blob: 834f0fe96f89f4acf707df504e1a244fc56d466b [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>
Paul Gortmakerafeacc82011-05-26 16:00:52 -040023#include <linux/export.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050028#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050029#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040030#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040031#include "fscache.h"
Fred Isaman94ad1c82011-03-01 01:34:14 +000032#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34#define NFSDBG_FACILITY NFSDBG_PAGECACHE
35
36#define MIN_POOL_WRITE (32)
37#define MIN_POOL_COMMIT (4)
38
39/*
40 * Local function declarations
41 */
Trond Myklebustc63c7b02007-04-02 19:29:52 -040042static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
43 struct inode *inode, int ioflags);
Fred Isamanf8512ad2008-03-19 11:24:39 -040044static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050045static const struct rpc_call_ops nfs_write_partial_ops;
46static const struct rpc_call_ops nfs_write_full_ops;
47static const struct rpc_call_ops nfs_commit_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Christoph Lametere18b8902006-12-06 20:33:20 -080049static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050050static mempool_t *nfs_wdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070051static mempool_t *nfs_commit_mempool;
52
Trond Myklebustc9d8f892008-04-15 16:56:39 -040053struct nfs_write_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070054{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080055 struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
Chuck Lever40859d72005-11-30 18:09:02 -050056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 if (p) {
58 memset(p, 0, sizeof(*p));
59 INIT_LIST_HEAD(&p->pages);
60 }
61 return p;
62}
Fred Isamane0c2b382011-03-23 13:27:53 +000063EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Trond Myklebust5e4424a2008-02-25 21:53:49 -080065void nfs_commit_free(struct nfs_write_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
Chuck Lever40859d72005-11-30 18:09:02 -050067 if (p && (p->pagevec != &p->page_array[0]))
68 kfree(p->pagevec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 mempool_free(p, nfs_commit_mempool);
70}
Fred Isamane0c2b382011-03-23 13:27:53 +000071EXPORT_SYMBOL_GPL(nfs_commit_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Trond Myklebust8d5658c2007-04-10 09:26:35 -040073struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050074{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080075 struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050076
77 if (p) {
78 memset(p, 0, sizeof(*p));
79 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070080 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040081 if (pagecount <= ARRAY_SIZE(p->page_array))
82 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050083 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040084 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
85 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050086 mempool_free(p, nfs_wdata_mempool);
87 p = NULL;
88 }
89 }
90 }
91 return p;
92}
93
Trond Myklebust1ae88b22009-08-12 09:12:30 -040094void nfs_writedata_free(struct nfs_write_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050095{
96 if (p && (p->pagevec != &p->page_array[0]))
97 kfree(p->pagevec);
98 mempool_free(p, nfs_wdata_mempool);
99}
100
Trond Myklebustdce81292011-07-13 15:59:19 -0400101void nfs_writedata_release(struct nfs_write_data *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102{
Fred Isaman5053aa52011-03-03 15:13:43 +0000103 put_lseg(wdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -0500104 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 nfs_writedata_free(wdata);
106}
107
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400108static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
109{
110 ctx->error = error;
111 smp_wmb();
112 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
113}
114
Trond Myklebust277459d2006-12-05 00:35:35 -0500115static struct nfs_page *nfs_page_find_request_locked(struct page *page)
116{
117 struct nfs_page *req = NULL;
118
119 if (PagePrivate(page)) {
120 req = (struct nfs_page *)page_private(page);
121 if (req != NULL)
Trond Myklebustc03b4022007-06-17 13:26:38 -0400122 kref_get(&req->wb_kref);
Trond Myklebust277459d2006-12-05 00:35:35 -0500123 }
124 return req;
125}
126
127static struct nfs_page *nfs_page_find_request(struct page *page)
128{
Trond Myklebust587142f2007-07-02 09:57:54 -0400129 struct inode *inode = page->mapping->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500130 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500131
Trond Myklebust587142f2007-07-02 09:57:54 -0400132 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500133 req = nfs_page_find_request_locked(page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400134 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500135 return req;
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138/* Adjust the file length if we're writing beyond the end */
139static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
140{
141 struct inode *inode = page->mapping->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400142 loff_t end, i_size;
143 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Trond Myklebusta3d01452008-06-11 12:21:19 -0400145 spin_lock(&inode->i_lock);
146 i_size = i_size_read(inode);
147 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 if (i_size > 0 && page->index < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400149 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
151 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400152 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400154 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
155out:
156 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157}
158
Trond Myklebusta301b772007-02-06 11:07:15 -0800159/* A writeback failed: mark the page as bad, and invalidate the page cache */
160static void nfs_set_pageerror(struct page *page)
161{
162 SetPageError(page);
163 nfs_zap_mapping(page->mapping->host, page->mapping);
164}
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166/* We can set the PG_uptodate flag if we see that a write request
167 * covers the full page.
168 */
169static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
170{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 if (PageUptodate(page))
172 return;
173 if (base != 0)
174 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500175 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500177 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178}
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180static int wb_priority(struct writeback_control *wbc)
181{
182 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400183 return FLUSH_HIGHPRI | FLUSH_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100184 if (wbc->for_kupdate || wbc->for_background)
Trond Myklebustb31268a2011-03-21 17:02:00 -0400185 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
186 return FLUSH_COND_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187}
188
189/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800190 * NFS congestion control
191 */
192
193int nfs_congestion_kb;
194
195#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
196#define NFS_CONGESTION_OFF_THRESH \
197 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
198
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400199static int nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800200{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400201 int ret = test_set_page_writeback(page);
202
203 if (!ret) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800204 struct inode *inode = page->mapping->host;
205 struct nfs_server *nfss = NFS_SERVER(inode);
206
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400207 page_cache_get(page);
Peter Zijlstra277866a2007-05-08 00:35:12 -0700208 if (atomic_long_inc_return(&nfss->writeback) >
Jens Axboe8aa7e842009-07-09 14:52:32 +0200209 NFS_CONGESTION_ON_THRESH) {
210 set_bdi_congested(&nfss->backing_dev_info,
211 BLK_RW_ASYNC);
212 }
Peter Zijlstra89a09142007-03-16 13:38:26 -0800213 }
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400214 return ret;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800215}
216
217static void nfs_end_page_writeback(struct page *page)
218{
219 struct inode *inode = page->mapping->host;
220 struct nfs_server *nfss = NFS_SERVER(inode);
221
222 end_page_writeback(page);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400223 page_cache_release(page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700224 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200225 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800226}
227
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400228static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
Trond Myklebuste261f512006-12-05 00:35:41 -0500229{
Trond Myklebust587142f2007-07-02 09:57:54 -0400230 struct inode *inode = page->mapping->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500231 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500232 int ret;
233
Trond Myklebust587142f2007-07-02 09:57:54 -0400234 spin_lock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400235 for (;;) {
Trond Myklebuste261f512006-12-05 00:35:41 -0500236 req = nfs_page_find_request_locked(page);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400237 if (req == NULL)
238 break;
Trond Myklebustacee4782008-01-22 17:13:07 -0500239 if (nfs_set_page_tag_locked(req))
Trond Myklebuste261f512006-12-05 00:35:41 -0500240 break;
241 /* Note: If we hold the page lock, as is the case in nfs_writepage,
Trond Myklebustacee4782008-01-22 17:13:07 -0500242 * then the call to nfs_set_page_tag_locked() will always
Trond Myklebuste261f512006-12-05 00:35:41 -0500243 * succeed provided that someone hasn't already marked the
244 * request as dirty (in which case we don't care).
245 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400246 spin_unlock(&inode->i_lock);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400247 if (!nonblock)
248 ret = nfs_wait_on_request(req);
249 else
250 ret = -EAGAIN;
Trond Myklebuste261f512006-12-05 00:35:41 -0500251 nfs_release_request(req);
252 if (ret != 0)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400253 return ERR_PTR(ret);
Trond Myklebust587142f2007-07-02 09:57:54 -0400254 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500255 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400256 spin_unlock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400257 return req;
258}
259
260/*
261 * Find an associated nfs write request, and prepare to flush it out
262 * May return an error if the user signalled nfs_wait_on_request().
263 */
264static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400265 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400266{
267 struct nfs_page *req;
268 int ret = 0;
269
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400270 req = nfs_find_and_lock_request(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400271 if (!req)
272 goto out;
273 ret = PTR_ERR(req);
274 if (IS_ERR(req))
275 goto out;
276
277 ret = nfs_set_page_writeback(page);
278 BUG_ON(ret != 0);
279 BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));
280
Fred Isamanf8512ad2008-03-19 11:24:39 -0400281 if (!nfs_pageio_add_request(pgio, req)) {
282 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400283 ret = pgio->pg_error;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400284 }
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400285out:
286 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500287}
288
Trond Myklebustf758c8852007-07-22 19:27:32 -0400289static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
290{
291 struct inode *inode = page->mapping->host;
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400292 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400293
294 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
295 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
296
297 nfs_pageio_cond_complete(pgio, page->index);
Wu Fengguang1b430be2010-10-26 14:21:26 -0700298 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400299 if (ret == -EAGAIN) {
300 redirty_page_for_writepage(wbc, page);
301 ret = 0;
302 }
303 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400304}
305
Trond Myklebuste261f512006-12-05 00:35:41 -0500306/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 * Write an mmapped page to the server.
308 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500309static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400311 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500312 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Trond Myklebustf758c8852007-07-22 19:27:32 -0400314 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
315 err = nfs_do_writepage(page, wbc, &pgio);
316 nfs_pageio_complete(&pgio);
317 if (err < 0)
318 return err;
319 if (pgio.pg_error < 0)
320 return pgio.pg_error;
321 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500322}
323
324int nfs_writepage(struct page *page, struct writeback_control *wbc)
325{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400326 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500327
Trond Myklebustf758c8852007-07-22 19:27:32 -0400328 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400330 return ret;
331}
332
333static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
334{
335 int ret;
336
337 ret = nfs_do_writepage(page, wbc, data);
338 unlock_page(page);
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
343{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400345 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400346 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 int err;
348
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400349 /* Stop dirtying of new pages while we sync */
350 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
351 nfs_wait_bit_killable, TASK_KILLABLE);
352 if (err)
353 goto out_err;
354
Chuck Lever91d5b472006-03-20 13:44:14 -0500355 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
356
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400357 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
Trond Myklebustf758c8852007-07-22 19:27:32 -0400358 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400359 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400360
361 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
362 smp_mb__after_clear_bit();
363 wake_up_bit(bitlock, NFS_INO_FLUSHING);
364
Trond Myklebustf758c8852007-07-22 19:27:32 -0400365 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400366 goto out_err;
367 err = pgio.pg_error;
368 if (err < 0)
369 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400370 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400371out_err:
372 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
375/*
376 * Insert a write request into an inode
377 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400378static int nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379{
380 struct nfs_inode *nfsi = NFS_I(inode);
381 int error;
382
Trond Myklebuste7d39062008-06-13 12:12:32 -0400383 error = radix_tree_preload(GFP_NOFS);
384 if (error != 0)
385 goto out;
386
387 /* Lock the request! */
388 nfs_lock_request_dontget(req);
389
390 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
Nick Piggin27852592008-02-04 23:48:37 -0800392 BUG_ON(error);
Trond Myklebust4d65c522011-03-25 14:15:11 -0400393 if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700394 inode->i_version++;
Trond Myklebust2df485a2010-12-07 22:39:17 -0500395 set_bit(PG_MAPPED, &req->wb_flags);
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400399 kref_get(&req->wb_kref);
Nick Piggin27852592008-02-04 23:48:37 -0800400 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
401 NFS_PAGE_TAG_LOCKED);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400402 spin_unlock(&inode->i_lock);
403 radix_tree_preload_end();
404out:
405 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406}
407
408/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800409 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 */
411static void nfs_inode_remove_request(struct nfs_page *req)
412{
Al Viro3d4ff432011-06-22 18:40:12 -0400413 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 struct nfs_inode *nfsi = NFS_I(inode);
415
416 BUG_ON (!NFS_WBACK_BUSY(req));
417
Trond Myklebust587142f2007-07-02 09:57:54 -0400418 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500419 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500420 ClearPagePrivate(req->wb_page);
Trond Myklebust2df485a2010-12-07 22:39:17 -0500421 clear_bit(PG_MAPPED, &req->wb_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
423 nfsi->npages--;
Trond Myklebust4d65c522011-03-25 14:15:11 -0400424 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 nfs_release_request(req);
426}
427
Trond Myklebust61822ab2006-12-05 00:35:42 -0500428static void
Fred6d884e82008-03-19 11:24:38 -0400429nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500430{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500431 __set_page_dirty_nobuffers(req->wb_page);
432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
435/*
436 * Add a request to the inode's commit list.
437 */
438static void
Fred Isamana861a1e2011-03-23 13:27:51 +0000439nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Al Viro3d4ff432011-06-22 18:40:12 -0400441 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 struct nfs_inode *nfsi = NFS_I(inode);
443
Trond Myklebust587142f2007-07-02 09:57:54 -0400444 spin_lock(&inode->i_lock);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400445 set_bit(PG_CLEAN, &(req)->wb_flags);
Trond Myklebust5c369682007-06-17 15:27:42 -0400446 radix_tree_tag_set(&nfsi->nfs_page_tree,
447 req->wb_index,
448 NFS_PAGE_TAG_COMMIT);
Trond Myklebustff778d02010-02-19 16:53:39 -0800449 nfsi->ncommit++;
Trond Myklebust587142f2007-07-02 09:57:54 -0400450 spin_unlock(&inode->i_lock);
Fred Isamana861a1e2011-03-23 13:27:51 +0000451 pnfs_mark_request_commit(req, lseg);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700452 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700453 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
Trond Myklebusta1803042006-12-05 00:45:19 -0500454 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400456
Trond Myklebuste468bae2008-06-13 13:25:22 -0400457static int
458nfs_clear_request_commit(struct nfs_page *req)
459{
460 struct page *page = req->wb_page;
461
462 if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
463 dec_zone_page_state(page, NR_UNSTABLE_NFS);
464 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
465 return 1;
466 }
467 return 0;
468}
469
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400470static inline
471int nfs_write_need_commit(struct nfs_write_data *data)
472{
Fred Isaman465d5242011-03-23 13:27:44 +0000473 if (data->verf.committed == NFS_DATA_SYNC)
474 return data->lseg == NULL;
475 else
476 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400477}
478
479static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000480int nfs_reschedule_unstable_write(struct nfs_page *req,
481 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400482{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400483 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Fred Isamana861a1e2011-03-23 13:27:51 +0000484 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400485 return 1;
486 }
487 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
Fred6d884e82008-03-19 11:24:38 -0400488 nfs_mark_request_dirty(req);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400489 return 1;
490 }
491 return 0;
492}
493#else
494static inline void
Fred Isamana861a1e2011-03-23 13:27:51 +0000495nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400496{
497}
498
Trond Myklebuste468bae2008-06-13 13:25:22 -0400499static inline int
500nfs_clear_request_commit(struct nfs_page *req)
501{
502 return 0;
503}
504
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400505static inline
506int nfs_write_need_commit(struct nfs_write_data *data)
507{
508 return 0;
509}
510
511static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000512int nfs_reschedule_unstable_write(struct nfs_page *req,
513 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400514{
515 return 0;
516}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517#endif
518
Trond Myklebust47c62562009-03-16 08:13:41 -0400519#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400520static int
521nfs_need_commit(struct nfs_inode *nfsi)
522{
523 return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
524}
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526/*
527 * nfs_scan_commit - Scan an inode for commit requests
528 * @inode: NFS inode to scan
529 * @dst: destination list
530 * @idx_start: lower bound of page->index to scan.
531 * @npages: idx_start + npages sets the upper bound to scan.
532 *
533 * Moves requests from the inode's 'commit' request list.
534 * The requests are *not* checked to ensure that they form a contiguous set.
535 */
536static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400537nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
539 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustff778d02010-02-19 16:53:39 -0800540 int ret;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000541
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400542 if (!nfs_need_commit(nfsi))
543 return 0;
544
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000545 spin_lock(&inode->i_lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800546 ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
547 if (ret > 0)
548 nfsi->ncommit -= ret;
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000549 spin_unlock(&inode->i_lock);
550
Trond Myklebust2928db12010-02-19 17:03:18 -0800551 if (nfs_need_commit(NFS_I(inode)))
552 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000553
Trond Myklebustff778d02010-02-19 16:53:39 -0800554 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500556#else
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400557static inline int nfs_need_commit(struct nfs_inode *nfsi)
558{
559 return 0;
560}
561
Trond Myklebustca52fec2007-04-17 17:22:13 -0400562static 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 -0500563{
564 return 0;
565}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566#endif
567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400569 * Search for an existing write request, and attempt to update
570 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400572 * If the attempt fails, then the existing request is flushed out
573 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400575static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
576 struct page *page,
577 unsigned int offset,
578 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400580 struct nfs_page *req;
581 unsigned int rqend;
582 unsigned int end;
583 int error;
584
585 if (!PagePrivate(page))
586 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400589 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 for (;;) {
Trond Myklebust277459d2006-12-05 00:35:35 -0500592 req = nfs_page_find_request_locked(page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400593 if (req == NULL)
594 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500595
Trond Myklebuste7d39062008-06-13 12:12:32 -0400596 rqend = req->wb_offset + req->wb_bytes;
597 /*
598 * Tell the caller to flush out the request if
599 * the offsets are non-contiguous.
600 * Note: nfs_flush_incompatible() will already
601 * have flushed out requests having wrong owners.
602 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400603 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400604 || end < req->wb_offset)
605 goto out_flushme;
606
607 if (nfs_set_page_tag_locked(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Trond Myklebuste7d39062008-06-13 12:12:32 -0400610 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400611 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400612 error = nfs_wait_on_request(req);
613 nfs_release_request(req);
614 if (error != 0)
615 goto out_err;
616 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Trond Myklebustff778d02010-02-19 16:53:39 -0800619 if (nfs_clear_request_commit(req) &&
Fred Isamana861a1e2011-03-23 13:27:51 +0000620 radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
621 req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) {
Trond Myklebustff778d02010-02-19 16:53:39 -0800622 NFS_I(inode)->ncommit--;
Fred Isamana861a1e2011-03-23 13:27:51 +0000623 pnfs_clear_request_commit(req);
624 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 /* Okay, the request matches. Update the region */
627 if (offset < req->wb_offset) {
628 req->wb_offset = offset;
629 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (end > rqend)
632 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400633 else
634 req->wb_bytes = rqend - req->wb_offset;
635out_unlock:
636 spin_unlock(&inode->i_lock);
637 return req;
638out_flushme:
639 spin_unlock(&inode->i_lock);
640 nfs_release_request(req);
641 error = nfs_wb_page(inode, page);
642out_err:
643 return ERR_PTR(error);
644}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Trond Myklebuste7d39062008-06-13 12:12:32 -0400646/*
647 * Try to update an existing write request, or create one if there is none.
648 *
649 * Note: Should always be called with the Page Lock held to prevent races
650 * if we have to add a new request. Also assumes that the caller has
651 * already called nfs_flush_incompatible() if necessary.
652 */
653static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
654 struct page *page, unsigned int offset, unsigned int bytes)
655{
656 struct inode *inode = page->mapping->host;
657 struct nfs_page *req;
658 int error;
659
660 req = nfs_try_to_update_request(inode, page, offset, bytes);
661 if (req != NULL)
662 goto out;
663 req = nfs_create_request(ctx, inode, page, offset, bytes);
664 if (IS_ERR(req))
665 goto out;
666 error = nfs_inode_add_request(inode, req);
667 if (error != 0) {
668 nfs_release_request(req);
669 req = ERR_PTR(error);
670 }
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400671out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400672 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
Trond Myklebuste7d39062008-06-13 12:12:32 -0400675static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
676 unsigned int offset, unsigned int count)
677{
678 struct nfs_page *req;
679
680 req = nfs_setup_write_request(ctx, page, offset, count);
681 if (IS_ERR(req))
682 return PTR_ERR(req);
683 /* Update file length */
684 nfs_grow_file(page, offset, count);
685 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400686 nfs_mark_request_dirty(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400687 nfs_clear_page_tag_locked(req);
688 return 0;
689}
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691int nfs_flush_incompatible(struct file *file, struct page *page)
692{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400693 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500695 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 /*
697 * Look for a request corresponding to this page. If there
698 * is one, and it belongs to another file, we flush it out
699 * before we try to copy anything into the page. Do this
700 * due to the lack of an ACCESS-type call in NFSv2.
701 * Also do the same if we find a request from an existing
702 * dropped page.
703 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500704 do {
705 req = nfs_page_find_request(page);
706 if (req == NULL)
707 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400708 do_flush = req->wb_page != page || req->wb_context != ctx ||
709 req->wb_lock_context->lockowner != current->files ||
710 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500712 if (!do_flush)
713 return 0;
714 status = nfs_wb_page(page->mapping->host, page);
715 } while (status == 0);
716 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500720 * If the page cache is marked as unsafe or invalid, then we can't rely on
721 * the PageUptodate() flag. In this case, we will need to turn off
722 * write optimisations that depend on the page contents being correct.
723 */
724static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
725{
726 return PageUptodate(page) &&
727 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
728}
729
730/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 * Update and possibly write a cached page of an NFS file.
732 *
733 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
734 * things with a page scheduled for an RPC call (e.g. invalidate it).
735 */
736int nfs_updatepage(struct file *file, struct page *page,
737 unsigned int offset, unsigned int count)
738{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400739 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 int status = 0;
742
Chuck Lever91d5b472006-03-20 13:44:14 -0500743 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
744
Chuck Lever48186c72008-06-11 17:56:05 -0400745 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800746 file->f_path.dentry->d_parent->d_name.name,
747 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400748 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500751 * is up to date, it may be more efficient to extend the write
752 * to cover the entire page in order to avoid fragmentation
753 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500755 if (nfs_write_pageuptodate(page, inode) &&
756 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100757 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500758 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
761
Trond Myklebuste21195a2006-12-05 00:35:39 -0500762 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400763 if (status < 0)
764 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700765 else
766 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Chuck Lever48186c72008-06-11 17:56:05 -0400768 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 return status;
771}
772
Fred Isamana861a1e2011-03-23 13:27:51 +0000773static void nfs_writepage_release(struct nfs_page *req,
774 struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400776 struct page *page = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Fred Isamana861a1e2011-03-23 13:27:51 +0000778 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400779 nfs_inode_remove_request(req);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400780 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400781 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782}
783
Trond Myklebust3ff75762007-07-14 15:40:00 -0400784static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785{
786 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
787 case FLUSH_HIGHPRI:
788 return RPC_PRIORITY_HIGH;
789 case FLUSH_LOWPRI:
790 return RPC_PRIORITY_LOW;
791 }
792 return RPC_PRIORITY_NORMAL;
793}
794
Fred Isamana69aef12011-03-03 15:13:47 +0000795int nfs_initiate_write(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000796 struct rpc_clnt *clnt,
797 const struct rpc_call_ops *call_ops,
798 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Andy Adamsond138d5d2011-03-03 15:13:41 +0000800 struct inode *inode = data->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400801 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400802 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400803 struct rpc_message msg = {
804 .rpc_argp = &data->args,
805 .rpc_resp = &data->res,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000806 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400807 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400808 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +0000809 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -0400810 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400811 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400812 .callback_ops = call_ops,
813 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500814 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -0400815 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400816 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400817 };
Trond Myklebust2c61be02010-04-09 19:54:50 -0400818 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
Andy Adamsond138d5d2011-03-03 15:13:41 +0000820 /* Set up the initial task struct. */
821 NFS_PROTO(inode)->write_setup(data, &msg);
822
823 dprintk("NFS: %5u initiated write call "
824 "(req %s/%lld, %u bytes @ offset %llu)\n",
825 data->task.tk_pid,
826 inode->i_sb->s_id,
827 (long long)NFS_FILEID(inode),
828 data->args.count,
829 (unsigned long long)data->args.offset);
830
831 task = rpc_run_task(&task_setup_data);
832 if (IS_ERR(task)) {
833 ret = PTR_ERR(task);
834 goto out;
835 }
836 if (how & FLUSH_SYNC) {
837 ret = rpc_wait_for_completion_task(task);
838 if (ret == 0)
839 ret = task->tk_status;
840 }
841 rpc_put_task(task);
842out:
843 return ret;
844}
Fred Isamana69aef12011-03-03 15:13:47 +0000845EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +0000846
847/*
848 * Set up the argument/result storage required for the RPC call.
849 */
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400850static void nfs_write_rpcsetup(struct nfs_page *req,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000851 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000852 unsigned int count, unsigned int offset,
853 int how)
854{
Al Viro3d4ff432011-06-22 18:40:12 -0400855 struct inode *inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 /* Set up the RPC argument and reply structs
858 * NB: take care not to mess about with data->commit et al. */
859
860 data->req = req;
Al Viro3d4ff432011-06-22 18:40:12 -0400861 data->inode = inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000862 data->cred = req->wb_context->cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 data->args.fh = NFS_FH(inode);
865 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -0400866 /* pnfs_set_layoutcommit needs this */
867 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 data->args.pgbase = req->wb_pgbase + offset;
869 data->args.pages = data->pagevec;
870 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500871 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400872 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400873 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -0400874 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
875 case 0:
876 break;
877 case FLUSH_COND_STABLE:
878 if (nfs_need_commit(NFS_I(inode)))
879 break;
880 default:
881 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 data->res.fattr = &data->fattr;
885 data->res.count = count;
886 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400887 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400888}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400890static int nfs_do_write(struct nfs_write_data *data,
891 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400892 int how)
893{
Stephen Rothwell5f00bcb2011-07-25 13:59:46 -0400894 struct inode *inode = data->args.context->dentry->d_inode;
Andy Adamson0382b742011-03-03 15:13:45 +0000895
Andy Adamsond138d5d2011-03-03 15:13:41 +0000896 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
Trond Myklebust275acaa2011-07-12 13:42:02 -0400899static int nfs_do_multiple_writes(struct list_head *head,
900 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -0400901 int how)
902{
903 struct nfs_write_data *data;
904 int ret = 0;
905
906 while (!list_empty(head)) {
907 int ret2;
908
909 data = list_entry(head->next, struct nfs_write_data, list);
910 list_del_init(&data->list);
911
Trond Myklebustdce81292011-07-13 15:59:19 -0400912 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400913 if (ret == 0)
914 ret = ret2;
915 }
916 return ret;
917}
918
Fred6d884e82008-03-19 11:24:38 -0400919/* If a nfs_flush_* function fails, it should remove reqs from @head and
920 * call this on each, which will prepare them to be retried on next
921 * writeback using standard nfs.
922 */
923static void nfs_redirty_request(struct nfs_page *req)
924{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400925 struct page *page = req->wb_page;
926
Fred6d884e82008-03-19 11:24:38 -0400927 nfs_mark_request_dirty(req);
Fred6d884e82008-03-19 11:24:38 -0400928 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400929 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400930}
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932/*
933 * Generate multiple small requests to write out a single
934 * contiguous dirty area on one page.
935 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400936static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Fred Isamanc76069b2011-03-03 15:13:48 +0000938 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 struct page *page = req->wb_page;
940 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -0400941 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700942 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400944 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 nfs_list_remove_request(req);
947
Trond Myklebustb31268a2011-03-21 17:02:00 -0400948 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
949 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
950 desc->pg_count > wsize))
951 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
952
953
Trond Myklebust275acaa2011-07-12 13:42:02 -0400954 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000955 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700956 do {
957 size_t len = min(nbytes, wsize);
958
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400959 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 if (!data)
961 goto out_bad;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400962 data->pagevec[0] = page;
Trond Myklebustf13c3622011-09-12 11:47:53 -0400963 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400964 list_add(&data->list, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700966 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400967 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700968 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 atomic_set(&req->wb_complete, requests);
Trond Myklebust50828d72011-07-12 13:42:02 -0400970 desc->pg_rpc_callops = &nfs_write_partial_ops;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400971 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973out_bad:
Trond Myklebust275acaa2011-07-12 13:42:02 -0400974 while (!list_empty(res)) {
975 data = list_entry(res->next, struct nfs_write_data, list);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400976 list_del(&data->list);
Fred Isaman0da2a4ac2011-01-19 14:18:50 -0500977 nfs_writedata_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500979 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 return -ENOMEM;
981}
982
983/*
984 * Create an RPC task for the given write request and kick it.
985 * The page must have been locked by the caller.
986 *
987 * It may happen that the page we're passed is not marked dirty.
988 * This is the case if nfs_updatepage detects a conflicting request
989 * that has been written but not committed.
990 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400991static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992{
993 struct nfs_page *req;
994 struct page **pages;
995 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000996 struct list_head *head = &desc->pg_list;
Peng Tao3b609182011-07-15 03:33:42 -0400997 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Fred Isamanc76069b2011-03-03 15:13:48 +0000999 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
1000 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +00001001 if (!data) {
1002 while (!list_empty(head)) {
1003 req = nfs_list_entry(head->next);
1004 nfs_list_remove_request(req);
1005 nfs_redirty_request(req);
1006 }
1007 ret = -ENOMEM;
1008 goto out;
1009 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 while (!list_empty(head)) {
1012 req = nfs_list_entry(head->next);
1013 nfs_list_remove_request(req);
1014 nfs_list_add_request(req, &data->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
1017 req = nfs_list_entry(data->pages.next);
1018
Trond Myklebustb31268a2011-03-21 17:02:00 -04001019 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1020 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1021 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 /* Set up the argument struct */
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001024 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001025 list_add(&data->list, res);
Trond Myklebust50828d72011-07-12 13:42:02 -04001026 desc->pg_rpc_callops = &nfs_write_full_ops;
Fred Isaman44b83792011-03-03 15:13:44 +00001027out:
Fred Isaman44b83792011-03-03 15:13:44 +00001028 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Trond Myklebustdce81292011-07-13 15:59:19 -04001031int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1032{
1033 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1034 return nfs_flush_multi(desc, head);
1035 return nfs_flush_one(desc, head);
1036}
1037
1038static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001039{
Trond Myklebust275acaa2011-07-12 13:42:02 -04001040 LIST_HEAD(head);
1041 int ret;
1042
Trond Myklebustdce81292011-07-13 15:59:19 -04001043 ret = nfs_generic_flush(desc, &head);
Trond Myklebust50828d72011-07-12 13:42:02 -04001044 if (ret == 0)
1045 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
Trond Myklebustdce81292011-07-13 15:59:19 -04001046 desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001047 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001048}
Trond Myklebust1751c362011-06-10 13:30:23 -04001049
1050static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1051 .pg_test = nfs_generic_pg_test,
1052 .pg_doio = nfs_generic_pg_writepages,
1053};
1054
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001055void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
Trond Myklebust1751c362011-06-10 13:30:23 -04001056 struct inode *inode, int ioflags)
1057{
1058 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1059 NFS_SERVER(inode)->wsize, ioflags);
1060}
1061
Trond Myklebustdce81292011-07-13 15:59:19 -04001062void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1063{
1064 pgio->pg_ops = &nfs_pageio_write_ops;
1065 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1066}
Trond Myklebust1f945352011-07-13 15:59:57 -04001067EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001068
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001069static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1070 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Trond Myklebust1751c362011-06-10 13:30:23 -04001072 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1073 nfs_pageio_init_write_mds(pgio, inode, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
1076/*
1077 * Handle a write reply that flushed part of a page.
1078 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001079static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001081 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Chuck Lever48186c72008-06-11 17:56:05 -04001083 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1084 task->tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001085 data->req->wb_context->dentry->d_inode->i_sb->s_id,
Chuck Lever48186c72008-06-11 17:56:05 -04001086 (long long)
Al Viro3d4ff432011-06-22 18:40:12 -04001087 NFS_FILEID(data->req->wb_context->dentry->d_inode),
Chuck Lever48186c72008-06-11 17:56:05 -04001088 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001090 nfs_writeback_done(task, data);
1091}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001092
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001093static void nfs_writeback_release_partial(void *calldata)
1094{
1095 struct nfs_write_data *data = calldata;
1096 struct nfs_page *req = data->req;
1097 struct page *page = req->wb_page;
1098 int status = data->task.tk_status;
1099
1100 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001101 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001102 nfs_context_set_write_error(req->wb_context, status);
1103 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001104 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 }
1106
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001107 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001108 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001109
Trond Myklebust587142f2007-07-02 09:57:54 -04001110 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001111 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1112 /* Do nothing we need to resend the writes */
1113 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1114 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1115 dprintk(" defer commit\n");
1116 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1117 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1118 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1119 dprintk(" server reboot detected\n");
1120 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001121 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001122 } else
1123 dprintk(" OK\n");
1124
1125out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001127 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001128 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
Andy Adamsondef6ed72009-04-01 09:22:26 -04001131#if defined(CONFIG_NFS_V4_1)
1132void nfs_write_prepare(struct rpc_task *task, void *calldata)
1133{
1134 struct nfs_write_data *data = calldata;
Andy Adamsondef6ed72009-04-01 09:22:26 -04001135
Trond Myklebust035168a2010-06-16 09:52:26 -04001136 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
1137 &data->args.seq_args,
Andy Adamsondef6ed72009-04-01 09:22:26 -04001138 &data->res.seq_res, 1, task))
1139 return;
1140 rpc_call_start(task);
1141}
1142#endif /* CONFIG_NFS_V4_1 */
1143
Trond Myklebust788e7a82006-03-20 13:44:27 -05001144static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001145#if defined(CONFIG_NFS_V4_1)
1146 .rpc_call_prepare = nfs_write_prepare,
1147#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001148 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001149 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001150};
1151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152/*
1153 * Handle a write reply that flushes a whole page.
1154 *
1155 * FIXME: There is an inherent race with invalidate_inode_pages and
1156 * writebacks since the page->count is kept > 1 for as long
1157 * as the page has a write request pending.
1158 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001159static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001161 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001163 nfs_writeback_done(task, data);
1164}
1165
1166static void nfs_writeback_release_full(void *calldata)
1167{
1168 struct nfs_write_data *data = calldata;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001169 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001170
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 /* Update attributes as result of writeback. */
1172 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001173 struct nfs_page *req = nfs_list_entry(data->pages.next);
1174 struct page *page = req->wb_page;
1175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Chuck Lever48186c72008-06-11 17:56:05 -04001178 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1179 data->task.tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001180 req->wb_context->dentry->d_inode->i_sb->s_id,
1181 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 req->wb_bytes,
1183 (long long)req_offset(req));
1184
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001185 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001186 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001187 nfs_context_set_write_error(req->wb_context, status);
1188 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001189 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001192 if (nfs_write_need_commit(data)) {
1193 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001194 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001195 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 goto next;
1197 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001198 dprintk(" OK\n");
1199remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001202 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001203 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001205 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
Trond Myklebust788e7a82006-03-20 13:44:27 -05001208static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001209#if defined(CONFIG_NFS_V4_1)
1210 .rpc_call_prepare = nfs_write_prepare,
1211#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001212 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001213 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001214};
1215
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/*
1218 * This function is called when the WRITE call is complete.
1219 */
Fred Isaman13602892011-02-11 15:42:38 +00001220void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 struct nfs_writeargs *argp = &data->args;
1223 struct nfs_writeres *resp = &data->res;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001224 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Chuck Levera3f565b2007-01-31 12:14:01 -05001226 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 task->tk_pid, task->tk_status);
1228
Chuck Leverf551e442006-09-20 14:33:04 -04001229 /*
1230 * ->write_done will attempt to use post-op attributes to detect
1231 * conflicting writes by other clients. A strict interpretation
1232 * of close-to-open would allow us to continue caching even if
1233 * another writer had changed the file, but some applications
1234 * depend on tighter cache coherency when writing.
1235 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001236 status = NFS_PROTO(data->inode)->write_done(task, data);
1237 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001238 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001239 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1242 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1243 /* We tried a write call, but the server did not
1244 * commit data to stable storage even though we
1245 * requested it.
1246 * Note: There is a known bug in Tru64 < 5.0 in which
1247 * the server reports NFS_DATA_SYNC, but performs
1248 * NFS_FILE_SYNC. We therefore implement this checking
1249 * as a dprintk() in order to avoid filling syslog.
1250 */
1251 static unsigned long complain;
1252
Fred Isamana69aef12011-03-03 15:13:47 +00001253 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001255 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 " (committed = %d) != (stable = %d)\n",
Rakib Mullick2b72c9c2011-11-01 12:23:42 +06001257 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 resp->verf->committed, argp->stable);
1259 complain = jiffies + 300 * HZ;
1260 }
1261 }
1262#endif
1263 /* Is this a short write? */
1264 if (task->tk_status >= 0 && resp->count < argp->count) {
1265 static unsigned long complain;
1266
Chuck Lever91d5b472006-03-20 13:44:14 -05001267 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 /* Has the server at least made some progress? */
1270 if (resp->count != 0) {
1271 /* Was this an NFSv2 write or an NFSv3 stable write? */
1272 if (resp->verf->committed != NFS_UNSTABLE) {
1273 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001274 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 argp->offset += resp->count;
1276 argp->pgbase += resp->count;
1277 argp->count -= resp->count;
1278 } else {
1279 /* Resend as a stable write in order to avoid
1280 * headaches in the case of a server crash.
1281 */
1282 argp->stable = NFS_FILE_SYNC;
1283 }
Trond Myklebustd00c5d42011-10-19 12:17:29 -07001284 rpc_restart_call_prepare(task);
Fred Isaman13602892011-02-11 15:42:38 +00001285 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
1287 if (time_before(complain, jiffies)) {
1288 printk(KERN_WARNING
1289 "NFS: Server wrote zero bytes, expected %u.\n",
1290 argp->count);
1291 complain = jiffies + 300 * HZ;
1292 }
1293 /* Can't do anything about it except throw an error. */
1294 task->tk_status = -EIO;
1295 }
Fred Isaman13602892011-02-11 15:42:38 +00001296 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
1298
1299
1300#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001301static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1302{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001303 int ret;
1304
Trond Myklebust71d0a612010-04-22 15:35:57 -04001305 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1306 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001307 if (!may_wait)
1308 return 0;
1309 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1310 NFS_INO_COMMIT,
1311 nfs_wait_bit_killable,
1312 TASK_KILLABLE);
1313 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001314}
1315
Fred Isamane0c2b382011-03-23 13:27:53 +00001316void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001317{
1318 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1319 smp_mb__after_clear_bit();
1320 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1321}
Fred Isamane0c2b382011-03-23 13:27:53 +00001322EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001323
Fred Isamane0c2b382011-03-23 13:27:53 +00001324void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Trond Myklebust383ba712008-02-19 20:04:20 -05001326 struct nfs_write_data *wdata = data;
1327
Fred Isaman988b6dc2011-03-23 13:27:52 +00001328 put_lseg(wdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -05001329 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 nfs_commit_free(wdata);
1331}
Fred Isamane0c2b382011-03-23 13:27:53 +00001332EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Fred Isamane0c2b382011-03-23 13:27:53 +00001334int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001335 const struct rpc_call_ops *call_ops,
1336 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
Trond Myklebust07737692007-10-25 18:42:54 -04001338 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001339 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001340 struct rpc_message msg = {
1341 .rpc_argp = &data->args,
1342 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001343 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001344 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001345 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001346 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001347 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001348 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001349 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001350 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001351 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001352 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001353 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001354 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001355 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001356 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Chuck Levera3f565b2007-01-31 12:14:01 -05001358 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001359
Trond Myklebust07737692007-10-25 18:42:54 -04001360 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001361 if (IS_ERR(task))
1362 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001363 if (how & FLUSH_SYNC)
1364 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001365 rpc_put_task(task);
1366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367}
Fred Isamane0c2b382011-03-23 13:27:53 +00001368EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001371 * Set up the argument/result storage required for the RPC call.
1372 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001373void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001374 struct list_head *head,
1375 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001376{
1377 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001378 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001379
1380 /* Set up the RPC argument and reply structs
1381 * NB: take care not to mess about with data->commit et al. */
1382
1383 list_splice_init(head, &data->pages);
1384
1385 data->inode = inode;
1386 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001387 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001388 data->mds_ops = &nfs_commit_ops;
1389
1390 data->args.fh = NFS_FH(data->inode);
1391 /* Note: we always request a commit of the entire inode */
1392 data->args.offset = 0;
1393 data->args.count = 0;
1394 data->args.context = get_nfs_open_context(first->wb_context);
1395 data->res.count = 0;
1396 data->res.fattr = &data->fattr;
1397 data->res.verf = &data->verf;
1398 nfs_fattr_init(&data->fattr);
1399}
Fred Isamane0c2b382011-03-23 13:27:53 +00001400EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001401
Fred Isamane0c2b382011-03-23 13:27:53 +00001402void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001403 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001404{
1405 struct nfs_page *req;
1406
1407 while (!list_empty(page_list)) {
1408 req = nfs_list_entry(page_list->next);
1409 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001410 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001411 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1412 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1413 BDI_RECLAIMABLE);
1414 nfs_clear_page_tag_locked(req);
1415 }
1416}
Fred Isamane0c2b382011-03-23 13:27:53 +00001417EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001418
Fred Isaman9ace33c2011-03-23 13:27:45 +00001419/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 * Commit dirty pages
1421 */
1422static int
Chuck Lever40859d72005-11-30 18:09:02 -05001423nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001427 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 if (!data)
1430 goto out_bad;
1431
1432 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001433 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001434 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001436 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001437 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 return -ENOMEM;
1439}
1440
1441/*
1442 * COMMIT call returned
1443 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001444static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001446 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Chuck Levera3f565b2007-01-31 12:14:01 -05001448 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 task->tk_pid, task->tk_status);
1450
Trond Myklebust788e7a82006-03-20 13:44:27 -05001451 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001452 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001453}
1454
Fred Isamane0c2b382011-03-23 13:27:53 +00001455void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001456{
Fred Isaman5917ce82011-03-23 13:27:48 +00001457 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001458 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 while (!list_empty(&data->pages)) {
1461 req = nfs_list_entry(data->pages.next);
1462 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -04001463 nfs_clear_request_commit(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Chuck Lever48186c72008-06-11 17:56:05 -04001465 dprintk("NFS: commit (%s/%lld %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001466 req->wb_context->dentry->d_sb->s_id,
1467 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 req->wb_bytes,
1469 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001470 if (status < 0) {
1471 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001473 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 goto next;
1475 }
1476
1477 /* Okay, COMMIT succeeded, apparently. Check the verifier
1478 * returned by the server against all stored verfs. */
1479 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1480 /* We have a match */
1481 nfs_inode_remove_request(req);
1482 dprintk(" OK\n");
1483 goto next;
1484 }
1485 /* We have a mismatch. Write the page again */
1486 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001487 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001489 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001491}
Fred Isamane0c2b382011-03-23 13:27:53 +00001492EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001493
1494static void nfs_commit_release(void *calldata)
1495{
1496 struct nfs_write_data *data = calldata;
1497
1498 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001499 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001500 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001502
1503static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001504#if defined(CONFIG_NFS_V4_1)
1505 .rpc_call_prepare = nfs_write_prepare,
1506#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001507 .rpc_call_done = nfs_commit_done,
1508 .rpc_release = nfs_commit_release,
1509};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Trond Myklebustb608b282010-07-30 15:31:54 -04001511int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001514 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001515 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Trond Myklebustb8413f92011-03-21 15:37:01 -04001517 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1518 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001519 goto out_mark_dirty;
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001520 res = nfs_scan_commit(inode, &head, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001522 int error;
1523
1524 error = pnfs_commit_list(inode, &head, how);
1525 if (error == PNFS_NOT_ATTEMPTED)
1526 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001527 if (error < 0)
1528 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001529 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001530 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001531 error = wait_on_bit(&NFS_I(inode)->flags,
1532 NFS_INO_COMMIT,
1533 nfs_wait_bit_killable,
1534 TASK_KILLABLE);
1535 if (error < 0)
1536 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001537 } else
1538 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001539 return res;
1540 /* Note: If we exit without ensuring that the commit is complete,
1541 * we must mark the inode as dirty. Otherwise, future calls to
1542 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1543 * that the data is on the disk.
1544 */
1545out_mark_dirty:
1546 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 return res;
1548}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001549
1550static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1551{
Trond Myklebust420e3642010-02-19 17:00:02 -08001552 struct nfs_inode *nfsi = NFS_I(inode);
1553 int flags = FLUSH_SYNC;
1554 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001555
Jeff Layton3236c3e2011-10-11 09:49:21 -04001556 /* no commits means nothing needs to be done */
1557 if (!nfsi->ncommit)
1558 return ret;
1559
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001560 if (wbc->sync_mode == WB_SYNC_NONE) {
1561 /* Don't commit yet if this is a non-blocking flush and there
1562 * are a lot of outstanding writes for this mapping.
1563 */
1564 if (nfsi->ncommit <= (nfsi->npages >> 1))
1565 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001566
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001567 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001568 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001569 }
1570
Trond Myklebust420e3642010-02-19 17:00:02 -08001571 ret = nfs_commit_inode(inode, flags);
1572 if (ret >= 0) {
1573 if (wbc->sync_mode == WB_SYNC_NONE) {
1574 if (ret < wbc->nr_to_write)
1575 wbc->nr_to_write -= ret;
1576 else
1577 wbc->nr_to_write = 0;
1578 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001579 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001580 }
1581out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001582 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1583 return ret;
1584}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001585#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001586static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1587{
1588 return 0;
1589}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590#endif
1591
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001592int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1593{
Andy Adamson863a3c62011-03-23 13:27:54 +00001594 int ret;
1595
1596 ret = nfs_commit_unstable_pages(inode, wbc);
1597 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001598 int status;
1599 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001600
Wu Fengguang846d5a02011-05-05 21:10:38 -06001601 if (wbc->sync_mode == WB_SYNC_NONE)
Andy Adamsonef311532011-03-12 02:58:10 -05001602 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001603
1604 status = pnfs_layoutcommit_inode(inode, sync);
1605 if (status < 0)
1606 return status;
1607 }
1608 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001609}
1610
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001611/*
1612 * flush the inode to disk.
1613 */
1614int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001615{
1616 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001617 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001618 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001619 .range_start = 0,
1620 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001621 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001622
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001623 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001624}
1625
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001626int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1627{
1628 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001629 int ret = 0;
1630
1631 BUG_ON(!PageLocked(page));
1632 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001633 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001634 req = nfs_page_find_request(page);
1635 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001636 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001637 if (nfs_lock_request_dontget(req)) {
1638 nfs_inode_remove_request(req);
1639 /*
1640 * In case nfs_inode_remove_request has marked the
1641 * page as being dirty
1642 */
1643 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1644 nfs_unlock_request(req);
1645 break;
1646 }
1647 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001648 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001649 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001650 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001651 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001652 return ret;
1653}
1654
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001655/*
1656 * Write back all requests on one page - we do this before reading it.
1657 */
1658int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001659{
1660 loff_t range_start = page_offset(page);
1661 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001662 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001663 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001664 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001665 .range_start = range_start,
1666 .range_end = range_end,
1667 };
1668 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001669
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001670 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001671 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001672 if (clear_page_dirty_for_io(page)) {
1673 ret = nfs_writepage_locked(page, &wbc);
1674 if (ret < 0)
1675 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001676 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001677 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001678 if (!PagePrivate(page))
1679 break;
1680 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001681 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001682 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001683 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001684 return 0;
1685out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001686 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001687}
1688
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001689#ifdef CONFIG_MIGRATION
1690int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001691 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001692{
Jeff Layton2da95652011-10-12 10:57:42 -04001693 /*
1694 * If PagePrivate is set, then the page is currently associated with
1695 * an in-progress read or write request. Don't try to migrate it.
1696 *
1697 * FIXME: we could do this in principle, but we'll need a way to ensure
1698 * that we can safely release the inode reference while holding
1699 * the page lock.
1700 */
1701 if (PagePrivate(page))
1702 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001703
Trond Myklebust7549ad52010-02-08 09:32:34 -05001704 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001705
Mel Gormana6bc32b2012-01-12 17:19:43 -08001706 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001707}
1708#endif
1709
David Howellsf7b422b2006-06-09 09:34:33 -04001710int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
1712 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1713 sizeof(struct nfs_write_data),
1714 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001715 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 if (nfs_wdata_cachep == NULL)
1717 return -ENOMEM;
1718
Matthew Dobson93d23412006-03-26 01:37:50 -08001719 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1720 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 if (nfs_wdata_mempool == NULL)
1722 return -ENOMEM;
1723
Matthew Dobson93d23412006-03-26 01:37:50 -08001724 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1725 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (nfs_commit_mempool == NULL)
1727 return -ENOMEM;
1728
Peter Zijlstra89a09142007-03-16 13:38:26 -08001729 /*
1730 * NFS congestion size, scale with available memory.
1731 *
1732 * 64MB: 8192k
1733 * 128MB: 11585k
1734 * 256MB: 16384k
1735 * 512MB: 23170k
1736 * 1GB: 32768k
1737 * 2GB: 46340k
1738 * 4GB: 65536k
1739 * 8GB: 92681k
1740 * 16GB: 131072k
1741 *
1742 * This allows larger machines to have larger/more transfers.
1743 * Limit the default to 256M
1744 */
1745 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1746 if (nfs_congestion_kb > 256*1024)
1747 nfs_congestion_kb = 256*1024;
1748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 return 0;
1750}
1751
David Brownell266bee82006-06-27 12:59:15 -07001752void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
1754 mempool_destroy(nfs_commit_mempool);
1755 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001756 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757}
1758