blob: 46aa4389ce135952084ed9a5df1dde35b5ec54b7 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/write.c
3 *
Trond Myklebust7c85d902006-12-13 15:23:48 -05004 * Write file data over NFS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
7 */
8
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/writeback.h>
Peter Zijlstra89a09142007-03-16 13:38:26 -080015#include <linux/swap.h>
Trond Myklebust074cc1d2009-08-10 08:54:13 -040016#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070022#include <linux/backing-dev.h>
23
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
26#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050027#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050028#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040029#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040030#include "fscache.h"
Fred Isaman94ad1c82011-03-01 01:34:14 +000031#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33#define NFSDBG_FACILITY NFSDBG_PAGECACHE
34
35#define MIN_POOL_WRITE (32)
36#define MIN_POOL_COMMIT (4)
37
38/*
39 * Local function declarations
40 */
Trond Myklebustc63c7b02007-04-02 19:29:52 -040041static void nfs_pageio_init_write(struct nfs_pageio_descriptor *desc,
42 struct inode *inode, int ioflags);
Fred Isamanf8512ad2008-03-19 11:24:39 -040043static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050044static const struct rpc_call_ops nfs_write_partial_ops;
45static const struct rpc_call_ops nfs_write_full_ops;
46static const struct rpc_call_ops nfs_commit_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Christoph Lametere18b8902006-12-06 20:33:20 -080048static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050049static mempool_t *nfs_wdata_mempool;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050static mempool_t *nfs_commit_mempool;
51
Trond Myklebustc9d8f892008-04-15 16:56:39 -040052struct nfs_write_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070053{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080054 struct nfs_write_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOFS);
Chuck Lever40859d72005-11-30 18:09:02 -050055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 if (p) {
57 memset(p, 0, sizeof(*p));
58 INIT_LIST_HEAD(&p->pages);
59 }
60 return p;
61}
Fred Isamane0c2b382011-03-23 13:27:53 +000062EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Trond Myklebust5e4424a2008-02-25 21:53:49 -080064void nfs_commit_free(struct nfs_write_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065{
Chuck Lever40859d72005-11-30 18:09:02 -050066 if (p && (p->pagevec != &p->page_array[0]))
67 kfree(p->pagevec);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 mempool_free(p, nfs_commit_mempool);
69}
Fred Isamane0c2b382011-03-23 13:27:53 +000070EXPORT_SYMBOL_GPL(nfs_commit_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Trond Myklebust8d5658c2007-04-10 09:26:35 -040072struct nfs_write_data *nfs_writedata_alloc(unsigned int pagecount)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050073{
Christoph Lametere6b4f8d2006-12-06 20:33:14 -080074 struct nfs_write_data *p = mempool_alloc(nfs_wdata_mempool, GFP_NOFS);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050075
76 if (p) {
77 memset(p, 0, sizeof(*p));
78 INIT_LIST_HEAD(&p->pages);
Trond Myklebuste9f7bee2006-09-08 09:48:54 -070079 p->npages = pagecount;
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040080 if (pagecount <= ARRAY_SIZE(p->page_array))
81 p->pagevec = p->page_array;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050082 else {
Chuck Lever0d0b5cb2006-05-25 01:40:53 -040083 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_NOFS);
84 if (!p->pagevec) {
Trond Myklebust3feb2d42006-03-20 13:44:37 -050085 mempool_free(p, nfs_wdata_mempool);
86 p = NULL;
87 }
88 }
89 }
90 return p;
91}
92
Trond Myklebust1ae88b22009-08-12 09:12:30 -040093void nfs_writedata_free(struct nfs_write_data *p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050094{
95 if (p && (p->pagevec != &p->page_array[0]))
96 kfree(p->pagevec);
97 mempool_free(p, nfs_wdata_mempool);
98}
99
Trond Myklebustdce81292011-07-13 15:59:19 -0400100void nfs_writedata_release(struct nfs_write_data *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Fred Isaman5053aa52011-03-03 15:13:43 +0000102 put_lseg(wdata->lseg);
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;
Trond Myklebustacee4782008-01-22 17:13:07 -0500238 if (nfs_set_page_tag_locked(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,
Trond Myklebustacee4782008-01-22 17:13:07 -0500241 * then the call to nfs_set_page_tag_locked() 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 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400377static int 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);
380 int error;
381
Trond Myklebuste7d39062008-06-13 12:12:32 -0400382 error = radix_tree_preload(GFP_NOFS);
383 if (error != 0)
384 goto out;
385
386 /* Lock the request! */
387 nfs_lock_request_dontget(req);
388
389 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 error = radix_tree_insert(&nfsi->nfs_page_tree, req->wb_index, req);
Nick Piggin27852592008-02-04 23:48:37 -0800391 BUG_ON(error);
Trond Myklebust4d65c522011-03-25 14:15:11 -0400392 if (!nfsi->npages && nfs_have_delegation(inode, FMODE_WRITE))
393 nfsi->change_attr++;
Trond Myklebust2df485a2010-12-07 22:39:17 -0500394 set_bit(PG_MAPPED, &req->wb_flags);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500395 SetPagePrivate(req->wb_page);
Trond Myklebust277459d2006-12-05 00:35:35 -0500396 set_page_private(req->wb_page, (unsigned long)req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400398 kref_get(&req->wb_kref);
Nick Piggin27852592008-02-04 23:48:37 -0800399 radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index,
400 NFS_PAGE_TAG_LOCKED);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400401 spin_unlock(&inode->i_lock);
402 radix_tree_preload_end();
403out:
404 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405}
406
407/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800408 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 */
410static void nfs_inode_remove_request(struct nfs_page *req)
411{
Al Viro3d4ff432011-06-22 18:40:12 -0400412 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 struct nfs_inode *nfsi = NFS_I(inode);
414
415 BUG_ON (!NFS_WBACK_BUSY(req));
416
Trond Myklebust587142f2007-07-02 09:57:54 -0400417 spin_lock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500418 set_page_private(req->wb_page, 0);
Trond Myklebustdeb7d632006-03-20 13:44:50 -0500419 ClearPagePrivate(req->wb_page);
Trond Myklebust2df485a2010-12-07 22:39:17 -0500420 clear_bit(PG_MAPPED, &req->wb_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 radix_tree_delete(&nfsi->nfs_page_tree, req->wb_index);
422 nfsi->npages--;
Trond Myklebust4d65c522011-03-25 14:15:11 -0400423 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 nfs_release_request(req);
425}
426
Trond Myklebust61822ab2006-12-05 00:35:42 -0500427static void
Fred6d884e82008-03-19 11:24:38 -0400428nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500429{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500430 __set_page_dirty_nobuffers(req->wb_page);
431}
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
434/*
435 * Add a request to the inode's commit list.
436 */
437static void
Fred Isamana861a1e2011-03-23 13:27:51 +0000438nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Al Viro3d4ff432011-06-22 18:40:12 -0400440 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 struct nfs_inode *nfsi = NFS_I(inode);
442
Trond Myklebust587142f2007-07-02 09:57:54 -0400443 spin_lock(&inode->i_lock);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400444 set_bit(PG_CLEAN, &(req)->wb_flags);
Trond Myklebust5c369682007-06-17 15:27:42 -0400445 radix_tree_tag_set(&nfsi->nfs_page_tree,
446 req->wb_index,
447 NFS_PAGE_TAG_COMMIT);
Trond Myklebustff778d02010-02-19 16:53:39 -0800448 nfsi->ncommit++;
Trond Myklebust587142f2007-07-02 09:57:54 -0400449 spin_unlock(&inode->i_lock);
Fred Isamana861a1e2011-03-23 13:27:51 +0000450 pnfs_mark_request_commit(req, lseg);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700451 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700452 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
Trond Myklebusta1803042006-12-05 00:45:19 -0500453 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400455
Trond Myklebuste468bae2008-06-13 13:25:22 -0400456static int
457nfs_clear_request_commit(struct nfs_page *req)
458{
459 struct page *page = req->wb_page;
460
461 if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
462 dec_zone_page_state(page, NR_UNSTABLE_NFS);
463 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
464 return 1;
465 }
466 return 0;
467}
468
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400469static inline
470int nfs_write_need_commit(struct nfs_write_data *data)
471{
Fred Isaman465d5242011-03-23 13:27:44 +0000472 if (data->verf.committed == NFS_DATA_SYNC)
473 return data->lseg == NULL;
474 else
475 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400476}
477
478static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000479int nfs_reschedule_unstable_write(struct nfs_page *req,
480 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400481{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400482 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Fred Isamana861a1e2011-03-23 13:27:51 +0000483 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400484 return 1;
485 }
486 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
Fred6d884e82008-03-19 11:24:38 -0400487 nfs_mark_request_dirty(req);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400488 return 1;
489 }
490 return 0;
491}
492#else
493static inline void
Fred Isamana861a1e2011-03-23 13:27:51 +0000494nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400495{
496}
497
Trond Myklebuste468bae2008-06-13 13:25:22 -0400498static inline int
499nfs_clear_request_commit(struct nfs_page *req)
500{
501 return 0;
502}
503
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400504static inline
505int nfs_write_need_commit(struct nfs_write_data *data)
506{
507 return 0;
508}
509
510static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000511int nfs_reschedule_unstable_write(struct nfs_page *req,
512 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400513{
514 return 0;
515}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516#endif
517
Trond Myklebust47c62562009-03-16 08:13:41 -0400518#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400519static int
520nfs_need_commit(struct nfs_inode *nfsi)
521{
522 return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
523}
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525/*
526 * nfs_scan_commit - Scan an inode for commit requests
527 * @inode: NFS inode to scan
528 * @dst: destination list
529 * @idx_start: lower bound of page->index to scan.
530 * @npages: idx_start + npages sets the upper bound to scan.
531 *
532 * Moves requests from the inode's 'commit' request list.
533 * The requests are *not* checked to ensure that they form a contiguous set.
534 */
535static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400536nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
538 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustff778d02010-02-19 16:53:39 -0800539 int ret;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000540
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400541 if (!nfs_need_commit(nfsi))
542 return 0;
543
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000544 spin_lock(&inode->i_lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800545 ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
546 if (ret > 0)
547 nfsi->ncommit -= ret;
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000548 spin_unlock(&inode->i_lock);
549
Trond Myklebust2928db12010-02-19 17:03:18 -0800550 if (nfs_need_commit(NFS_I(inode)))
551 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000552
Trond Myklebustff778d02010-02-19 16:53:39 -0800553 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500555#else
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400556static inline int nfs_need_commit(struct nfs_inode *nfsi)
557{
558 return 0;
559}
560
Trond Myklebustca52fec2007-04-17 17:22:13 -0400561static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500562{
563 return 0;
564}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565#endif
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400568 * Search for an existing write request, and attempt to update
569 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400571 * If the attempt fails, then the existing request is flushed out
572 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400574static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
575 struct page *page,
576 unsigned int offset,
577 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400579 struct nfs_page *req;
580 unsigned int rqend;
581 unsigned int end;
582 int error;
583
584 if (!PagePrivate(page))
585 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400588 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 for (;;) {
Trond Myklebust277459d2006-12-05 00:35:35 -0500591 req = nfs_page_find_request_locked(page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400592 if (req == NULL)
593 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500594
Trond Myklebuste7d39062008-06-13 12:12:32 -0400595 rqend = req->wb_offset + req->wb_bytes;
596 /*
597 * Tell the caller to flush out the request if
598 * the offsets are non-contiguous.
599 * Note: nfs_flush_incompatible() will already
600 * have flushed out requests having wrong owners.
601 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400602 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400603 || end < req->wb_offset)
604 goto out_flushme;
605
606 if (nfs_set_page_tag_locked(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Trond Myklebuste7d39062008-06-13 12:12:32 -0400609 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400610 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400611 error = nfs_wait_on_request(req);
612 nfs_release_request(req);
613 if (error != 0)
614 goto out_err;
615 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617
Trond Myklebustff778d02010-02-19 16:53:39 -0800618 if (nfs_clear_request_commit(req) &&
Fred Isamana861a1e2011-03-23 13:27:51 +0000619 radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
620 req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) {
Trond Myklebustff778d02010-02-19 16:53:39 -0800621 NFS_I(inode)->ncommit--;
Fred Isamana861a1e2011-03-23 13:27:51 +0000622 pnfs_clear_request_commit(req);
623 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 /* Okay, the request matches. Update the region */
626 if (offset < req->wb_offset) {
627 req->wb_offset = offset;
628 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (end > rqend)
631 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400632 else
633 req->wb_bytes = rqend - req->wb_offset;
634out_unlock:
635 spin_unlock(&inode->i_lock);
636 return req;
637out_flushme:
638 spin_unlock(&inode->i_lock);
639 nfs_release_request(req);
640 error = nfs_wb_page(inode, page);
641out_err:
642 return ERR_PTR(error);
643}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Trond Myklebuste7d39062008-06-13 12:12:32 -0400645/*
646 * Try to update an existing write request, or create one if there is none.
647 *
648 * Note: Should always be called with the Page Lock held to prevent races
649 * if we have to add a new request. Also assumes that the caller has
650 * already called nfs_flush_incompatible() if necessary.
651 */
652static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
653 struct page *page, unsigned int offset, unsigned int bytes)
654{
655 struct inode *inode = page->mapping->host;
656 struct nfs_page *req;
657 int error;
658
659 req = nfs_try_to_update_request(inode, page, offset, bytes);
660 if (req != NULL)
661 goto out;
662 req = nfs_create_request(ctx, inode, page, offset, bytes);
663 if (IS_ERR(req))
664 goto out;
665 error = nfs_inode_add_request(inode, req);
666 if (error != 0) {
667 nfs_release_request(req);
668 req = ERR_PTR(error);
669 }
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400670out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400671 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
Trond Myklebuste7d39062008-06-13 12:12:32 -0400674static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
675 unsigned int offset, unsigned int count)
676{
677 struct nfs_page *req;
678
679 req = nfs_setup_write_request(ctx, page, offset, count);
680 if (IS_ERR(req))
681 return PTR_ERR(req);
682 /* Update file length */
683 nfs_grow_file(page, offset, count);
684 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400685 nfs_mark_request_dirty(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400686 nfs_clear_page_tag_locked(req);
687 return 0;
688}
689
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690int nfs_flush_incompatible(struct file *file, struct page *page)
691{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400692 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500694 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 /*
696 * Look for a request corresponding to this page. If there
697 * is one, and it belongs to another file, we flush it out
698 * before we try to copy anything into the page. Do this
699 * due to the lack of an ACCESS-type call in NFSv2.
700 * Also do the same if we find a request from an existing
701 * dropped page.
702 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500703 do {
704 req = nfs_page_find_request(page);
705 if (req == NULL)
706 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400707 do_flush = req->wb_page != page || req->wb_context != ctx ||
708 req->wb_lock_context->lockowner != current->files ||
709 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500711 if (!do_flush)
712 return 0;
713 status = nfs_wb_page(page->mapping->host, page);
714 } while (status == 0);
715 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716}
717
718/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500719 * If the page cache is marked as unsafe or invalid, then we can't rely on
720 * the PageUptodate() flag. In this case, we will need to turn off
721 * write optimisations that depend on the page contents being correct.
722 */
723static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
724{
725 return PageUptodate(page) &&
726 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
727}
728
729/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 * Update and possibly write a cached page of an NFS file.
731 *
732 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
733 * things with a page scheduled for an RPC call (e.g. invalidate it).
734 */
735int nfs_updatepage(struct file *file, struct page *page,
736 unsigned int offset, unsigned int count)
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 inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 int status = 0;
741
Chuck Lever91d5b472006-03-20 13:44:14 -0500742 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
743
Chuck Lever48186c72008-06-11 17:56:05 -0400744 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800745 file->f_path.dentry->d_parent->d_name.name,
746 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400747 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500750 * is up to date, it may be more efficient to extend the write
751 * to cover the entire page in order to avoid fragmentation
752 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500754 if (nfs_write_pageuptodate(page, inode) &&
755 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100756 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500757 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
760
Trond Myklebuste21195a2006-12-05 00:35:39 -0500761 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400762 if (status < 0)
763 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700764 else
765 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Chuck Lever48186c72008-06-11 17:56:05 -0400767 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 return status;
770}
771
Fred Isamana861a1e2011-03-23 13:27:51 +0000772static void nfs_writepage_release(struct nfs_page *req,
773 struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400775 struct page *page = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
Fred Isamana861a1e2011-03-23 13:27:51 +0000777 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400778 nfs_inode_remove_request(req);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400779 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400780 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
Trond Myklebust3ff75762007-07-14 15:40:00 -0400783static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784{
785 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
786 case FLUSH_HIGHPRI:
787 return RPC_PRIORITY_HIGH;
788 case FLUSH_LOWPRI:
789 return RPC_PRIORITY_LOW;
790 }
791 return RPC_PRIORITY_NORMAL;
792}
793
Fred Isamana69aef12011-03-03 15:13:47 +0000794int nfs_initiate_write(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000795 struct rpc_clnt *clnt,
796 const struct rpc_call_ops *call_ops,
797 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798{
Andy Adamsond138d5d2011-03-03 15:13:41 +0000799 struct inode *inode = data->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400800 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400801 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400802 struct rpc_message msg = {
803 .rpc_argp = &data->args,
804 .rpc_resp = &data->res,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000805 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400806 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400807 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +0000808 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -0400809 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400810 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400811 .callback_ops = call_ops,
812 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500813 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -0400814 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400815 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400816 };
Trond Myklebust2c61be02010-04-09 19:54:50 -0400817 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Andy Adamsond138d5d2011-03-03 15:13:41 +0000819 /* Set up the initial task struct. */
820 NFS_PROTO(inode)->write_setup(data, &msg);
821
822 dprintk("NFS: %5u initiated write call "
823 "(req %s/%lld, %u bytes @ offset %llu)\n",
824 data->task.tk_pid,
825 inode->i_sb->s_id,
826 (long long)NFS_FILEID(inode),
827 data->args.count,
828 (unsigned long long)data->args.offset);
829
830 task = rpc_run_task(&task_setup_data);
831 if (IS_ERR(task)) {
832 ret = PTR_ERR(task);
833 goto out;
834 }
835 if (how & FLUSH_SYNC) {
836 ret = rpc_wait_for_completion_task(task);
837 if (ret == 0)
838 ret = task->tk_status;
839 }
840 rpc_put_task(task);
841out:
842 return ret;
843}
Fred Isamana69aef12011-03-03 15:13:47 +0000844EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +0000845
846/*
847 * Set up the argument/result storage required for the RPC call.
848 */
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400849static void nfs_write_rpcsetup(struct nfs_page *req,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000850 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000851 unsigned int count, unsigned int offset,
852 int how)
853{
Al Viro3d4ff432011-06-22 18:40:12 -0400854 struct inode *inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /* Set up the RPC argument and reply structs
857 * NB: take care not to mess about with data->commit et al. */
858
859 data->req = req;
Al Viro3d4ff432011-06-22 18:40:12 -0400860 data->inode = inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000861 data->cred = req->wb_context->cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 data->args.fh = NFS_FH(inode);
864 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -0400865 /* pnfs_set_layoutcommit needs this */
866 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 data->args.pgbase = req->wb_pgbase + offset;
868 data->args.pages = data->pagevec;
869 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500870 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400871 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400872 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -0400873 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
874 case 0:
875 break;
876 case FLUSH_COND_STABLE:
877 if (nfs_need_commit(NFS_I(inode)))
878 break;
879 default:
880 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400881 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
883 data->res.fattr = &data->fattr;
884 data->res.count = count;
885 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400886 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400887}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400889static int nfs_do_write(struct nfs_write_data *data,
890 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400891 int how)
892{
Stephen Rothwell5f00bcb2011-07-25 13:59:46 -0400893 struct inode *inode = data->args.context->dentry->d_inode;
Andy Adamson0382b742011-03-03 15:13:45 +0000894
Andy Adamsond138d5d2011-03-03 15:13:41 +0000895 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
897
Trond Myklebust275acaa2011-07-12 13:42:02 -0400898static int nfs_do_multiple_writes(struct list_head *head,
899 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -0400900 int how)
901{
902 struct nfs_write_data *data;
903 int ret = 0;
904
905 while (!list_empty(head)) {
906 int ret2;
907
908 data = list_entry(head->next, struct nfs_write_data, list);
909 list_del_init(&data->list);
910
Trond Myklebustdce81292011-07-13 15:59:19 -0400911 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400912 if (ret == 0)
913 ret = ret2;
914 }
915 return ret;
916}
917
Fred6d884e82008-03-19 11:24:38 -0400918/* If a nfs_flush_* function fails, it should remove reqs from @head and
919 * call this on each, which will prepare them to be retried on next
920 * writeback using standard nfs.
921 */
922static void nfs_redirty_request(struct nfs_page *req)
923{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400924 struct page *page = req->wb_page;
925
Fred6d884e82008-03-19 11:24:38 -0400926 nfs_mark_request_dirty(req);
Fred6d884e82008-03-19 11:24:38 -0400927 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400928 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400929}
930
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931/*
932 * Generate multiple small requests to write out a single
933 * contiguous dirty area on one page.
934 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400935static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936{
Fred Isamanc76069b2011-03-03 15:13:48 +0000937 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 struct page *page = req->wb_page;
939 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -0400940 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700941 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400943 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
945 nfs_list_remove_request(req);
946
Trond Myklebustb31268a2011-03-21 17:02:00 -0400947 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
948 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
949 desc->pg_count > wsize))
950 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
951
952
Trond Myklebust275acaa2011-07-12 13:42:02 -0400953 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000954 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700955 do {
956 size_t len = min(nbytes, wsize);
957
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400958 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 if (!data)
960 goto out_bad;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400961 data->pagevec[0] = page;
Trond Myklebustf13c3622011-09-12 11:47:53 -0400962 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400963 list_add(&data->list, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700965 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400966 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700967 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 atomic_set(&req->wb_complete, requests);
Trond Myklebust50828d72011-07-12 13:42:02 -0400969 desc->pg_rpc_callops = &nfs_write_partial_ops;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400970 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
972out_bad:
Trond Myklebust275acaa2011-07-12 13:42:02 -0400973 while (!list_empty(res)) {
974 data = list_entry(res->next, struct nfs_write_data, list);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400975 list_del(&data->list);
Fred Isaman0da2a4ac2011-01-19 14:18:50 -0500976 nfs_writedata_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500978 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 return -ENOMEM;
980}
981
982/*
983 * Create an RPC task for the given write request and kick it.
984 * The page must have been locked by the caller.
985 *
986 * It may happen that the page we're passed is not marked dirty.
987 * This is the case if nfs_updatepage detects a conflicting request
988 * that has been written but not committed.
989 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400990static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
992 struct nfs_page *req;
993 struct page **pages;
994 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000995 struct list_head *head = &desc->pg_list;
Peng Tao3b609182011-07-15 03:33:42 -0400996 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Fred Isamanc76069b2011-03-03 15:13:48 +0000998 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
999 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +00001000 if (!data) {
1001 while (!list_empty(head)) {
1002 req = nfs_list_entry(head->next);
1003 nfs_list_remove_request(req);
1004 nfs_redirty_request(req);
1005 }
1006 ret = -ENOMEM;
1007 goto out;
1008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 while (!list_empty(head)) {
1011 req = nfs_list_entry(head->next);
1012 nfs_list_remove_request(req);
1013 nfs_list_add_request(req, &data->pages);
1014 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
1017 req = nfs_list_entry(data->pages.next);
1018
Trond Myklebustb31268a2011-03-21 17:02:00 -04001019 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1020 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1021 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 /* Set up the argument struct */
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001024 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001025 list_add(&data->list, res);
Trond Myklebust50828d72011-07-12 13:42:02 -04001026 desc->pg_rpc_callops = &nfs_write_full_ops;
Fred Isaman44b83792011-03-03 15:13:44 +00001027out:
Fred Isaman44b83792011-03-03 15:13:44 +00001028 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Trond Myklebustdce81292011-07-13 15:59:19 -04001031int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1032{
1033 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1034 return nfs_flush_multi(desc, head);
1035 return nfs_flush_one(desc, head);
1036}
1037
1038static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001039{
Trond Myklebust275acaa2011-07-12 13:42:02 -04001040 LIST_HEAD(head);
1041 int ret;
1042
Trond Myklebustdce81292011-07-13 15:59:19 -04001043 ret = nfs_generic_flush(desc, &head);
Trond Myklebust50828d72011-07-12 13:42:02 -04001044 if (ret == 0)
1045 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
Trond Myklebustdce81292011-07-13 15:59:19 -04001046 desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001047 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001048}
Trond Myklebust1751c362011-06-10 13:30:23 -04001049
1050static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1051 .pg_test = nfs_generic_pg_test,
1052 .pg_doio = nfs_generic_pg_writepages,
1053};
1054
Trond Myklebust1f945352011-07-13 15:59:57 -04001055static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
Trond Myklebust1751c362011-06-10 13:30:23 -04001056 struct inode *inode, int ioflags)
1057{
1058 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1059 NFS_SERVER(inode)->wsize, ioflags);
1060}
1061
Trond Myklebustdce81292011-07-13 15:59:19 -04001062void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1063{
1064 pgio->pg_ops = &nfs_pageio_write_ops;
1065 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1066}
Trond Myklebust1f945352011-07-13 15:59:57 -04001067EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001068
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001069static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1070 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
Trond Myklebust1751c362011-06-10 13:30:23 -04001072 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1073 nfs_pageio_init_write_mds(pgio, inode, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074}
1075
1076/*
1077 * Handle a write reply that flushed part of a page.
1078 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001079static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001081 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Chuck Lever48186c72008-06-11 17:56:05 -04001083 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1084 task->tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001085 data->req->wb_context->dentry->d_inode->i_sb->s_id,
Chuck Lever48186c72008-06-11 17:56:05 -04001086 (long long)
Al Viro3d4ff432011-06-22 18:40:12 -04001087 NFS_FILEID(data->req->wb_context->dentry->d_inode),
Chuck Lever48186c72008-06-11 17:56:05 -04001088 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001090 nfs_writeback_done(task, data);
1091}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001092
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001093static void nfs_writeback_release_partial(void *calldata)
1094{
1095 struct nfs_write_data *data = calldata;
1096 struct nfs_page *req = data->req;
1097 struct page *page = req->wb_page;
1098 int status = data->task.tk_status;
1099
1100 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001101 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001102 nfs_context_set_write_error(req->wb_context, status);
1103 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001104 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 }
1106
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001107 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001108 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001109
Trond Myklebust587142f2007-07-02 09:57:54 -04001110 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001111 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1112 /* Do nothing we need to resend the writes */
1113 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1114 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1115 dprintk(" defer commit\n");
1116 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1117 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1118 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1119 dprintk(" server reboot detected\n");
1120 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001121 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001122 } else
1123 dprintk(" OK\n");
1124
1125out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001127 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001128 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
Andy Adamsondef6ed72009-04-01 09:22:26 -04001131#if defined(CONFIG_NFS_V4_1)
1132void nfs_write_prepare(struct rpc_task *task, void *calldata)
1133{
1134 struct nfs_write_data *data = calldata;
Andy Adamsondef6ed72009-04-01 09:22:26 -04001135
Trond Myklebust035168a2010-06-16 09:52:26 -04001136 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
1137 &data->args.seq_args,
Andy Adamsondef6ed72009-04-01 09:22:26 -04001138 &data->res.seq_res, 1, task))
1139 return;
1140 rpc_call_start(task);
1141}
1142#endif /* CONFIG_NFS_V4_1 */
1143
Trond Myklebust788e7a82006-03-20 13:44:27 -05001144static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001145#if defined(CONFIG_NFS_V4_1)
1146 .rpc_call_prepare = nfs_write_prepare,
1147#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001148 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001149 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001150};
1151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152/*
1153 * Handle a write reply that flushes a whole page.
1154 *
1155 * FIXME: There is an inherent race with invalidate_inode_pages and
1156 * writebacks since the page->count is kept > 1 for as long
1157 * as the page has a write request pending.
1158 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001159static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001161 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001163 nfs_writeback_done(task, data);
1164}
1165
1166static void nfs_writeback_release_full(void *calldata)
1167{
1168 struct nfs_write_data *data = calldata;
1169 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001170
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 /* Update attributes as result of writeback. */
1172 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001173 struct nfs_page *req = nfs_list_entry(data->pages.next);
1174 struct page *page = req->wb_page;
1175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Chuck Lever48186c72008-06-11 17:56:05 -04001178 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1179 data->task.tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001180 req->wb_context->dentry->d_inode->i_sb->s_id,
1181 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 req->wb_bytes,
1183 (long long)req_offset(req));
1184
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001185 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001186 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001187 nfs_context_set_write_error(req->wb_context, status);
1188 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001189 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001192 if (nfs_write_need_commit(data)) {
1193 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001194 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001195 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 goto next;
1197 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001198 dprintk(" OK\n");
1199remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001202 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001203 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001205 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
Trond Myklebust788e7a82006-03-20 13:44:27 -05001208static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001209#if defined(CONFIG_NFS_V4_1)
1210 .rpc_call_prepare = nfs_write_prepare,
1211#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001212 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001213 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001214};
1215
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/*
1218 * This function is called when the WRITE call is complete.
1219 */
Fred Isaman13602892011-02-11 15:42:38 +00001220void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 struct nfs_writeargs *argp = &data->args;
1223 struct nfs_writeres *resp = &data->res;
Andy Adamsoneedc0202009-04-01 09:22:41 -04001224 struct nfs_server *server = NFS_SERVER(data->inode);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001225 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
Chuck Levera3f565b2007-01-31 12:14:01 -05001227 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 task->tk_pid, task->tk_status);
1229
Chuck Leverf551e442006-09-20 14:33:04 -04001230 /*
1231 * ->write_done will attempt to use post-op attributes to detect
1232 * conflicting writes by other clients. A strict interpretation
1233 * of close-to-open would allow us to continue caching even if
1234 * another writer had changed the file, but some applications
1235 * depend on tighter cache coherency when writing.
1236 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001237 status = NFS_PROTO(data->inode)->write_done(task, data);
1238 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001239 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001240 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1243 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1244 /* We tried a write call, but the server did not
1245 * commit data to stable storage even though we
1246 * requested it.
1247 * Note: There is a known bug in Tru64 < 5.0 in which
1248 * the server reports NFS_DATA_SYNC, but performs
1249 * NFS_FILE_SYNC. We therefore implement this checking
1250 * as a dprintk() in order to avoid filling syslog.
1251 */
1252 static unsigned long complain;
1253
Fred Isamana69aef12011-03-03 15:13:47 +00001254 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001256 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 " (committed = %d) != (stable = %d)\n",
Andy Adamsoneedc0202009-04-01 09:22:41 -04001258 server->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 resp->verf->committed, argp->stable);
1260 complain = jiffies + 300 * HZ;
1261 }
1262 }
1263#endif
1264 /* Is this a short write? */
1265 if (task->tk_status >= 0 && resp->count < argp->count) {
1266 static unsigned long complain;
1267
Chuck Lever91d5b472006-03-20 13:44:14 -05001268 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1269
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 /* Has the server at least made some progress? */
1271 if (resp->count != 0) {
1272 /* Was this an NFSv2 write or an NFSv3 stable write? */
1273 if (resp->verf->committed != NFS_UNSTABLE) {
1274 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001275 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 argp->offset += resp->count;
1277 argp->pgbase += resp->count;
1278 argp->count -= resp->count;
1279 } else {
1280 /* Resend as a stable write in order to avoid
1281 * headaches in the case of a server crash.
1282 */
1283 argp->stable = NFS_FILE_SYNC;
1284 }
Trond Myklebust0110ee12009-12-07 09:00:24 -05001285 nfs_restart_rpc(task, server->nfs_client);
Fred Isaman13602892011-02-11 15:42:38 +00001286 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288 if (time_before(complain, jiffies)) {
1289 printk(KERN_WARNING
1290 "NFS: Server wrote zero bytes, expected %u.\n",
1291 argp->count);
1292 complain = jiffies + 300 * HZ;
1293 }
1294 /* Can't do anything about it except throw an error. */
1295 task->tk_status = -EIO;
1296 }
Fred Isaman13602892011-02-11 15:42:38 +00001297 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298}
1299
1300
1301#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001302static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1303{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001304 int ret;
1305
Trond Myklebust71d0a612010-04-22 15:35:57 -04001306 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1307 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001308 if (!may_wait)
1309 return 0;
1310 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1311 NFS_INO_COMMIT,
1312 nfs_wait_bit_killable,
1313 TASK_KILLABLE);
1314 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001315}
1316
Fred Isamane0c2b382011-03-23 13:27:53 +00001317void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001318{
1319 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1320 smp_mb__after_clear_bit();
1321 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1322}
Fred Isamane0c2b382011-03-23 13:27:53 +00001323EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001324
Fred Isamane0c2b382011-03-23 13:27:53 +00001325void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
Trond Myklebust383ba712008-02-19 20:04:20 -05001327 struct nfs_write_data *wdata = data;
1328
Fred Isaman988b6dc2011-03-23 13:27:52 +00001329 put_lseg(wdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -05001330 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 nfs_commit_free(wdata);
1332}
Fred Isamane0c2b382011-03-23 13:27:53 +00001333EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Fred Isamane0c2b382011-03-23 13:27:53 +00001335int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001336 const struct rpc_call_ops *call_ops,
1337 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
Trond Myklebust07737692007-10-25 18:42:54 -04001339 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001340 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001341 struct rpc_message msg = {
1342 .rpc_argp = &data->args,
1343 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001344 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001345 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001346 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001347 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001348 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001349 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001350 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001351 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001352 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001353 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001354 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001355 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001356 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001357 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
Chuck Levera3f565b2007-01-31 12:14:01 -05001359 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001360
Trond Myklebust07737692007-10-25 18:42:54 -04001361 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001362 if (IS_ERR(task))
1363 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001364 if (how & FLUSH_SYNC)
1365 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001366 rpc_put_task(task);
1367 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368}
Fred Isamane0c2b382011-03-23 13:27:53 +00001369EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
1371/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001372 * Set up the argument/result storage required for the RPC call.
1373 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001374void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001375 struct list_head *head,
1376 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001377{
1378 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001379 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001380
1381 /* Set up the RPC argument and reply structs
1382 * NB: take care not to mess about with data->commit et al. */
1383
1384 list_splice_init(head, &data->pages);
1385
1386 data->inode = inode;
1387 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001388 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001389 data->mds_ops = &nfs_commit_ops;
1390
1391 data->args.fh = NFS_FH(data->inode);
1392 /* Note: we always request a commit of the entire inode */
1393 data->args.offset = 0;
1394 data->args.count = 0;
1395 data->args.context = get_nfs_open_context(first->wb_context);
1396 data->res.count = 0;
1397 data->res.fattr = &data->fattr;
1398 data->res.verf = &data->verf;
1399 nfs_fattr_init(&data->fattr);
1400}
Fred Isamane0c2b382011-03-23 13:27:53 +00001401EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001402
Fred Isamane0c2b382011-03-23 13:27:53 +00001403void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001404 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001405{
1406 struct nfs_page *req;
1407
1408 while (!list_empty(page_list)) {
1409 req = nfs_list_entry(page_list->next);
1410 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001411 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001412 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1413 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1414 BDI_RECLAIMABLE);
1415 nfs_clear_page_tag_locked(req);
1416 }
1417}
Fred Isamane0c2b382011-03-23 13:27:53 +00001418EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001419
Fred Isaman9ace33c2011-03-23 13:27:45 +00001420/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 * Commit dirty pages
1422 */
1423static int
Chuck Lever40859d72005-11-30 18:09:02 -05001424nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001428 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 if (!data)
1431 goto out_bad;
1432
1433 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001434 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001435 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001437 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001438 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return -ENOMEM;
1440}
1441
1442/*
1443 * COMMIT call returned
1444 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001445static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001447 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Chuck Levera3f565b2007-01-31 12:14:01 -05001449 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 task->tk_pid, task->tk_status);
1451
Trond Myklebust788e7a82006-03-20 13:44:27 -05001452 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001453 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001454}
1455
Fred Isamane0c2b382011-03-23 13:27:53 +00001456void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001457{
Fred Isaman5917ce82011-03-23 13:27:48 +00001458 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001459 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 while (!list_empty(&data->pages)) {
1462 req = nfs_list_entry(data->pages.next);
1463 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -04001464 nfs_clear_request_commit(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
Chuck Lever48186c72008-06-11 17:56:05 -04001466 dprintk("NFS: commit (%s/%lld %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001467 req->wb_context->dentry->d_sb->s_id,
1468 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 req->wb_bytes,
1470 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001471 if (status < 0) {
1472 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001474 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 goto next;
1476 }
1477
1478 /* Okay, COMMIT succeeded, apparently. Check the verifier
1479 * returned by the server against all stored verfs. */
1480 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1481 /* We have a match */
1482 nfs_inode_remove_request(req);
1483 dprintk(" OK\n");
1484 goto next;
1485 }
1486 /* We have a mismatch. Write the page again */
1487 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001488 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001490 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001492}
Fred Isamane0c2b382011-03-23 13:27:53 +00001493EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001494
1495static void nfs_commit_release(void *calldata)
1496{
1497 struct nfs_write_data *data = calldata;
1498
1499 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001500 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001501 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001503
1504static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001505#if defined(CONFIG_NFS_V4_1)
1506 .rpc_call_prepare = nfs_write_prepare,
1507#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001508 .rpc_call_done = nfs_commit_done,
1509 .rpc_release = nfs_commit_release,
1510};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511
Trond Myklebustb608b282010-07-30 15:31:54 -04001512int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001515 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001516 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517
Trond Myklebustb8413f92011-03-21 15:37:01 -04001518 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1519 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001520 goto out_mark_dirty;
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001521 res = nfs_scan_commit(inode, &head, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001523 int error;
1524
1525 error = pnfs_commit_list(inode, &head, how);
1526 if (error == PNFS_NOT_ATTEMPTED)
1527 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001528 if (error < 0)
1529 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001530 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001531 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001532 error = wait_on_bit(&NFS_I(inode)->flags,
1533 NFS_INO_COMMIT,
1534 nfs_wait_bit_killable,
1535 TASK_KILLABLE);
1536 if (error < 0)
1537 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001538 } else
1539 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001540 return res;
1541 /* Note: If we exit without ensuring that the commit is complete,
1542 * we must mark the inode as dirty. Otherwise, future calls to
1543 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1544 * that the data is on the disk.
1545 */
1546out_mark_dirty:
1547 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 return res;
1549}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001550
1551static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1552{
Trond Myklebust420e3642010-02-19 17:00:02 -08001553 struct nfs_inode *nfsi = NFS_I(inode);
1554 int flags = FLUSH_SYNC;
1555 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001556
Jeff Layton3236c3e2011-10-11 09:49:21 -04001557 /* no commits means nothing needs to be done */
1558 if (!nfsi->ncommit)
1559 return ret;
1560
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001561 if (wbc->sync_mode == WB_SYNC_NONE) {
1562 /* Don't commit yet if this is a non-blocking flush and there
1563 * are a lot of outstanding writes for this mapping.
1564 */
1565 if (nfsi->ncommit <= (nfsi->npages >> 1))
1566 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001567
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001568 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001569 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001570 }
1571
Trond Myklebust420e3642010-02-19 17:00:02 -08001572 ret = nfs_commit_inode(inode, flags);
1573 if (ret >= 0) {
1574 if (wbc->sync_mode == WB_SYNC_NONE) {
1575 if (ret < wbc->nr_to_write)
1576 wbc->nr_to_write -= ret;
1577 else
1578 wbc->nr_to_write = 0;
1579 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001580 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001581 }
1582out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001583 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1584 return ret;
1585}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001586#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001587static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1588{
1589 return 0;
1590}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591#endif
1592
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001593int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1594{
Andy Adamson863a3c62011-03-23 13:27:54 +00001595 int ret;
1596
1597 ret = nfs_commit_unstable_pages(inode, wbc);
1598 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001599 int status;
1600 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001601
Wu Fengguang846d5a02011-05-05 21:10:38 -06001602 if (wbc->sync_mode == WB_SYNC_NONE)
Andy Adamsonef311532011-03-12 02:58:10 -05001603 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001604
1605 status = pnfs_layoutcommit_inode(inode, sync);
1606 if (status < 0)
1607 return status;
1608 }
1609 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001610}
1611
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001612/*
1613 * flush the inode to disk.
1614 */
1615int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001616{
1617 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001618 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001619 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001620 .range_start = 0,
1621 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001622 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001623
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001624 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001625}
1626
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001627int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1628{
1629 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001630 int ret = 0;
1631
1632 BUG_ON(!PageLocked(page));
1633 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001634 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001635 req = nfs_page_find_request(page);
1636 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001637 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001638 if (nfs_lock_request_dontget(req)) {
1639 nfs_inode_remove_request(req);
1640 /*
1641 * In case nfs_inode_remove_request has marked the
1642 * page as being dirty
1643 */
1644 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1645 nfs_unlock_request(req);
1646 break;
1647 }
1648 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001649 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001650 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001651 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001652 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001653 return ret;
1654}
1655
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001656/*
1657 * Write back all requests on one page - we do this before reading it.
1658 */
1659int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001660{
1661 loff_t range_start = page_offset(page);
1662 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001663 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001664 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001665 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001666 .range_start = range_start,
1667 .range_end = range_end,
1668 };
1669 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001670
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001671 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001672 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001673 if (clear_page_dirty_for_io(page)) {
1674 ret = nfs_writepage_locked(page, &wbc);
1675 if (ret < 0)
1676 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001677 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001678 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001679 if (!PagePrivate(page))
1680 break;
1681 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001682 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001683 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001684 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001685 return 0;
1686out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001687 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001688}
1689
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001690#ifdef CONFIG_MIGRATION
1691int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1692 struct page *page)
1693{
1694 struct nfs_page *req;
1695 int ret;
1696
Trond Myklebust7549ad52010-02-08 09:32:34 -05001697 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001698
Trond Myklebustcfb506e2010-07-30 15:31:57 -04001699 req = nfs_find_and_lock_request(page, false);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001700 ret = PTR_ERR(req);
1701 if (IS_ERR(req))
1702 goto out;
1703
1704 ret = migrate_page(mapping, newpage, page);
1705 if (!req)
1706 goto out;
1707 if (ret)
1708 goto out_unlock;
1709 page_cache_get(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001710 spin_lock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001711 req->wb_page = newpage;
1712 SetPagePrivate(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001713 set_page_private(newpage, (unsigned long)req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001714 ClearPagePrivate(page);
1715 set_page_private(page, 0);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001716 spin_unlock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001717 page_cache_release(page);
1718out_unlock:
1719 nfs_clear_page_tag_locked(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001720out:
1721 return ret;
1722}
1723#endif
1724
David Howellsf7b422b2006-06-09 09:34:33 -04001725int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726{
1727 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1728 sizeof(struct nfs_write_data),
1729 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001730 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 if (nfs_wdata_cachep == NULL)
1732 return -ENOMEM;
1733
Matthew Dobson93d23412006-03-26 01:37:50 -08001734 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1735 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 if (nfs_wdata_mempool == NULL)
1737 return -ENOMEM;
1738
Matthew Dobson93d23412006-03-26 01:37:50 -08001739 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1740 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 if (nfs_commit_mempool == NULL)
1742 return -ENOMEM;
1743
Peter Zijlstra89a09142007-03-16 13:38:26 -08001744 /*
1745 * NFS congestion size, scale with available memory.
1746 *
1747 * 64MB: 8192k
1748 * 128MB: 11585k
1749 * 256MB: 16384k
1750 * 512MB: 23170k
1751 * 1GB: 32768k
1752 * 2GB: 46340k
1753 * 4GB: 65536k
1754 * 8GB: 92681k
1755 * 16GB: 131072k
1756 *
1757 * This allows larger machines to have larger/more transfers.
1758 * Limit the default to 256M
1759 */
1760 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1761 if (nfs_congestion_kb > 256*1024)
1762 nfs_congestion_kb = 256*1024;
1763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 return 0;
1765}
1766
David Brownell266bee82006-06-27 12:59:15 -07001767void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
1769 mempool_destroy(nfs_commit_mempool);
1770 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001771 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772}
1773