blob: c07462320f6b5c41c09ba2ff054e75048951691f [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);
Fred Isamanca138f32012-04-05 15:26:36 -0400685 if (req)
686 nfs_clear_request_commit(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400687 return req;
688out_flushme:
689 spin_unlock(&inode->i_lock);
690 nfs_release_request(req);
691 error = nfs_wb_page(inode, page);
692out_err:
693 return ERR_PTR(error);
694}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Trond Myklebuste7d39062008-06-13 12:12:32 -0400696/*
697 * Try to update an existing write request, or create one if there is none.
698 *
699 * Note: Should always be called with the Page Lock held to prevent races
700 * if we have to add a new request. Also assumes that the caller has
701 * already called nfs_flush_incompatible() if necessary.
702 */
703static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
704 struct page *page, unsigned int offset, unsigned int bytes)
705{
706 struct inode *inode = page->mapping->host;
707 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400708
709 req = nfs_try_to_update_request(inode, page, offset, bytes);
710 if (req != NULL)
711 goto out;
712 req = nfs_create_request(ctx, inode, page, offset, bytes);
713 if (IS_ERR(req))
714 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500715 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400716out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400717 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
Trond Myklebuste7d39062008-06-13 12:12:32 -0400720static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
721 unsigned int offset, unsigned int count)
722{
723 struct nfs_page *req;
724
725 req = nfs_setup_write_request(ctx, page, offset, count);
726 if (IS_ERR(req))
727 return PTR_ERR(req);
728 /* Update file length */
729 nfs_grow_file(page, offset, count);
730 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400731 nfs_mark_request_dirty(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500732 nfs_unlock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400733 return 0;
734}
735
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736int nfs_flush_incompatible(struct file *file, struct page *page)
737{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400738 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500740 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 /*
742 * Look for a request corresponding to this page. If there
743 * is one, and it belongs to another file, we flush it out
744 * before we try to copy anything into the page. Do this
745 * due to the lack of an ACCESS-type call in NFSv2.
746 * Also do the same if we find a request from an existing
747 * dropped page.
748 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500749 do {
750 req = nfs_page_find_request(page);
751 if (req == NULL)
752 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400753 do_flush = req->wb_page != page || req->wb_context != ctx ||
754 req->wb_lock_context->lockowner != current->files ||
755 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500757 if (!do_flush)
758 return 0;
759 status = nfs_wb_page(page->mapping->host, page);
760 } while (status == 0);
761 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
764/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500765 * If the page cache is marked as unsafe or invalid, then we can't rely on
766 * the PageUptodate() flag. In this case, we will need to turn off
767 * write optimisations that depend on the page contents being correct.
768 */
769static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
770{
771 return PageUptodate(page) &&
772 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
773}
774
775/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 * Update and possibly write a cached page of an NFS file.
777 *
778 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
779 * things with a page scheduled for an RPC call (e.g. invalidate it).
780 */
781int nfs_updatepage(struct file *file, struct page *page,
782 unsigned int offset, unsigned int count)
783{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400784 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 int status = 0;
787
Chuck Lever91d5b472006-03-20 13:44:14 -0500788 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
789
Chuck Lever48186c72008-06-11 17:56:05 -0400790 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800791 file->f_path.dentry->d_parent->d_name.name,
792 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400793 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500796 * is up to date, it may be more efficient to extend the write
797 * to cover the entire page in order to avoid fragmentation
798 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500800 if (nfs_write_pageuptodate(page, inode) &&
801 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100802 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500803 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 }
806
Trond Myklebuste21195a2006-12-05 00:35:39 -0500807 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400808 if (status < 0)
809 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700810 else
811 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Chuck Lever48186c72008-06-11 17:56:05 -0400813 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 return status;
816}
817
Fred Isamana861a1e2011-03-23 13:27:51 +0000818static void nfs_writepage_release(struct nfs_page *req,
819 struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400821 struct page *page = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
Fred Isamana861a1e2011-03-23 13:27:51 +0000823 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400824 nfs_inode_remove_request(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500825 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400826 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
Trond Myklebust3ff75762007-07-14 15:40:00 -0400829static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830{
831 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
832 case FLUSH_HIGHPRI:
833 return RPC_PRIORITY_HIGH;
834 case FLUSH_LOWPRI:
835 return RPC_PRIORITY_LOW;
836 }
837 return RPC_PRIORITY_NORMAL;
838}
839
Fred Isamana69aef12011-03-03 15:13:47 +0000840int nfs_initiate_write(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000841 struct rpc_clnt *clnt,
842 const struct rpc_call_ops *call_ops,
843 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
Andy Adamsond138d5d2011-03-03 15:13:41 +0000845 struct inode *inode = data->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400846 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400847 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400848 struct rpc_message msg = {
849 .rpc_argp = &data->args,
850 .rpc_resp = &data->res,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000851 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400852 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400853 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +0000854 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -0400855 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400856 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400857 .callback_ops = call_ops,
858 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500859 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -0400860 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400861 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400862 };
Trond Myklebust2c61be02010-04-09 19:54:50 -0400863 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Andy Adamsond138d5d2011-03-03 15:13:41 +0000865 /* Set up the initial task struct. */
866 NFS_PROTO(inode)->write_setup(data, &msg);
867
868 dprintk("NFS: %5u initiated write call "
869 "(req %s/%lld, %u bytes @ offset %llu)\n",
870 data->task.tk_pid,
871 inode->i_sb->s_id,
872 (long long)NFS_FILEID(inode),
873 data->args.count,
874 (unsigned long long)data->args.offset);
875
876 task = rpc_run_task(&task_setup_data);
877 if (IS_ERR(task)) {
878 ret = PTR_ERR(task);
879 goto out;
880 }
881 if (how & FLUSH_SYNC) {
882 ret = rpc_wait_for_completion_task(task);
883 if (ret == 0)
884 ret = task->tk_status;
885 }
886 rpc_put_task(task);
887out:
888 return ret;
889}
Fred Isamana69aef12011-03-03 15:13:47 +0000890EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +0000891
892/*
893 * Set up the argument/result storage required for the RPC call.
894 */
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400895static void nfs_write_rpcsetup(struct nfs_page *req,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000896 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000897 unsigned int count, unsigned int offset,
898 int how)
899{
Al Viro3d4ff432011-06-22 18:40:12 -0400900 struct inode *inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 /* Set up the RPC argument and reply structs
903 * NB: take care not to mess about with data->commit et al. */
904
905 data->req = req;
Al Viro3d4ff432011-06-22 18:40:12 -0400906 data->inode = inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000907 data->cred = req->wb_context->cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 data->args.fh = NFS_FH(inode);
910 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -0400911 /* pnfs_set_layoutcommit needs this */
912 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 data->args.pgbase = req->wb_pgbase + offset;
914 data->args.pages = data->pagevec;
915 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500916 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400917 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400918 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -0400919 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
920 case 0:
921 break;
922 case FLUSH_COND_STABLE:
923 if (nfs_need_commit(NFS_I(inode)))
924 break;
925 default:
926 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929 data->res.fattr = &data->fattr;
930 data->res.count = count;
931 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400932 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400933}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400935static int nfs_do_write(struct nfs_write_data *data,
936 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400937 int how)
938{
Stephen Rothwell5f00bcb2011-07-25 13:59:46 -0400939 struct inode *inode = data->args.context->dentry->d_inode;
Andy Adamson0382b742011-03-03 15:13:45 +0000940
Andy Adamsond138d5d2011-03-03 15:13:41 +0000941 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942}
943
Trond Myklebust275acaa2011-07-12 13:42:02 -0400944static int nfs_do_multiple_writes(struct list_head *head,
945 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -0400946 int how)
947{
948 struct nfs_write_data *data;
949 int ret = 0;
950
951 while (!list_empty(head)) {
952 int ret2;
953
954 data = list_entry(head->next, struct nfs_write_data, list);
955 list_del_init(&data->list);
956
Trond Myklebustdce81292011-07-13 15:59:19 -0400957 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400958 if (ret == 0)
959 ret = ret2;
960 }
961 return ret;
962}
963
Fred6d884e82008-03-19 11:24:38 -0400964/* If a nfs_flush_* function fails, it should remove reqs from @head and
965 * call this on each, which will prepare them to be retried on next
966 * writeback using standard nfs.
967 */
968static void nfs_redirty_request(struct nfs_page *req)
969{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400970 struct page *page = req->wb_page;
971
Fred6d884e82008-03-19 11:24:38 -0400972 nfs_mark_request_dirty(req);
Fred Isaman9994b622012-03-08 17:29:34 -0500973 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400974 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400975}
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977/*
978 * Generate multiple small requests to write out a single
979 * contiguous dirty area on one page.
980 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400981static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
Fred Isamanc76069b2011-03-03 15:13:48 +0000983 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 struct page *page = req->wb_page;
985 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -0400986 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700987 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400989 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990
991 nfs_list_remove_request(req);
992
Trond Myklebustb31268a2011-03-21 17:02:00 -0400993 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
994 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
995 desc->pg_count > wsize))
996 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
997
998
Trond Myklebust275acaa2011-07-12 13:42:02 -0400999 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +00001000 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001001 do {
1002 size_t len = min(nbytes, wsize);
1003
Trond Myklebust8d5658c2007-04-10 09:26:35 -04001004 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 if (!data)
1006 goto out_bad;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001007 data->pagevec[0] = page;
Trond Myklebustf13c3622011-09-12 11:47:53 -04001008 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001009 list_add(&data->list, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001011 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001012 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001013 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 atomic_set(&req->wb_complete, requests);
Trond Myklebust50828d72011-07-12 13:42:02 -04001015 desc->pg_rpc_callops = &nfs_write_partial_ops;
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001016 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018out_bad:
Trond Myklebust275acaa2011-07-12 13:42:02 -04001019 while (!list_empty(res)) {
1020 data = list_entry(res->next, struct nfs_write_data, list);
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001021 list_del(&data->list);
Fred Isaman8ccd2712012-04-20 14:47:35 -04001022 nfs_writedata_release(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 }
Trond Myklebust61822ab2006-12-05 00:35:42 -05001024 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 return -ENOMEM;
1026}
1027
1028/*
1029 * Create an RPC task for the given write request and kick it.
1030 * The page must have been locked by the caller.
1031 *
1032 * It may happen that the page we're passed is not marked dirty.
1033 * This is the case if nfs_updatepage detects a conflicting request
1034 * that has been written but not committed.
1035 */
Trond Myklebust275acaa2011-07-12 13:42:02 -04001036static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037{
1038 struct nfs_page *req;
1039 struct page **pages;
1040 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +00001041 struct list_head *head = &desc->pg_list;
Peng Tao3b609182011-07-15 03:33:42 -04001042 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Fred Isamanc76069b2011-03-03 15:13:48 +00001044 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
1045 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +00001046 if (!data) {
1047 while (!list_empty(head)) {
1048 req = nfs_list_entry(head->next);
1049 nfs_list_remove_request(req);
1050 nfs_redirty_request(req);
1051 }
1052 ret = -ENOMEM;
1053 goto out;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 while (!list_empty(head)) {
1057 req = nfs_list_entry(head->next);
1058 nfs_list_remove_request(req);
1059 nfs_list_add_request(req, &data->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
1062 req = nfs_list_entry(data->pages.next);
1063
Trond Myklebustb31268a2011-03-21 17:02:00 -04001064 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1065 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1066 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1067
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 /* Set up the argument struct */
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001069 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001070 list_add(&data->list, res);
Trond Myklebust50828d72011-07-12 13:42:02 -04001071 desc->pg_rpc_callops = &nfs_write_full_ops;
Fred Isaman44b83792011-03-03 15:13:44 +00001072out:
Fred Isaman44b83792011-03-03 15:13:44 +00001073 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
Trond Myklebustdce81292011-07-13 15:59:19 -04001076int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1077{
1078 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1079 return nfs_flush_multi(desc, head);
1080 return nfs_flush_one(desc, head);
1081}
1082
1083static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001084{
Trond Myklebust275acaa2011-07-12 13:42:02 -04001085 LIST_HEAD(head);
1086 int ret;
1087
Trond Myklebustdce81292011-07-13 15:59:19 -04001088 ret = nfs_generic_flush(desc, &head);
Trond Myklebust50828d72011-07-12 13:42:02 -04001089 if (ret == 0)
1090 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
Trond Myklebustdce81292011-07-13 15:59:19 -04001091 desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001092 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001093}
Trond Myklebust1751c362011-06-10 13:30:23 -04001094
1095static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1096 .pg_test = nfs_generic_pg_test,
1097 .pg_doio = nfs_generic_pg_writepages,
1098};
1099
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001100void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
Trond Myklebust1751c362011-06-10 13:30:23 -04001101 struct inode *inode, int ioflags)
1102{
1103 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1104 NFS_SERVER(inode)->wsize, ioflags);
1105}
1106
Trond Myklebustdce81292011-07-13 15:59:19 -04001107void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1108{
1109 pgio->pg_ops = &nfs_pageio_write_ops;
1110 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1111}
Trond Myklebust1f945352011-07-13 15:59:57 -04001112EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001113
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001114static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1115 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
Trond Myklebust1751c362011-06-10 13:30:23 -04001117 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1118 nfs_pageio_init_write_mds(pgio, inode, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
1121/*
1122 * Handle a write reply that flushed part of a page.
1123 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001124static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001126 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Chuck Lever48186c72008-06-11 17:56:05 -04001128 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1129 task->tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001130 data->req->wb_context->dentry->d_inode->i_sb->s_id,
Chuck Lever48186c72008-06-11 17:56:05 -04001131 (long long)
Al Viro3d4ff432011-06-22 18:40:12 -04001132 NFS_FILEID(data->req->wb_context->dentry->d_inode),
Chuck Lever48186c72008-06-11 17:56:05 -04001133 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001135 nfs_writeback_done(task, data);
1136}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001137
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001138static void nfs_writeback_release_partial(void *calldata)
1139{
1140 struct nfs_write_data *data = calldata;
1141 struct nfs_page *req = data->req;
1142 struct page *page = req->wb_page;
1143 int status = data->task.tk_status;
1144
1145 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001146 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001147 nfs_context_set_write_error(req->wb_context, status);
1148 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001149 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001152 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001153 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001154
Trond Myklebust587142f2007-07-02 09:57:54 -04001155 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001156 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1157 /* Do nothing we need to resend the writes */
1158 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1159 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1160 dprintk(" defer commit\n");
1161 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1162 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1163 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1164 dprintk(" server reboot detected\n");
1165 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001166 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001167 } else
1168 dprintk(" OK\n");
1169
1170out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001172 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001173 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
Andy Adamsondef6ed72009-04-01 09:22:26 -04001176void nfs_write_prepare(struct rpc_task *task, void *calldata)
1177{
1178 struct nfs_write_data *data = calldata;
Bryan Schumakerc6cb80d2012-03-19 14:54:39 -04001179 NFS_PROTO(data->inode)->write_rpc_prepare(task, data);
Andy Adamsondef6ed72009-04-01 09:22:26 -04001180}
Andy Adamsondef6ed72009-04-01 09:22:26 -04001181
Trond Myklebust788e7a82006-03-20 13:44:27 -05001182static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001183 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001184 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001185 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001186};
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/*
1189 * Handle a write reply that flushes a whole page.
1190 *
1191 * FIXME: There is an inherent race with invalidate_inode_pages and
1192 * writebacks since the page->count is kept > 1 for as long
1193 * as the page has a write request pending.
1194 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001195static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001197 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001199 nfs_writeback_done(task, data);
1200}
1201
1202static void nfs_writeback_release_full(void *calldata)
1203{
1204 struct nfs_write_data *data = calldata;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001205 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 /* Update attributes as result of writeback. */
1208 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001209 struct nfs_page *req = nfs_list_entry(data->pages.next);
1210 struct page *page = req->wb_page;
1211
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
Chuck Lever48186c72008-06-11 17:56:05 -04001214 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1215 data->task.tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001216 req->wb_context->dentry->d_inode->i_sb->s_id,
1217 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 req->wb_bytes,
1219 (long long)req_offset(req));
1220
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001221 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001222 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001223 nfs_context_set_write_error(req->wb_context, status);
1224 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001225 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001228 if (nfs_write_need_commit(data)) {
1229 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001230 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001231 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 goto next;
1233 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001234 dprintk(" OK\n");
1235remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 next:
Fred Isaman9994b622012-03-08 17:29:34 -05001238 nfs_unlock_request(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001239 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001241 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242}
1243
Trond Myklebust788e7a82006-03-20 13:44:27 -05001244static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001245 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001246 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001247 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001248};
1249
1250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251/*
1252 * This function is called when the WRITE call is complete.
1253 */
Fred Isaman13602892011-02-11 15:42:38 +00001254void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 struct nfs_writeargs *argp = &data->args;
1257 struct nfs_writeres *resp = &data->res;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001258 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
Chuck Levera3f565b2007-01-31 12:14:01 -05001260 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 task->tk_pid, task->tk_status);
1262
Chuck Leverf551e442006-09-20 14:33:04 -04001263 /*
1264 * ->write_done will attempt to use post-op attributes to detect
1265 * conflicting writes by other clients. A strict interpretation
1266 * of close-to-open would allow us to continue caching even if
1267 * another writer had changed the file, but some applications
1268 * depend on tighter cache coherency when writing.
1269 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001270 status = NFS_PROTO(data->inode)->write_done(task, data);
1271 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001272 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001273 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1276 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1277 /* We tried a write call, but the server did not
1278 * commit data to stable storage even though we
1279 * requested it.
1280 * Note: There is a known bug in Tru64 < 5.0 in which
1281 * the server reports NFS_DATA_SYNC, but performs
1282 * NFS_FILE_SYNC. We therefore implement this checking
1283 * as a dprintk() in order to avoid filling syslog.
1284 */
1285 static unsigned long complain;
1286
Fred Isamana69aef12011-03-03 15:13:47 +00001287 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001289 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 " (committed = %d) != (stable = %d)\n",
Rakib Mullick2b72c9c2011-11-01 12:23:42 +06001291 NFS_SERVER(data->inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 resp->verf->committed, argp->stable);
1293 complain = jiffies + 300 * HZ;
1294 }
1295 }
1296#endif
1297 /* Is this a short write? */
1298 if (task->tk_status >= 0 && resp->count < argp->count) {
1299 static unsigned long complain;
1300
Chuck Lever91d5b472006-03-20 13:44:14 -05001301 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 /* Has the server at least made some progress? */
1304 if (resp->count != 0) {
1305 /* Was this an NFSv2 write or an NFSv3 stable write? */
1306 if (resp->verf->committed != NFS_UNSTABLE) {
1307 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001308 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 argp->offset += resp->count;
1310 argp->pgbase += resp->count;
1311 argp->count -= resp->count;
1312 } else {
1313 /* Resend as a stable write in order to avoid
1314 * headaches in the case of a server crash.
1315 */
1316 argp->stable = NFS_FILE_SYNC;
1317 }
Trond Myklebustd00c5d42011-10-19 12:17:29 -07001318 rpc_restart_call_prepare(task);
Fred Isaman13602892011-02-11 15:42:38 +00001319 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 }
1321 if (time_before(complain, jiffies)) {
1322 printk(KERN_WARNING
1323 "NFS: Server wrote zero bytes, expected %u.\n",
1324 argp->count);
1325 complain = jiffies + 300 * HZ;
1326 }
1327 /* Can't do anything about it except throw an error. */
1328 task->tk_status = -EIO;
1329 }
Fred Isaman13602892011-02-11 15:42:38 +00001330 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331}
1332
1333
1334#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001335static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1336{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001337 int ret;
1338
Trond Myklebust71d0a612010-04-22 15:35:57 -04001339 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1340 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001341 if (!may_wait)
1342 return 0;
1343 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1344 NFS_INO_COMMIT,
1345 nfs_wait_bit_killable,
1346 TASK_KILLABLE);
1347 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001348}
1349
Fred Isamane0c2b382011-03-23 13:27:53 +00001350void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001351{
1352 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1353 smp_mb__after_clear_bit();
1354 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1355}
Fred Isamane0c2b382011-03-23 13:27:53 +00001356EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001357
Fred Isamane0c2b382011-03-23 13:27:53 +00001358void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359{
Trond Myklebust383ba712008-02-19 20:04:20 -05001360 struct nfs_write_data *wdata = data;
1361
1362 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 nfs_commit_free(wdata);
1364}
Fred Isamane0c2b382011-03-23 13:27:53 +00001365EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Fred Isamane0c2b382011-03-23 13:27:53 +00001367int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001368 const struct rpc_call_ops *call_ops,
1369 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370{
Trond Myklebust07737692007-10-25 18:42:54 -04001371 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001372 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001373 struct rpc_message msg = {
1374 .rpc_argp = &data->args,
1375 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001376 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001377 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001378 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001379 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001380 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001381 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001382 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001383 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001384 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001385 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001386 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001387 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001388 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001389 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Chuck Levera3f565b2007-01-31 12:14:01 -05001391 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001392
Trond Myklebust07737692007-10-25 18:42:54 -04001393 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001394 if (IS_ERR(task))
1395 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001396 if (how & FLUSH_SYNC)
1397 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001398 rpc_put_task(task);
1399 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400}
Fred Isamane0c2b382011-03-23 13:27:53 +00001401EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
1403/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001404 * Set up the argument/result storage required for the RPC call.
1405 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001406void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001407 struct list_head *head,
1408 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001409{
1410 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001411 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001412
1413 /* Set up the RPC argument and reply structs
1414 * NB: take care not to mess about with data->commit et al. */
1415
1416 list_splice_init(head, &data->pages);
1417
1418 data->inode = inode;
1419 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001420 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001421 data->mds_ops = &nfs_commit_ops;
1422
1423 data->args.fh = NFS_FH(data->inode);
1424 /* Note: we always request a commit of the entire inode */
1425 data->args.offset = 0;
1426 data->args.count = 0;
1427 data->args.context = get_nfs_open_context(first->wb_context);
1428 data->res.count = 0;
1429 data->res.fattr = &data->fattr;
1430 data->res.verf = &data->verf;
1431 nfs_fattr_init(&data->fattr);
1432}
Fred Isamane0c2b382011-03-23 13:27:53 +00001433EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001434
Fred Isamane0c2b382011-03-23 13:27:53 +00001435void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001436 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001437{
1438 struct nfs_page *req;
1439
1440 while (!list_empty(page_list)) {
1441 req = nfs_list_entry(page_list->next);
1442 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001443 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001444 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1445 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1446 BDI_RECLAIMABLE);
Fred Isaman9994b622012-03-08 17:29:34 -05001447 nfs_unlock_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001448 }
1449}
Fred Isamane0c2b382011-03-23 13:27:53 +00001450EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001451
Fred Isaman9ace33c2011-03-23 13:27:45 +00001452/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 * Commit dirty pages
1454 */
1455static int
Chuck Lever40859d72005-11-30 18:09:02 -05001456nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
1458 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001460 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461
1462 if (!data)
1463 goto out_bad;
1464
1465 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001466 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001467 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001469 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001470 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return -ENOMEM;
1472}
1473
1474/*
1475 * COMMIT call returned
1476 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001477static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001479 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Chuck Levera3f565b2007-01-31 12:14:01 -05001481 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 task->tk_pid, task->tk_status);
1483
Trond Myklebust788e7a82006-03-20 13:44:27 -05001484 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001485 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001486}
1487
Fred Isamane0c2b382011-03-23 13:27:53 +00001488void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001489{
Fred Isaman5917ce82011-03-23 13:27:48 +00001490 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001491 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 while (!list_empty(&data->pages)) {
1494 req = nfs_list_entry(data->pages.next);
1495 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001496 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497
Chuck Lever48186c72008-06-11 17:56:05 -04001498 dprintk("NFS: commit (%s/%lld %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001499 req->wb_context->dentry->d_sb->s_id,
1500 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 req->wb_bytes,
1502 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001503 if (status < 0) {
1504 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001506 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 goto next;
1508 }
1509
1510 /* Okay, COMMIT succeeded, apparently. Check the verifier
1511 * returned by the server against all stored verfs. */
1512 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1513 /* We have a match */
1514 nfs_inode_remove_request(req);
1515 dprintk(" OK\n");
1516 goto next;
1517 }
1518 /* We have a mismatch. Write the page again */
1519 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001520 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 next:
Fred Isaman9994b622012-03-08 17:29:34 -05001522 nfs_unlock_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001524}
Fred Isamane0c2b382011-03-23 13:27:53 +00001525EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001526
1527static void nfs_commit_release(void *calldata)
1528{
1529 struct nfs_write_data *data = calldata;
1530
1531 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001532 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001533 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001535
1536static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001537 .rpc_call_prepare = nfs_write_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001538 .rpc_call_done = nfs_commit_done,
1539 .rpc_release = nfs_commit_release,
1540};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Trond Myklebustb608b282010-07-30 15:31:54 -04001542int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001545 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001546 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547
Trond Myklebustb8413f92011-03-21 15:37:01 -04001548 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1549 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001550 goto out_mark_dirty;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001551 res = nfs_scan_commit(inode, &head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001553 int error;
1554
1555 error = pnfs_commit_list(inode, &head, how);
1556 if (error == PNFS_NOT_ATTEMPTED)
1557 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001558 if (error < 0)
1559 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001560 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001561 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001562 error = wait_on_bit(&NFS_I(inode)->flags,
1563 NFS_INO_COMMIT,
1564 nfs_wait_bit_killable,
1565 TASK_KILLABLE);
1566 if (error < 0)
1567 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001568 } else
1569 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001570 return res;
1571 /* Note: If we exit without ensuring that the commit is complete,
1572 * we must mark the inode as dirty. Otherwise, future calls to
1573 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1574 * that the data is on the disk.
1575 */
1576out_mark_dirty:
1577 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return res;
1579}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001580
1581static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1582{
Trond Myklebust420e3642010-02-19 17:00:02 -08001583 struct nfs_inode *nfsi = NFS_I(inode);
1584 int flags = FLUSH_SYNC;
1585 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001586
Jeff Layton3236c3e2011-10-11 09:49:21 -04001587 /* no commits means nothing needs to be done */
1588 if (!nfsi->ncommit)
1589 return ret;
1590
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001591 if (wbc->sync_mode == WB_SYNC_NONE) {
1592 /* Don't commit yet if this is a non-blocking flush and there
1593 * are a lot of outstanding writes for this mapping.
1594 */
1595 if (nfsi->ncommit <= (nfsi->npages >> 1))
1596 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001597
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001598 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001599 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001600 }
1601
Trond Myklebust420e3642010-02-19 17:00:02 -08001602 ret = nfs_commit_inode(inode, flags);
1603 if (ret >= 0) {
1604 if (wbc->sync_mode == WB_SYNC_NONE) {
1605 if (ret < wbc->nr_to_write)
1606 wbc->nr_to_write -= ret;
1607 else
1608 wbc->nr_to_write = 0;
1609 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001610 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001611 }
1612out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001613 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1614 return ret;
1615}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001616#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001617static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1618{
1619 return 0;
1620}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621#endif
1622
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001623int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1624{
Andy Adamson863a3c62011-03-23 13:27:54 +00001625 int ret;
1626
1627 ret = nfs_commit_unstable_pages(inode, wbc);
1628 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001629 int status;
1630 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001631
Wu Fengguang846d5a02011-05-05 21:10:38 -06001632 if (wbc->sync_mode == WB_SYNC_NONE)
Andy Adamsonef311532011-03-12 02:58:10 -05001633 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001634
1635 status = pnfs_layoutcommit_inode(inode, sync);
1636 if (status < 0)
1637 return status;
1638 }
1639 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001640}
1641
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001642/*
1643 * flush the inode to disk.
1644 */
1645int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001646{
1647 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001648 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001649 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001650 .range_start = 0,
1651 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001652 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001653
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001654 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001655}
1656
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001657int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1658{
1659 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001660 int ret = 0;
1661
1662 BUG_ON(!PageLocked(page));
1663 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001664 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001665 req = nfs_page_find_request(page);
1666 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001667 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001668 if (nfs_lock_request_dontget(req)) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001669 nfs_clear_request_commit(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001670 nfs_inode_remove_request(req);
1671 /*
1672 * In case nfs_inode_remove_request has marked the
1673 * page as being dirty
1674 */
1675 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1676 nfs_unlock_request(req);
1677 break;
1678 }
1679 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001680 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001681 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001682 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001683 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001684 return ret;
1685}
1686
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001687/*
1688 * Write back all requests on one page - we do this before reading it.
1689 */
1690int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001691{
1692 loff_t range_start = page_offset(page);
1693 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001694 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001695 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001696 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001697 .range_start = range_start,
1698 .range_end = range_end,
1699 };
1700 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001701
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001702 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001703 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001704 if (clear_page_dirty_for_io(page)) {
1705 ret = nfs_writepage_locked(page, &wbc);
1706 if (ret < 0)
1707 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001708 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001709 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001710 if (!PagePrivate(page))
1711 break;
1712 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001713 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001714 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001715 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001716 return 0;
1717out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001718 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001719}
1720
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001721#ifdef CONFIG_MIGRATION
1722int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001723 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001724{
Jeff Layton2da95652011-10-12 10:57:42 -04001725 /*
1726 * If PagePrivate is set, then the page is currently associated with
1727 * an in-progress read or write request. Don't try to migrate it.
1728 *
1729 * FIXME: we could do this in principle, but we'll need a way to ensure
1730 * that we can safely release the inode reference while holding
1731 * the page lock.
1732 */
1733 if (PagePrivate(page))
1734 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001735
Trond Myklebust7549ad52010-02-08 09:32:34 -05001736 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001737
Mel Gormana6bc32b2012-01-12 17:19:43 -08001738 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001739}
1740#endif
1741
David Howellsf7b422b2006-06-09 09:34:33 -04001742int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743{
1744 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1745 sizeof(struct nfs_write_data),
1746 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001747 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 if (nfs_wdata_cachep == NULL)
1749 return -ENOMEM;
1750
Matthew Dobson93d23412006-03-26 01:37:50 -08001751 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1752 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 if (nfs_wdata_mempool == NULL)
1754 return -ENOMEM;
1755
Matthew Dobson93d23412006-03-26 01:37:50 -08001756 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1757 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 if (nfs_commit_mempool == NULL)
1759 return -ENOMEM;
1760
Peter Zijlstra89a09142007-03-16 13:38:26 -08001761 /*
1762 * NFS congestion size, scale with available memory.
1763 *
1764 * 64MB: 8192k
1765 * 128MB: 11585k
1766 * 256MB: 16384k
1767 * 512MB: 23170k
1768 * 1GB: 32768k
1769 * 2GB: 46340k
1770 * 4GB: 65536k
1771 * 8GB: 92681k
1772 * 16GB: 131072k
1773 *
1774 * This allows larger machines to have larger/more transfers.
1775 * Limit the default to 256M
1776 */
1777 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1778 if (nfs_congestion_kb > 256*1024)
1779 nfs_congestion_kb = 256*1024;
1780
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 return 0;
1782}
1783
David Brownell266bee82006-06-27 12:59:15 -07001784void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785{
1786 mempool_destroy(nfs_commit_mempool);
1787 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001788 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789}
1790