SYNC;
1422 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001423
Trond Myklebust420e3642010-02-19 17:00:02 -08001424 /* Don't commit yet if this is a non-blocking flush and there are
1425 * lots of outstanding writes for this mapping.
1426 */
1427 if (wbc->sync_mode == WB_SYNC_NONE &&
1428 nfsi->ncommit <= (nfsi->npages >> 1))
1429 goto out_mark_dirty;
1430
Trond Myklebust5bad5ab2010-02-19 17:02:24 -08001431 if (wbc->nonblocking || wbc->for_background)
Trond Myklebust420e3642010-02-19 17:00:02 -08001432 flags = 0;
1433 ret = nfs_commit_inode(inode, flags);
1434 if (ret >= 0) {
1435 if (wbc->sync_mode == WB_SYNC_NONE) {
1436 if (ret < wbc->nr_to_write)
1437 wbc->nr_to_write -= ret;
1438 else
1439 wbc->nr_to_write = 0;
1440 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001441 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001442 }
1443out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001444 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1445 return ret;
1446}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001447#else
1448static inline int nfs_commit_list(struct inode *inode, struct list_head *head, int how)
1449{
1450 return 0;
1451}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001452
1453static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1454{
1455 return 0;
1456}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457#endif
1458
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001459int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1460{
1461 return nfs_commit_unstable_pages(inode, wbc);
1462}
1463
Trond Myklebust1c759502006-10-09 16:18:38 -04001464long nfs_sync_mapping_wait(struct address_space *mapping, struct writeback_control *wbc, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
Trond Myklebust1c759502006-10-09 16:18:38 -04001466 struct inode *inode = mapping->host;
Trond Myklebustca52fec2007-04-17 17:22:13 -04001467 pgoff_t idx_start, idx_end;
Trond Myklebust1c759502006-10-09 16:18:38 -04001468 unsigned int npages = 0;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001469 LIST_HEAD(head);
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001470 int nocommit = how & FLUSH_NOCOMMIT;
Trond Myklebust3f442542006-09-17 14:46:44 -04001471 long pages, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Trond Myklebust1c759502006-10-09 16:18:38 -04001473 /* FIXME */
1474 if (wbc->range_cyclic)
1475 idx_start = 0;
1476 else {
1477 idx_start = wbc->range_start >> PAGE_CACHE_SHIFT;
1478 idx_end = wbc->range_end >> PAGE_CACHE_SHIFT;
1479 if (idx_end > idx_start) {
Trond Myklebustca52fec2007-04-17 17:22:13 -04001480 pgoff_t l_npages = 1 + idx_end - idx_start;
Trond Myklebust1c759502006-10-09 16:18:38 -04001481 npages = l_npages;
1482 if (sizeof(npages) != sizeof(l_npages) &&
Trond Myklebustca52fec2007-04-17 17:22:13 -04001483 (pgoff_t)npages != l_npages)
Trond Myklebust1c759502006-10-09 16:18:38 -04001484 npages = 0;
1485 }
1486 }
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001487 how &= ~FLUSH_NOCOMMIT;
Trond Myklebust587142f2007-07-02 09:57:54 -04001488 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 do {
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001490 ret = nfs_wait_on_requests_locked(inode, idx_start, npages);
1491 if (ret != 0)
Trond Myklebust70b9ecb2006-01-03 09:55:34 +01001492 continue;
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001493 if (nocommit)
1494 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001495 pages = nfs_scan_commit(inode, &head, idx_start, npages);
Trond Myklebust724c4392007-04-17 17:22:13 -04001496 if (pages == 0)
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001497 break;
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001498 if (how & FLUSH_INVALIDATE) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001499 spin_unlock(&inode->i_lock);
Trond Myklebust83715ad2006-07-05 13:17:12 -04001500 nfs_cancel_commit_list(&head);
Trond Myklebuste8e058e2006-11-15 17:31:56 -05001501 ret = pages;
Trond Myklebust587142f2007-07-02 09:57:54 -04001502 spin_lock(&inode->i_lock);
Trond Myklebustd2ccddf2006-05-31 01:13:38 -04001503 continue;
1504 }
1505 pages += nfs_scan_commit(inode, &head, 0, 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001506 spin_unlock(&inode->i_lock);
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001507 ret = nfs_commit_list(inode, &head, how);
Trond Myklebust587142f2007-07-02 09:57:54 -04001508 spin_lock(&inode->i_lock);
1509
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001510 } while (ret >= 0);
Trond Myklebust587142f2007-07-02 09:57:54 -04001511 spin_unlock(&inode->i_lock);
Trond Myklebustc42de9d2006-03-20 13:44:51 -05001512 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513}
1514
Trond Myklebust34901f72007-07-25 14:09:54 -04001515static int __nfs_write_mapping(struct address_space *mapping, struct writeback_control *wbc, int how)
Trond Myklebust1c759502006-10-09 16:18:38 -04001516{
Trond Myklebust1c759502006-10-09 16:18:38 -04001517 int ret;
1518
Trond Myklebust34901f72007-07-25 14:09:54 -04001519 ret = nfs_writepages(mapping, wbc);
Trond Myklebust61822ab2006-12-05 00:35:42 -05001520 if (ret < 0)
1521 goto out;
Trond Myklebust34901f72007-07-25 14:09:54 -04001522 ret = nfs_sync_mapping_wait(mapping, wbc, how);
Trond Myklebusted90ef52007-07-20 13:13:28 -04001523 if (ret < 0)
1524 goto out;
1525 return 0;
Trond Myklebust61822ab2006-12-05 00:35:42 -05001526out:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001527 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Trond Myklebust1c759502006-10-09 16:18:38 -04001528 return ret;
1529}
1530
Trond Myklebust34901f72007-07-25 14:09:54 -04001531/* Two pass sync: first using WB_SYNC_NONE, then WB_SYNC_ALL */
1532static int nfs_write_mapping(struct address_space *mapping, int how)
1533{
1534 struct writeback_control wbc = {
1535 .bdi = mapping->backing_dev_info,
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001536 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001537 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001538 .range_start = 0,
1539 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001540 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001541
Trond Myklebust34901f72007-07-25 14:09:54 -04001542 return __nfs_write_mapping(mapping, &wbc, how);
1543}
1544
Trond Myklebusted90ef52007-07-20 13:13:28 -04001545/*
1546 * flush the inode to disk.
1547 */
1548int nfs_wb_all(struct inode *inode)
Trond Myklebust1c759502006-10-09 16:18:38 -04001549{
Trond Myklebusted90ef52007-07-20 13:13:28 -04001550 return nfs_write_mapping(inode->i_mapping, 0);
1551}
Trond Myklebust1c759502006-10-09 16:18:38 -04001552
Trond Myklebusted90ef52007-07-20 13:13:28 -04001553int nfs_wb_nocommit(struct inode *inode)
1554{
1555 return nfs_write_mapping(inode->i_mapping, FLUSH_NOCOMMIT);
Trond Myklebust1c759502006-10-09 16:18:38 -04001556}
1557
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001558int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1559{
1560 struct nfs_page *req;
1561 loff_t range_start = page_offset(page);
1562 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1563 struct writeback_control wbc = {
1564 .bdi = page->mapping->backing_dev_info,
1565 .sync_mode = WB_SYNC_ALL,
1566 .nr_to_write = LONG_MAX,
1567 .range_start = range_start,
1568 .range_end = range_end,
1569 };
1570 int ret = 0;
1571
1572 BUG_ON(!PageLocked(page));
1573 for (;;) {
1574 req = nfs_page_find_request(page);
1575 if (req == NULL)
1576 goto out;
Trond Myklebuste468bae2008-06-13 13:25:22 -04001577 if (test_bit(PG_CLEAN, &req->wb_flags)) {
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001578 nfs_release_request(req);
1579 break;
1580 }
1581 if (nfs_lock_request_dontget(req)) {
1582 nfs_inode_remove_request(req);
1583 /*
1584 * In case nfs_inode_remove_request has marked the
1585 * page as being dirty
1586 */
1587 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1588 nfs_unlock_request(req);
1589 break;
1590 }
1591 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001592 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001593 if (ret < 0)
1594 goto out;
1595 }
1596 if (!PagePrivate(page))
1597 return 0;
1598 ret = nfs_sync_mapping_wait(page->mapping, &wbc, FLUSH_INVALIDATE);
1599out:
1600 return ret;
1601}
1602
Adrian Bunk5334eb12007-11-21 15:04:31 -08001603static int nfs_wb_page_priority(struct inode *inode, struct page *page,
1604 int how)
Trond Myklebust1c759502006-10-09 16:18:38 -04001605{
1606 loff_t range_start = page_offset(page);
1607 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001608 struct writeback_control wbc = {
1609 .bdi = page->mapping->backing_dev_info,
1610 .sync_mode = WB_SYNC_ALL,
1611 .nr_to_write = LONG_MAX,
1612 .range_start = range_start,
1613 .range_end = range_end,
1614 };
1615 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001616
Trond Myklebust73e33022008-04-11 16:03:54 -04001617 do {
1618 if (clear_page_dirty_for_io(page)) {
1619 ret = nfs_writepage_locked(page, &wbc);
1620 if (ret < 0)
1621 goto out_error;
1622 } else if (!PagePrivate(page))
1623 break;
1624 ret = nfs_sync_mapping_wait(page->mapping, &wbc, how);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001625 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001626 goto out_error;
1627 } while (PagePrivate(page));
1628 return 0;
1629out_error:
Trond Myklebuste507d9e2006-12-05 00:35:42 -05001630 __mark_inode_dirty(inode, I_DIRTY_PAGES);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001631 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001632}
1633
1634/*
1635 * Write back all requests on one page - we do this before reading it.
1636 */
1637int nfs_wb_page(struct inode *inode, struct page* page)
1638{
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001639 return nfs_wb_page_priority(inode, page, FLUSH_STABLE);
Trond Myklebust1c759502006-10-09 16:18:38 -04001640}
1641
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001642#ifdef CONFIG_MIGRATION
1643int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1644 struct page *page)
1645{
1646 struct nfs_page *req;
1647 int ret;
1648
Trond Myklebust7549ad52010-02-08 09:32:34 -05001649 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001650
1651 req = nfs_find_and_lock_request(page);
1652 ret = PTR_ERR(req);
1653 if (IS_ERR(req))
1654 goto out;
1655
1656 ret = migrate_page(mapping, newpage, page);
1657 if (!req)
1658 goto out;
1659 if (ret)
1660 goto out_unlock;
1661 page_cache_get(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001662 spin_lock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001663 req->wb_page = newpage;
1664 SetPagePrivate(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001665 set_page_private(newpage, (unsigned long)req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001666 ClearPagePrivate(page);
1667 set_page_private(page, 0);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001668 spin_unlock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001669 page_cache_release(page);
1670out_unlock:
1671 nfs_clear_page_tag_locked(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001672out:
1673 return ret;
1674}
1675#endif
1676
David Howellsf7b422b2006-06-09 09:34:33 -04001677int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678{
1679 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1680 sizeof(struct nfs_write_data),
1681 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001682 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 if (nfs_wdata_cachep == NULL)
1684 return -ENOMEM;
1685
Matthew Dobson93d23412006-03-26 01:37:50 -08001686 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1687 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 if (nfs_wdata_mempool == NULL)
1689 return -ENOMEM;
1690
Matthew Dobson93d23412006-03-26 01:37:50 -08001691 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1692 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 if (nfs_commit_mempool == NULL)
1694 return -ENOMEM;
1695
Peter Zijlstra89a09142007-03-16 13:38:26 -08001696 /*
1697 * NFS congestion size, scale with available memory.
1698 *
1699 * 64MB: 8192k
1700 * 128MB: 11585k
1701 * 256MB: 16384k
1702 * 512MB: 23170k
1703 * 1GB: 32768k
1704 * 2GB: 46340k
1705 * 4GB: 65536k
1706 * 8GB: 92681k
1707 * 16GB: 131072k
1708 *
1709 * This allows larger machines to have larger/more transfers.
1710 * Limit the default to 256M
1711 */
1712 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1713 if (nfs_congestion_kb > 256*1024)
1714 nfs_congestion_kb = 256*1024;
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return 0;
1717}
1718
David Brownell266bee82006-06-27 12:59:15 -07001719void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720{
1721 mempool_destroy(nfs_commit_mempool);
1722 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001723 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725