blob: 2c68818f68ac056b8587c22bf40f8f5d229a6403 [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{
Trond Myklebust383ba712008-02-19 20:04:20 -0500103 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 nfs_writedata_free(wdata);
105}
106
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400107static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
108{
109 ctx->error = error;
110 smp_wmb();
111 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
112}
113
Trond Myklebust277459d2006-12-05 00:35:35 -0500114static struct nfs_page *nfs_page_find_request_locked(struct page *page)
115{
116 struct nfs_page *req = NULL;
117
118 if (PagePrivate(page)) {
119 req = (struct nfs_page *)page_private(page);
120 if (req != NULL)
Trond Myklebustc03b4022007-06-17 13:26:38 -0400121 kref_get(&req->wb_kref);
Trond Myklebust277459d2006-12-05 00:35:35 -0500122 }
123 return req;
124}
125
126static struct nfs_page *nfs_page_find_request(struct page *page)
127{
Trond Myklebust587142f2007-07-02 09:57:54 -0400128 struct inode *inode = page->mapping->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500129 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500130
Trond Myklebust587142f2007-07-02 09:57:54 -0400131 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500132 req = nfs_page_find_request_locked(page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400133 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500134 return req;
135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137/* Adjust the file length if we're writing beyond the end */
138static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
139{
140 struct inode *inode = page->mapping->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400141 loff_t end, i_size;
142 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Trond Myklebusta3d01452008-06-11 12:21:19 -0400144 spin_lock(&inode->i_lock);
145 i_size = i_size_read(inode);
146 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 if (i_size > 0 && page->index < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400148 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + ((loff_t)offset+count);
150 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400151 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400153 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
154out:
155 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
Trond Myklebusta301b772007-02-06 11:07:15 -0800158/* A writeback failed: mark the page as bad, and invalidate the page cache */
159static void nfs_set_pageerror(struct page *page)
160{
161 SetPageError(page);
162 nfs_zap_mapping(page->mapping->host, page->mapping);
163}
164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165/* We can set the PG_uptodate flag if we see that a write request
166 * covers the full page.
167 */
168static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
169{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 if (PageUptodate(page))
171 return;
172 if (base != 0)
173 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500174 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500176 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179static int wb_priority(struct writeback_control *wbc)
180{
181 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400182 return FLUSH_HIGHPRI | FLUSH_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100183 if (wbc->for_kupdate || wbc->for_background)
Trond Myklebustb31268a2011-03-21 17:02:00 -0400184 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
185 return FLUSH_COND_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186}
187
188/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800189 * NFS congestion control
190 */
191
192int nfs_congestion_kb;
193
194#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
195#define NFS_CONGESTION_OFF_THRESH \
196 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
197
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400198static int nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800199{
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400200 int ret = test_set_page_writeback(page);
201
202 if (!ret) {
Peter Zijlstra89a09142007-03-16 13:38:26 -0800203 struct inode *inode = page->mapping->host;
204 struct nfs_server *nfss = NFS_SERVER(inode);
205
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400206 page_cache_get(page);
Peter Zijlstra277866a2007-05-08 00:35:12 -0700207 if (atomic_long_inc_return(&nfss->writeback) >
Jens Axboe8aa7e842009-07-09 14:52:32 +0200208 NFS_CONGESTION_ON_THRESH) {
209 set_bdi_congested(&nfss->backing_dev_info,
210 BLK_RW_ASYNC);
211 }
Peter Zijlstra89a09142007-03-16 13:38:26 -0800212 }
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400213 return ret;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800214}
215
216static void nfs_end_page_writeback(struct page *page)
217{
218 struct inode *inode = page->mapping->host;
219 struct nfs_server *nfss = NFS_SERVER(inode);
220
221 end_page_writeback(page);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400222 page_cache_release(page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700223 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200224 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800225}
226
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400227static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
Trond Myklebuste261f512006-12-05 00:35:41 -0500228{
Trond Myklebust587142f2007-07-02 09:57:54 -0400229 struct inode *inode = page->mapping->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500230 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500231 int ret;
232
Trond Myklebust587142f2007-07-02 09:57:54 -0400233 spin_lock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400234 for (;;) {
Trond Myklebuste261f512006-12-05 00:35:41 -0500235 req = nfs_page_find_request_locked(page);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400236 if (req == NULL)
237 break;
Fred Isaman9994b622012-03-08 17:29:34 -0500238 if (nfs_lock_request_dontget(req))
Trond Myklebuste261f512006-12-05 00:35:41 -0500239 break;
240 /* Note: If we hold the page lock, as is the case in nfs_writepage,
Fred Isaman9994b622012-03-08 17:29:34 -0500241 * then the call to nfs_lock_request_dontget() will always
Trond Myklebuste261f512006-12-05 00:35:41 -0500242 * succeed provided that someone hasn't already marked the
243 * request as dirty (in which case we don't care).
244 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400245 spin_unlock(&inode->i_lock);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400246 if (!nonblock)
247 ret = nfs_wait_on_request(req);
248 else
249 ret = -EAGAIN;
Trond Myklebuste261f512006-12-05 00:35:41 -0500250 nfs_release_request(req);
251 if (ret != 0)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400252 return ERR_PTR(ret);
Trond Myklebust587142f2007-07-02 09:57:54 -0400253 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500254 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400255 spin_unlock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400256 return req;
257}
258
259/*
260 * Find an associated nfs write request, and prepare to flush it out
261 * May return an error if the user signalled nfs_wait_on_request().
262 */
263static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400264 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400265{
266 struct nfs_page *req;
267 int ret = 0;
268
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400269 req = nfs_find_and_lock_request(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400270 if (!req)
271 goto out;
272 ret = PTR_ERR(req);
273 if (IS_ERR(req))
274 goto out;
275
276 ret = nfs_set_page_writeback(page);
277 BUG_ON(ret != 0);
278 BUG_ON(test_bit(PG_CLEAN, &req->wb_flags));
279
Fred Isamanf8512ad2008-03-19 11:24:39 -0400280 if (!nfs_pageio_add_request(pgio, req)) {
281 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400282 ret = pgio->pg_error;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400283 }
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400284out:
285 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500286}
287
Trond Myklebustf758c8852007-07-22 19:27:32 -0400288static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
289{
290 struct inode *inode = page->mapping->host;
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400291 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400292
293 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
294 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
295
296 nfs_pageio_cond_complete(pgio, page->index);
Wu Fengguang1b430be2010-10-26 14:21:26 -0700297 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400298 if (ret == -EAGAIN) {
299 redirty_page_for_writepage(wbc, page);
300 ret = 0;
301 }
302 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400303}
304
Trond Myklebuste261f512006-12-05 00:35:41 -0500305/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * Write an mmapped page to the server.
307 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500308static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400310 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500311 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Trond Myklebustf758c8852007-07-22 19:27:32 -0400313 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc));
314 err = nfs_do_writepage(page, wbc, &pgio);
315 nfs_pageio_complete(&pgio);
316 if (err < 0)
317 return err;
318 if (pgio.pg_error < 0)
319 return pgio.pg_error;
320 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500321}
322
323int nfs_writepage(struct page *page, struct writeback_control *wbc)
324{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400325 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500326
Trond Myklebustf758c8852007-07-22 19:27:32 -0400327 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400329 return ret;
330}
331
332static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
333{
334 int ret;
335
336 ret = nfs_do_writepage(page, wbc, data);
337 unlock_page(page);
338 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
342{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400344 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400345 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 int err;
347
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400348 /* Stop dirtying of new pages while we sync */
349 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
350 nfs_wait_bit_killable, TASK_KILLABLE);
351 if (err)
352 goto out_err;
353
Chuck Lever91d5b472006-03-20 13:44:14 -0500354 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
355
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400356 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc));
Trond Myklebustf758c8852007-07-22 19:27:32 -0400357 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400358 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400359
360 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
361 smp_mb__after_clear_bit();
362 wake_up_bit(bitlock, NFS_INO_FLUSHING);
363
Trond Myklebustf758c8852007-07-22 19:27:32 -0400364 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400365 goto out_err;
366 err = pgio.pg_error;
367 if (err < 0)
368 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400369 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400370out_err:
371 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
374/*
375 * Insert a write request into an inode
376 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500377static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378{
379 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400380
381 /* Lock the request! */
382 nfs_lock_request_dontget(req);
383
384 spin_lock(&inode->i_lock);
Trond Myklebust4d65c522011-03-25 14:15:11 -0400385 if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700386 inode->i_version++;
Trond Myklebust2df485a2010-12-07 22:39:17 -0500387 set_bit(PG_MAPPED, &req->wb_flags);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500388 SetPagePrivate(req->wb_page);
Trond Myklebust277459d2006-12-05 00:35:35 -0500389 set_page_private(req->wb_page, (unsigned long)req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400391 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400392 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393}
394
395/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800396 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 */
398static void nfs_inode_remove_request(struct nfs_page *req)
399{
Al Viro3d4ff432011-06-22 18:40:12 -0400400 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 struct nfs_inode *nfsi = NFS_I(inode);
402
403 BUG_ON (!NFS_WBACK_BUSY(req));
404
Trond Myklebust587142f2007-07-02 09:57:54 -0400405 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500406 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500407 ClearPagePrivate(req->wb_page);
Trond Myklebust2df485a2010-12-07 22:39:17 -0500408 clear_bit(PG_MAPPED, &req->wb_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 nfsi->npages--;
Trond Myklebust4d65c522011-03-25 14:15:11 -0400410 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 nfs_release_request(req);
412}
413
Trond Myklebust61822ab2006-12-05 00:35:42 -0500414static void
Fred6d884e82008-03-19 11:24:38 -0400415nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500416{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500417 __set_page_dirty_nobuffers(req->wb_page);
418}
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400421/**
422 * nfs_request_add_commit_list - add request to a commit list
423 * @req: pointer to a struct nfs_page
424 * @head: commit list head
425 *
426 * This sets the PG_CLEAN bit, updates the inode global count of
427 * number of outstanding requests requiring a commit as well as
428 * the MM page stats.
429 *
430 * The caller must _not_ hold the inode->i_lock, but must be
431 * holding the nfs_page lock.
432 */
433void
434nfs_request_add_commit_list(struct nfs_page *req, struct list_head *head)
435{
436 struct inode *inode = req->wb_context->dentry->d_inode;
437
438 set_bit(PG_CLEAN, &(req)->wb_flags);
439 spin_lock(&inode->i_lock);
440 nfs_list_add_request(req, head);
441 NFS_I(inode)->ncommit++;
442 spin_unlock(&inode->i_lock);
443 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
444 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
445 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
446}
447EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
448
449/**
450 * nfs_request_remove_commit_list - Remove request from a commit list
451 * @req: pointer to a nfs_page
452 *
453 * This clears the PG_CLEAN bit, and updates the inode global count of
454 * number of outstanding requests requiring a commit
455 * It does not update the MM page stats.
456 *
457 * The caller _must_ hold the inode->i_lock and the nfs_page lock.
458 */
459void
460nfs_request_remove_commit_list(struct nfs_page *req)
461{
462 struct inode *inode = req->wb_context->dentry->d_inode;
463
464 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
465 return;
466 nfs_list_remove_request(req);
467 NFS_I(inode)->ncommit--;
468}
469EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
470
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472/*
473 * Add a request to the inode's commit list.
474 */
475static void
Fred Isamana861a1e2011-03-23 13:27:51 +0000476nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Al Viro3d4ff432011-06-22 18:40:12 -0400478 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Trond Myklebust8dd37752012-03-15 17:16:40 -0400480 if (pnfs_mark_request_commit(req, lseg))
481 return;
482 nfs_request_add_commit_list(req, &NFS_I(inode)->commit_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400484
Fred Isamand6d6dc72012-03-08 17:29:35 -0500485static void
486nfs_clear_page_commit(struct page *page)
487{
488 dec_zone_page_state(page, NR_UNSTABLE_NFS);
489 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
490}
491
Trond Myklebust8dd37752012-03-15 17:16:40 -0400492static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400493nfs_clear_request_commit(struct nfs_page *req)
494{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400495 if (test_bit(PG_CLEAN, &req->wb_flags)) {
496 struct inode *inode = req->wb_context->dentry->d_inode;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400497
Trond Myklebust8dd37752012-03-15 17:16:40 -0400498 if (!pnfs_clear_request_commit(req)) {
499 spin_lock(&inode->i_lock);
500 nfs_request_remove_commit_list(req);
501 spin_unlock(&inode->i_lock);
502 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500503 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400504 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400505}
506
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400507static inline
508int nfs_write_need_commit(struct nfs_write_data *data)
509{
Fred Isaman465d5242011-03-23 13:27:44 +0000510 if (data->verf.committed == NFS_DATA_SYNC)
511 return data->lseg == NULL;
512 else
513 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400514}
515
516static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000517int nfs_reschedule_unstable_write(struct nfs_page *req,
518 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400519{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400520 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Fred Isamana861a1e2011-03-23 13:27:51 +0000521 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400522 return 1;
523 }
524 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
Fred6d884e82008-03-19 11:24:38 -0400525 nfs_mark_request_dirty(req);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400526 return 1;
527 }
528 return 0;
529}
530#else
Trond Myklebust8dd37752012-03-15 17:16:40 -0400531static void
Fred Isamana861a1e2011-03-23 13:27:51 +0000532nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400533{
534}
535
Trond Myklebust8dd37752012-03-15 17:16:40 -0400536static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400537nfs_clear_request_commit(struct nfs_page *req)
538{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400539}
540
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400541static inline
542int nfs_write_need_commit(struct nfs_write_data *data)
543{
544 return 0;
545}
546
547static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000548int nfs_reschedule_unstable_write(struct nfs_page *req,
549 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400550{
551 return 0;
552}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553#endif
554
Trond Myklebust47c62562009-03-16 08:13:41 -0400555#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400556static int
557nfs_need_commit(struct nfs_inode *nfsi)
558{
Fred Isamand6d6dc72012-03-08 17:29:35 -0500559 return nfsi->ncommit > 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400560}
561
Fred Isamand6d6dc72012-03-08 17:29:35 -0500562/* i_lock held by caller */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400563static int
Trond Myklebust3b3be882012-03-17 11:59:30 -0400564nfs_scan_commit_list(struct list_head *src, struct list_head *dst, int max,
565 spinlock_t *lock)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500566{
567 struct nfs_page *req, *tmp;
568 int ret = 0;
569
570 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400571 if (!nfs_lock_request(req))
572 continue;
Trond Myklebust3b3be882012-03-17 11:59:30 -0400573 if (cond_resched_lock(lock))
574 list_safe_reset_next(req, tmp, wb_list);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400575 nfs_request_remove_commit_list(req);
576 nfs_list_add_request(req, dst);
577 ret++;
578 if (ret == max)
579 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500580 }
581 return ret;
582}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584/*
585 * nfs_scan_commit - Scan an inode for commit requests
586 * @inode: NFS inode to scan
587 * @dst: destination list
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 *
589 * Moves requests from the inode's 'commit' request list.
590 * The requests are *not* checked to ensure that they form a contiguous set.
591 */
592static int
Fred Isamand6d6dc72012-03-08 17:29:35 -0500593nfs_scan_commit(struct inode *inode, struct list_head *dst)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 struct nfs_inode *nfsi = NFS_I(inode);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500596 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400597
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000598 spin_lock(&inode->i_lock);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500599 if (nfsi->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400600 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500601
Trond Myklebust3b3be882012-03-17 11:59:30 -0400602 ret = nfs_scan_commit_list(&nfsi->commit_list, dst, max,
603 &inode->i_lock);
604 ret += pnfs_scan_commit_lists(inode, max - ret,
605 &inode->i_lock);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500606 }
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000607 spin_unlock(&inode->i_lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800608 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500610
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500611#else
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400612static inline int nfs_need_commit(struct nfs_inode *nfsi)
613{
614 return 0;
615}
616
Fred Isamand6d6dc72012-03-08 17:29:35 -0500617static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500618{
619 return 0;
620}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621#endif
622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400624 * Search for an existing write request, and attempt to update
625 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400627 * If the attempt fails, then the existing request is flushed out
628 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400630static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
631 struct page *page,
632 unsigned int offset,
633 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400635 struct nfs_page *req;
636 unsigned int rqend;
637 unsigned int end;
638 int error;
639
640 if (!PagePrivate(page))
641 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400644 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 for (;;) {
Trond Myklebust277459d2006-12-05 00:35:35 -0500647 req = nfs_page_find_request_locked(page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400648 if (req == NULL)
649 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500650
Trond Myklebuste7d39062008-06-13 12:12:32 -0400651 rqend = req->wb_offset + req->wb_bytes;
652 /*
653 * Tell the caller to flush out the request if
654 * the offsets are non-contiguous.
655 * Note: nfs_flush_incompatible() will already
656 * have flushed out requests having wrong owners.
657 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400658 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400659 || end < req->wb_offset)
660 goto out_flushme;
661
Fred Isaman9994b622012-03-08 17:29:34 -0500662 if (nfs_lock_request_dontget(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Trond Myklebuste7d39062008-06-13 12:12:32 -0400665 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400666 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400667 error = nfs_wait_on_request(req);
668 nfs_release_request(req);
669 if (error != 0)
670 goto out_err;
671 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673
674 /* Okay, the request matches. Update the region */
675 if (offset < req->wb_offset) {
676 req->wb_offset = offset;
677 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 if (end > rqend)
680 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400681 else
682 req->wb_bytes = rqend - req->wb_offset;
683out_unlock:
684 spin_unlock(&inode->i_lock);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400685 nfs_clear_request_commit(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400686 return req;
687out_flushme:
688 spin_unlock(&inode->i_lock);
689 nfs_release_request(req);
690 error = nfs_wb_page(inode, page);
691out_err:
692 return ERR_PTR(error);
693}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Trond Myklebuste7d39062008-06-13 12:12:32 -0400695/*
696 * Try to update an existing write request, or create one if there is none.
697 *
698 * Note: Should always be called with the Page Lock held to prevent races
699 * if we have to add a new request. Also assumes that the caller has
700 * already called nfs_flush_incompatible() if necessary.
701 */
702static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
703 struct page *page, unsigned int offset, unsigned int bytes)
704{
705 struct inode *inode = page->mapping->host;
706 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400707
708 req = nfs_try_to_update_request(inode, page, offset, bytes);
709 if (req != NULL)
710 goto out;
711 req = nfs_create_request(ctx, inode, page, offset, bytes);
712 if (IS_ERR(req))
713 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500714 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400715out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400716 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Trond Myklebuste7d39062008-06-13 12:12:32 -0400719static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
720 unsigned int offset, unsigned int count)
721{
722 struct nfs_page *req;
723
724 req = nfs_setup_write_request(ctx, page, offset, count);
725 if (IS_ERR(req))
726 return PTR_ERR(req);
727 /* Update file length */
728 nfs_grow_file(page, offset, count);
729 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400730 nfs_mark_request_dirty(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500731 nfs_unlock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400732 return 0;
733}
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735int nfs_flush_incompatible(struct file *file, struct page *page)
736{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400737 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500739 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 /*
741 * Look for a request corresponding to this page. If there
742 * is one, and it belongs to another file, we flush it out
743 * before we try to copy anything into the page. Do this
744 * due to the lack of an ACCESS-type call in NFSv2.
745 * Also do the same if we find a request from an existing
746 * dropped page.
747 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500748 do {
749 req = nfs_page_find_request(page);
750 if (req == NULL)
751 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400752 do_flush = req->wb_page != page || req->wb_context != ctx ||
753 req->wb_lock_context->lockowner != current->files ||
754 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500756 if (!do_flush)
757 return 0;
758 status = nfs_wb_page(page->mapping->host, page);
759 } while (status == 0);
760 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
763/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500764 * If the page cache is marked as unsafe or invalid, then we can't rely on
765 * the PageUptodate() flag. In this case, we will need to turn off
766 * write optimisations that depend on the page contents being correct.
767 */
768static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
769{
770 return PageUptodate(page) &&
771 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
772}
773
774/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * Update and possibly write a cached page of an NFS file.
776 *
777 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
778 * things with a page scheduled for an RPC call (e.g. invalidate it).
779 */
780int nfs_updatepage(struct file *file, struct page *page,
781 unsigned int offset, unsigned int count)
782{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400783 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 int status = 0;
786
Chuck Lever91d5b472006-03-20 13:44:14 -0500787 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
788
Chuck Lever48186c72008-06-11 17:56:05 -0400789 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800790 file->f_path.dentry->d_parent->d_name.name,
791 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400792 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500795 * is up to date, it may be more efficient to extend the write
796 * to cover the entire page in order to avoid fragmentation
797 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500799 if (nfs_write_pageuptodate(page, inode) &&
800 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100801 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500802 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
805
Trond Myklebuste21195a2006-12-05 00:35:39 -0500806 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400807 if (status < 0)
808 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700809 else
810 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
Chuck Lever48186c72008-06-11 17:56:05 -0400812 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return status;
815}
816
Fred Isamana861a1e2011-03-23 13:27:51 +0000817static void nfs_writepage_release(struct nfs_page *req,
818 struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400820 struct page *page = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Fred Isamana861a1e2011-03-23 13:27:51 +0000822 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400823 nfs_inode_remove_request(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500824 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400825 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826}
827
Trond Myklebust3ff75762007-07-14 15:40:00 -0400828static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829{
830 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
831 case FLUSH_HIGHPRI:
832 return RPC_PRIORITY_HIGH;
833 case FLUSH_LOWPRI:
834 return RPC_PRIORITY_LOW;
835 }
836 return RPC_PRIORITY_NORMAL;
837}
838
Fred Isamana69aef12011-03-03 15:13:47 +0000839int nfs_initiate_write(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000840 struct rpc_clnt *clnt,
841 const struct rpc_call_ops *call_ops,
842 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
Andy Adamsond138d5d2011-03-03 15:13:41 +0000844 struct inode *inode = data->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400845 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400846 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400847 struct rpc_message msg = {
848 .rpc_argp = &data->args,
849 .rpc_resp = &data->res,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000850 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400851 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400852 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +0000853 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -0400854 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400855 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400856 .callback_ops = call_ops,
857 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500858 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -0400859 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400860 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400861 };
Trond Myklebust2c61be02010-04-09 19:54:50 -0400862 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Andy Adamsond138d5d2011-03-03 15:13:41 +0000864 /* Set up the initial task struct. */
865 NFS_PROTO(inode)->write_setup(data, &msg);
866
867 dprintk("NFS: %5u initiated write call "
868 "(req %s/%lld, %u bytes @ offset %llu)\n",
869 data->task.tk_pid,
870 inode->i_sb->s_id,
871 (long long)NFS_FILEID(inode),
872 data->args.count,
873 (unsigned long long)data->args.offset);
874
875 task = rpc_run_task(&task_setup_data);
876 if (IS_ERR(task)) {
877 ret = PTR_ERR(task);
878 goto out;
879 }
880 if (how & FLUSH_SYNC) {
881 ret = rpc_wait_for_completion_task(task);
882 if (ret == 0)
883 ret = task->tk_status;
884 }
885 rpc_put_task(task);
886out:
887 return ret;
888}
Fred Isamana69aef12011-03-03 15:13:47 +0000889EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +0000890
891/*
892 * Set up the argument/result storage required for the RPC call.
893 */
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400894static void nfs_write_rpcsetup(struct nfs_page *req,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000895 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000896 unsigned int count, unsigned int offset,
897 int how)
898{
Al Viro3d4ff432011-06-22 18:40:12 -0400899 struct inode *inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 /* Set up the RPC argument and reply structs
902 * NB: take care not to mess about with data->commit et al. */
903
904 data->req = req;
Al Viro3d4ff432011-06-22 18:40:12 -0400905 data->inode = inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000906 data->cred = req->wb_context->cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 data->args.fh = NFS_FH(inode);
909 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -0400910 /* pnfs_set_layoutcommit needs this */
911 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 data->args.pgbase = req->wb_pgbase + offset;
913 data->args.pages = data->pagevec;
914 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500915 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400916 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400917 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -0400918 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
919 case 0:
920 break;
921 case FLUSH_COND_STABLE:
922 if (nfs_need_commit(NFS_I(inode)))
923 break;
924 default:
925 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
928 data->res.fattr = &data->fattr;
929 data->res.count = count;
930 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400931 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400932}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400934static int nfs_do_write(struct nfs_write_data *data,
935 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400936 int how)
937{
Stephen Rothwell5f00bcb2011-07-25 13:59:46 -0400938 struct inode *inode = data->args.context->dentry->d_inode;
Andy Adamson0382b742011-03-03 15:13:45 +0000939
Andy Adamsond138d5d2011-03-03 15:13:41 +0000940 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941}
942
Trond Myklebust275acaa2011-07-12 13:42:02 -0400943static int nfs_do_multiple_writes(struct list_head *head,
944 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -0400945 int how)
946{
947 struct nfs_write_data *data;
948 int ret = 0;
949
950 while (!list_empty(head)) {
951 int ret2;
952
953 data = list_entry(head->next, struct nfs_write_data, list);
954 list_del_init(&data->list);
955
Trond Myklebustdce81292011-07-13 15:59:19 -0400956 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400957 if (ret == 0)
958 ret = ret2;
959 }
960 return ret;
961}
962
Fred6d884e82008-03-19 11:24:38 -0400963/* If a nfs_flush_* function fails, it should remove reqs from @head and
964 * call this on each, which will prepare them to be retried on next
965 * writeback using standard nfs.
966 */
967static void nfs_redirty_request(struct nfs_page *req)
968{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400969 struct page *page = req->wb_page;
970
Fred6d884e82008-03-19 11:24:38 -0400971 nfs_mark_request_dirty(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500972 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400973 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400974}
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976/*
977 * Generate multiple small requests to write out a single
978 * contiguous dirty area on one page.
979 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400980static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Fred Isamanc76069b2011-03-03 15:13:48 +0000982 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 struct page *page = req->wb_page;
984 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -0400985 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700986 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400988 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
990 nfs_list_remove_request(req);
991
Trond Myklebustb31268a2011-03-21 17:02:00 -0400992 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
993 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
994 desc->pg_count > wsize))
995 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
996
997
Trond Myklebust275acaa2011-07-12 13:42:02 -0400998 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000999 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001000 do {
1001 size_t len = min(nbytes, wsize);
1002
Trond Myklebust8d5658c2007-04-10 09:26:35 -04001003 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (!data)
1005 goto out_bad;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001006 data->pagevec[0] = page;
Trond Myklebustf13c3622011-09-12 11:47:53 -04001007 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001008 list_add(&data->list, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001010 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001011 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001012 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 atomic_set(&req->wb_complete, requests);
Trond Myklebust50828d72011-07-12 13:42:02 -04001014 desc->pg_rpc_callops = &nfs_write_partial_ops;
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001015 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017out_bad:
Trond Myklebust275acaa2011-07-12 13:42:02 -04001018 while (!list_empty(res)) {
1019 data = list_entry(res->next, struct nfs_write_data, list);
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001020 list_del(&data->list);
Fred Isaman0da2a4ac2011-01-19 14:18:50 -05001021 nfs_writedata_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
Trond Myklebust61822ab2006-12-05 00:35:42 -05001023 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 return -ENOMEM;
1025}
1026
1027/*
1028 * Create an RPC task for the given write request and kick it.
1029 * The page must have been locked by the caller.
1030 *
1031 * It may happen that the page we're passed is not marked dirty.
1032 * This is the case if nfs_updatepage detects a conflicting request
1033 * that has been written but not committed.
1034 */
Trond Myklebust275acaa2011-07-12 13:42:02 -04001035static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
1037 struct nfs_page *req;
1038 struct page **pages;
1039 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +00001040 struct list_head *head = &desc->pg_list;
Peng Tao3b609182011-07-15 03:33:42 -04001041 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Fred Isamanc76069b2011-03-03 15:13:48 +00001043 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
1044 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +00001045 if (!data) {
1046 while (!list_empty(head)) {
1047 req = nfs_list_entry(head->next);
1048 nfs_list_remove_request(req);
1049 nfs_redirty_request(req);
1050 }
1051 ret = -ENOMEM;
1052 goto out;
1053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 while (!list_empty(head)) {
1056 req = nfs_list_entry(head->next);
1057 nfs_list_remove_request(req);
1058 nfs_list_add_request(req, &data->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
1061 req = nfs_list_entry(data->pages.next);
1062
Trond Myklebustb31268a2011-03-21 17:02:00 -04001063 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1064 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1065 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1066
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 /* Set up the argument struct */
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001068 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001069 list_add(&data->list, res);
Trond Myklebust50828d72011-07-12 13:42:02 -04001070 desc->pg_rpc_callops = &nfs_write_full_ops;
Fred Isaman44b83792011-03-03 15:13:44 +00001071out:
Fred Isaman44b83792011-03-03 15:13:44 +00001072 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
Trond Myklebustdce81292011-07-13 15:59:19 -04001075int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1076{
1077 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1078 return nfs_flush_multi(desc, head);
1079 return nfs_flush_one(desc, head);
1080}
1081
1082static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001083{
Trond Myklebust275acaa2011-07-12 13:42:02 -04001084 LIST_HEAD(head);
1085 int ret;
1086
Trond Myklebustdce81292011-07-13 15:59:19 -04001087 ret = nfs_generic_flush(desc, &head);
Trond Myklebust50828d72011-07-12 13:42:02 -04001088 if (ret == 0)
1089 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
Trond Myklebustdce81292011-07-13 15:59:19 -04001090 desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001091 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001092}
Trond Myklebust1751c362011-06-10 13:30:23 -04001093
1094static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1095 .pg_test = nfs_generic_pg_test,
1096 .pg_doio = nfs_generic_pg_writepages,
1097};
1098
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001099void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
Trond Myklebust1751c362011-06-10 13:30:23 -04001100 struct inode *inode, int ioflags)
1101{
1102 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1103 NFS_SERVER(inode)->wsize, ioflags);
1104}
1105
Trond Myklebustdce81292011-07-13 15:59:19 -04001106void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1107{
1108 pgio->pg_ops = &nfs_pageio_write_ops;
1109 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1110}
Trond Myklebust1f945352011-07-13 15:59:57 -04001111EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001112
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001113static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1114 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115{
Trond Myklebust1751c362011-06-10 13:30:23 -04001116 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1117 nfs_pageio_init_write_mds(pgio, inode, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118}
1119
1120/*
1121 * Handle a write reply that flushed part of a page.
1122 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001123static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001125 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Chuck Lever48186c72008-06-11 17:56:05 -04001127 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1128 task->tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001129 data->req->wb_context->dentry->d_inode->i_sb->s_id,
Chuck Lever48186c72008-06-11 17:56:05 -04001130 (long long)
Al Viro3d4ff432011-06-22 18:40:12 -04001131 NFS_FILEID(data->req->wb_context->dentry->d_inode),
Chuck Lever48186c72008-06-11 17:56:05 -04001132 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001134 nfs_writeback_done(task, data);
1135}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001136
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001137static void nfs_writeback_release_partial(void *calldata)
1138{
1139 struct nfs_write_data *data = calldata;
1140 struct nfs_page *req = data->req;
1141 struct page *page = req->wb_page;
1142 int status = data->task.tk_status;
1143
1144 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001145 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001146 nfs_context_set_write_error(req->wb_context, status);
1147 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001148 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 }
1150
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001151 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001152 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001153
Trond Myklebust587142f2007-07-02 09:57:54 -04001154 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001155 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1156 /* Do nothing we need to resend the writes */
1157 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1158 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1159 dprintk(" defer commit\n");
1160 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1161 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1162 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1163 dprintk(" server reboot detected\n");
1164 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001165 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001166 } else
1167 dprintk(" OK\n");
1168
1169out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001171 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001172 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
Andy Adamsondef6ed72009-04-01 09:22:26 -04001175void nfs_write_prepare(struct rpc_task *task, void *calldata)
1176{
1177 struct nfs_write_data *data = calldata;
Bryan Schumakerc6cb80d2012-03-19 14:54:39 -04001178 NFS_PROTO(data->inode)->write_rpc_prepare(task, data);
Andy Adamsondef6ed72009-04-01 09:22:26 -04001179}
Andy Adamsondef6ed72009-04-01 09:22:26 -04001180
Trond Myklebust788e7a82006-03-20 13:44:27 -05001181static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001182 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001183 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001184 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001185};
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187/*
1188 * Handle a write reply that flushes a whole page.
1189 *
1190 * FIXME: There is an inherent race with invalidate_inode_pages and
1191 * writebacks since the page->count is kept > 1 for as long
1192 * as the page has a write request pending.
1193 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001194static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001196 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001198 nfs_writeback_done(task, data);
1199}
1200
1201static void nfs_writeback_release_full(void *calldata)
1202{
1203 struct nfs_write_data *data = calldata;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001204 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 /* Update attributes as result of writeback. */
1207 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001208 struct nfs_page *req = nfs_list_entry(data->pages.next);
1209 struct page *page = req->wb_page;
1210
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
Chuck Lever48186c72008-06-11 17:56:05 -04001213 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1214 data->task.tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001215 req->wb_context->dentry->d_inode->i_sb->s_id,
1216 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 req->wb_bytes,
1218 (long long)req_offset(req));
1219
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001220 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001221 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001222 nfs_context_set_write_error(req->wb_context, status);
1223 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001224 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001227 if (nfs_write_need_commit(data)) {
1228 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001229 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001230 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 goto next;
1232 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001233 dprintk(" OK\n");
1234remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 next:
Fred Isaman9994b622012-03-08 17:29:34 -05001237 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001238 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001240 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
Trond Myklebust788e7a82006-03-20 13:44:27 -05001243static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001244 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001245 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001246 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001247};
1248
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250/*
1251 * This function is called when the WRITE call is complete.
1252 */
Fred Isaman13602892011-02-11 15:42:38 +00001253void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 struct nfs_writeargs *argp = &data->args;
1256 struct nfs_writeres *resp = &data->res;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001257 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Chuck Levera3f565b2007-01-31 12:14:01 -05001259 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 task->tk_pid, task->tk_status);
1261
Chuck Leverf551e442006-09-20 14:33:04 -04001262 /*
1263 * ->write_done will attempt to use post-op attributes to detect
1264 * conflicting writes by other clients. A strict interpretation
1265 * of close-to-open would allow us to continue caching even if
1266 * another writer had changed the file, but some applications
1267 * depend on tighter cache coherency when writing.
1268 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001269 status = NFS_PROTO(data->inode)->write_done(task, data);
1270 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001271 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001272 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1275 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1276 /* We tried a write call, but the server did not
1277 * commit data to stable storage even though we
1278 * requested it.
1279 * Note: There is a known bug in Tru64 < 5.0 in which
1280 * the server reports NFS_DATA_SYNC, but performs
1281 * NFS_FILE_SYNC. We therefore implement this checking
1282 * as a dprintk() in order to avoid filling syslog.
1283 */
1284 static unsigned long complain;
1285
Fred Isamana69aef12011-03-03 15:13:47 +00001286 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001288 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 " (committed = %d) != (stable = %d)\n",
Rakib Mullick2b72c9c2011-11-01 12:23:42 +06001290 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 resp->verf->committed, argp->stable);
1292 complain = jiffies + 300 * HZ;
1293 }
1294 }
1295#endif
1296 /* Is this a short write? */
1297 if (task->tk_status >= 0 && resp->count < argp->count) {
1298 static unsigned long complain;
1299
Chuck Lever91d5b472006-03-20 13:44:14 -05001300 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 /* Has the server at least made some progress? */
1303 if (resp->count != 0) {
1304 /* Was this an NFSv2 write or an NFSv3 stable write? */
1305 if (resp->verf->committed != NFS_UNSTABLE) {
1306 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001307 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 argp->offset += resp->count;
1309 argp->pgbase += resp->count;
1310 argp->count -= resp->count;
1311 } else {
1312 /* Resend as a stable write in order to avoid
1313 * headaches in the case of a server crash.
1314 */
1315 argp->stable = NFS_FILE_SYNC;
1316 }
Trond Myklebustd00c5d42011-10-19 12:17:29 -07001317 rpc_restart_call_prepare(task);
Fred Isaman13602892011-02-11 15:42:38 +00001318 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 }
1320 if (time_before(complain, jiffies)) {
1321 printk(KERN_WARNING
1322 "NFS: Server wrote zero bytes, expected %u.\n",
1323 argp->count);
1324 complain = jiffies + 300 * HZ;
1325 }
1326 /* Can't do anything about it except throw an error. */
1327 task->tk_status = -EIO;
1328 }
Fred Isaman13602892011-02-11 15:42:38 +00001329 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330}
1331
1332
1333#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001334static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1335{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001336 int ret;
1337
Trond Myklebust71d0a612010-04-22 15:35:57 -04001338 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1339 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001340 if (!may_wait)
1341 return 0;
1342 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1343 NFS_INO_COMMIT,
1344 nfs_wait_bit_killable,
1345 TASK_KILLABLE);
1346 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001347}
1348
Fred Isamane0c2b382011-03-23 13:27:53 +00001349void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001350{
1351 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1352 smp_mb__after_clear_bit();
1353 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1354}
Fred Isamane0c2b382011-03-23 13:27:53 +00001355EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001356
Fred Isamane0c2b382011-03-23 13:27:53 +00001357void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
Trond Myklebust383ba712008-02-19 20:04:20 -05001359 struct nfs_write_data *wdata = data;
1360
1361 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 nfs_commit_free(wdata);
1363}
Fred Isamane0c2b382011-03-23 13:27:53 +00001364EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Fred Isamane0c2b382011-03-23 13:27:53 +00001366int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001367 const struct rpc_call_ops *call_ops,
1368 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369{
Trond Myklebust07737692007-10-25 18:42:54 -04001370 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001371 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001372 struct rpc_message msg = {
1373 .rpc_argp = &data->args,
1374 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001375 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001376 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001377 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001378 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001379 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001380 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001381 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001382 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001383 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001384 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001385 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001386 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001387 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001388 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
Chuck Levera3f565b2007-01-31 12:14:01 -05001390 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001391
Trond Myklebust07737692007-10-25 18:42:54 -04001392 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001393 if (IS_ERR(task))
1394 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001395 if (how & FLUSH_SYNC)
1396 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001397 rpc_put_task(task);
1398 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
Fred Isamane0c2b382011-03-23 13:27:53 +00001400EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001403 * Set up the argument/result storage required for the RPC call.
1404 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001405void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001406 struct list_head *head,
1407 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001408{
1409 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001410 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001411
1412 /* Set up the RPC argument and reply structs
1413 * NB: take care not to mess about with data->commit et al. */
1414
1415 list_splice_init(head, &data->pages);
1416
1417 data->inode = inode;
1418 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001419 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001420 data->mds_ops = &nfs_commit_ops;
1421
1422 data->args.fh = NFS_FH(data->inode);
1423 /* Note: we always request a commit of the entire inode */
1424 data->args.offset = 0;
1425 data->args.count = 0;
1426 data->args.context = get_nfs_open_context(first->wb_context);
1427 data->res.count = 0;
1428 data->res.fattr = &data->fattr;
1429 data->res.verf = &data->verf;
1430 nfs_fattr_init(&data->fattr);
1431}
Fred Isamane0c2b382011-03-23 13:27:53 +00001432EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001433
Fred Isamane0c2b382011-03-23 13:27:53 +00001434void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001435 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001436{
1437 struct nfs_page *req;
1438
1439 while (!list_empty(page_list)) {
1440 req = nfs_list_entry(page_list->next);
1441 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001442 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001443 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1444 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1445 BDI_RECLAIMABLE);
Fred Isaman9994b622012-03-08 17:29:34 -05001446 nfs_unlock_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001447 }
1448}
Fred Isamane0c2b382011-03-23 13:27:53 +00001449EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001450
Fred Isaman9ace33c2011-03-23 13:27:45 +00001451/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 * Commit dirty pages
1453 */
1454static int
Chuck Lever40859d72005-11-30 18:09:02 -05001455nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
1457 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001459 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
1461 if (!data)
1462 goto out_bad;
1463
1464 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001465 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001466 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001468 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001469 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 return -ENOMEM;
1471}
1472
1473/*
1474 * COMMIT call returned
1475 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001476static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001478 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Chuck Levera3f565b2007-01-31 12:14:01 -05001480 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 task->tk_pid, task->tk_status);
1482
Trond Myklebust788e7a82006-03-20 13:44:27 -05001483 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001484 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001485}
1486
Fred Isamane0c2b382011-03-23 13:27:53 +00001487void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001488{
Fred Isaman5917ce82011-03-23 13:27:48 +00001489 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001490 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 while (!list_empty(&data->pages)) {
1493 req = nfs_list_entry(data->pages.next);
1494 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001495 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Chuck Lever48186c72008-06-11 17:56:05 -04001497 dprintk("NFS: commit (%s/%lld %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001498 req->wb_context->dentry->d_sb->s_id,
1499 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 req->wb_bytes,
1501 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001502 if (status < 0) {
1503 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001505 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 goto next;
1507 }
1508
1509 /* Okay, COMMIT succeeded, apparently. Check the verifier
1510 * returned by the server against all stored verfs. */
1511 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1512 /* We have a match */
1513 nfs_inode_remove_request(req);
1514 dprintk(" OK\n");
1515 goto next;
1516 }
1517 /* We have a mismatch. Write the page again */
1518 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001519 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 next:
Fred Isaman9994b622012-03-08 17:29:34 -05001521 nfs_unlock_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001523}
Fred Isamane0c2b382011-03-23 13:27:53 +00001524EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001525
1526static void nfs_commit_release(void *calldata)
1527{
1528 struct nfs_write_data *data = calldata;
1529
1530 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001531 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001532 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001534
1535static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001536 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001537 .rpc_call_done = nfs_commit_done,
1538 .rpc_release = nfs_commit_release,
1539};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Trond Myklebustb608b282010-07-30 15:31:54 -04001541int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001544 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001545 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Trond Myklebustb8413f92011-03-21 15:37:01 -04001547 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1548 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001549 goto out_mark_dirty;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001550 res = nfs_scan_commit(inode, &head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001552 int error;
1553
1554 error = pnfs_commit_list(inode, &head, how);
1555 if (error == PNFS_NOT_ATTEMPTED)
1556 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001557 if (error < 0)
1558 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001559 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001560 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001561 error = wait_on_bit(&NFS_I(inode)->flags,
1562 NFS_INO_COMMIT,
1563 nfs_wait_bit_killable,
1564 TASK_KILLABLE);
1565 if (error < 0)
1566 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001567 } else
1568 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001569 return res;
1570 /* Note: If we exit without ensuring that the commit is complete,
1571 * we must mark the inode as dirty. Otherwise, future calls to
1572 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1573 * that the data is on the disk.
1574 */
1575out_mark_dirty:
1576 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return res;
1578}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001579
1580static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1581{
Trond Myklebust420e3642010-02-19 17:00:02 -08001582 struct nfs_inode *nfsi = NFS_I(inode);
1583 int flags = FLUSH_SYNC;
1584 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001585
Jeff Layton3236c3e2011-10-11 09:49:21 -04001586 /* no commits means nothing needs to be done */
1587 if (!nfsi->ncommit)
1588 return ret;
1589
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001590 if (wbc->sync_mode == WB_SYNC_NONE) {
1591 /* Don't commit yet if this is a non-blocking flush and there
1592 * are a lot of outstanding writes for this mapping.
1593 */
1594 if (nfsi->ncommit <= (nfsi->npages >> 1))
1595 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001596
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001597 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001598 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001599 }
1600
Trond Myklebust420e3642010-02-19 17:00:02 -08001601 ret = nfs_commit_inode(inode, flags);
1602 if (ret >= 0) {
1603 if (wbc->sync_mode == WB_SYNC_NONE) {
1604 if (ret < wbc->nr_to_write)
1605 wbc->nr_to_write -= ret;
1606 else
1607 wbc->nr_to_write = 0;
1608 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001609 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001610 }
1611out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001612 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1613 return ret;
1614}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001615#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001616static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1617{
1618 return 0;
1619}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620#endif
1621
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001622int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1623{
Andy Adamson863a3c62011-03-23 13:27:54 +00001624 int ret;
1625
1626 ret = nfs_commit_unstable_pages(inode, wbc);
1627 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001628 int status;
1629 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001630
Wu Fengguang846d5a02011-05-05 21:10:38 -06001631 if (wbc->sync_mode == WB_SYNC_NONE)
Andy Adamsonef311532011-03-12 02:58:10 -05001632 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001633
1634 status = pnfs_layoutcommit_inode(inode, sync);
1635 if (status < 0)
1636 return status;
1637 }
1638 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001639}
1640
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001641/*
1642 * flush the inode to disk.
1643 */
1644int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001645{
1646 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001647 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001648 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001649 .range_start = 0,
1650 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001651 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001652
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001653 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001654}
1655
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001656int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1657{
1658 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001659 int ret = 0;
1660
1661 BUG_ON(!PageLocked(page));
1662 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001663 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001664 req = nfs_page_find_request(page);
1665 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001666 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001667 if (nfs_lock_request_dontget(req)) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001668 nfs_clear_request_commit(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001669 nfs_inode_remove_request(req);
1670 /*
1671 * In case nfs_inode_remove_request has marked the
1672 * page as being dirty
1673 */
1674 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1675 nfs_unlock_request(req);
1676 break;
1677 }
1678 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001679 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001680 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001681 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001682 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001683 return ret;
1684}
1685
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001686/*
1687 * Write back all requests on one page - we do this before reading it.
1688 */
1689int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001690{
1691 loff_t range_start = page_offset(page);
1692 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001693 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001694 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001695 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001696 .range_start = range_start,
1697 .range_end = range_end,
1698 };
1699 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001700
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001701 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001702 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001703 if (clear_page_dirty_for_io(page)) {
1704 ret = nfs_writepage_locked(page, &wbc);
1705 if (ret < 0)
1706 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001707 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001708 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001709 if (!PagePrivate(page))
1710 break;
1711 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001712 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001713 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001714 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001715 return 0;
1716out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001717 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001718}
1719
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001720#ifdef CONFIG_MIGRATION
1721int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001722 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001723{
Jeff Layton2da95652011-10-12 10:57:42 -04001724 /*
1725 * If PagePrivate is set, then the page is currently associated with
1726 * an in-progress read or write request. Don't try to migrate it.
1727 *
1728 * FIXME: we could do this in principle, but we'll need a way to ensure
1729 * that we can safely release the inode reference while holding
1730 * the page lock.
1731 */
1732 if (PagePrivate(page))
1733 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001734
Trond Myklebust7549ad52010-02-08 09:32:34 -05001735 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001736
Mel Gormana6bc32b2012-01-12 17:19:43 -08001737 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001738}
1739#endif
1740
David Howellsf7b422b2006-06-09 09:34:33 -04001741int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
1743 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1744 sizeof(struct nfs_write_data),
1745 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001746 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 if (nfs_wdata_cachep == NULL)
1748 return -ENOMEM;
1749
Matthew Dobson93d23412006-03-26 01:37:50 -08001750 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1751 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 if (nfs_wdata_mempool == NULL)
1753 return -ENOMEM;
1754
Matthew Dobson93d23412006-03-26 01:37:50 -08001755 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1756 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (nfs_commit_mempool == NULL)
1758 return -ENOMEM;
1759
Peter Zijlstra89a09142007-03-16 13:38:26 -08001760 /*
1761 * NFS congestion size, scale with available memory.
1762 *
1763 * 64MB: 8192k
1764 * 128MB: 11585k
1765 * 256MB: 16384k
1766 * 512MB: 23170k
1767 * 1GB: 32768k
1768 * 2GB: 46340k
1769 * 4GB: 65536k
1770 * 8GB: 92681k
1771 * 16GB: 131072k
1772 *
1773 * This allows larger machines to have larger/more transfers.
1774 * Limit the default to 256M
1775 */
1776 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1777 if (nfs_congestion_kb > 256*1024)
1778 nfs_congestion_kb = 256*1024;
1779
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 return 0;
1781}
1782
David Brownell266bee82006-06-27 12:59:15 -07001783void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784{
1785 mempool_destroy(nfs_commit_mempool);
1786 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001787 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788}
1789