blob: c9bd2a6b7d4ba757729fa28fa185d84edd745d70 [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);
Trond Myklebustb80c3cb2010-04-09 19:07:07 -0400431 __mark_inode_dirty(req->wb_page->mapping->host, I_DIRTY_DATASYNC);
Trond Myklebust61822ab2006-12-05 00:35:42 -0500432}
433
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
435/*
436 * Add a request to the inode's commit list.
437 */
438static void
Fred Isamana861a1e2011-03-23 13:27:51 +0000439nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Al Viro3d4ff432011-06-22 18:40:12 -0400441 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 struct nfs_inode *nfsi = NFS_I(inode);
443
Trond Myklebust587142f2007-07-02 09:57:54 -0400444 spin_lock(&inode->i_lock);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400445 set_bit(PG_CLEAN, &(req)->wb_flags);
Trond Myklebust5c369682007-06-17 15:27:42 -0400446 radix_tree_tag_set(&nfsi->nfs_page_tree,
447 req->wb_index,
448 NFS_PAGE_TAG_COMMIT);
Trond Myklebustff778d02010-02-19 16:53:39 -0800449 nfsi->ncommit++;
Trond Myklebust587142f2007-07-02 09:57:54 -0400450 spin_unlock(&inode->i_lock);
Fred Isamana861a1e2011-03-23 13:27:51 +0000451 pnfs_mark_request_commit(req, lseg);
Christoph Lameterfd39fc82006-06-30 01:55:40 -0700452 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Peter Zijlstrac9e51e42007-10-16 23:25:47 -0700453 inc_bdi_stat(req->wb_page->mapping->backing_dev_info, BDI_RECLAIMABLE);
Trond Myklebusta1803042006-12-05 00:45:19 -0500454 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400456
Trond Myklebuste468bae2008-06-13 13:25:22 -0400457static int
458nfs_clear_request_commit(struct nfs_page *req)
459{
460 struct page *page = req->wb_page;
461
462 if (test_and_clear_bit(PG_CLEAN, &(req)->wb_flags)) {
463 dec_zone_page_state(page, NR_UNSTABLE_NFS);
464 dec_bdi_stat(page->mapping->backing_dev_info, BDI_RECLAIMABLE);
465 return 1;
466 }
467 return 0;
468}
469
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400470static inline
471int nfs_write_need_commit(struct nfs_write_data *data)
472{
Fred Isaman465d5242011-03-23 13:27:44 +0000473 if (data->verf.committed == NFS_DATA_SYNC)
474 return data->lseg == NULL;
475 else
476 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400477}
478
479static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000480int nfs_reschedule_unstable_write(struct nfs_page *req,
481 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400482{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400483 if (test_and_clear_bit(PG_NEED_COMMIT, &req->wb_flags)) {
Fred Isamana861a1e2011-03-23 13:27:51 +0000484 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400485 return 1;
486 }
487 if (test_and_clear_bit(PG_NEED_RESCHED, &req->wb_flags)) {
Fred6d884e82008-03-19 11:24:38 -0400488 nfs_mark_request_dirty(req);
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400489 return 1;
490 }
491 return 0;
492}
493#else
494static inline void
Fred Isamana861a1e2011-03-23 13:27:51 +0000495nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400496{
497}
498
Trond Myklebuste468bae2008-06-13 13:25:22 -0400499static inline int
500nfs_clear_request_commit(struct nfs_page *req)
501{
502 return 0;
503}
504
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400505static inline
506int nfs_write_need_commit(struct nfs_write_data *data)
507{
508 return 0;
509}
510
511static inline
Fred Isamana861a1e2011-03-23 13:27:51 +0000512int nfs_reschedule_unstable_write(struct nfs_page *req,
513 struct nfs_write_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400514{
515 return 0;
516}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517#endif
518
Trond Myklebust47c62562009-03-16 08:13:41 -0400519#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400520static int
521nfs_need_commit(struct nfs_inode *nfsi)
522{
523 return radix_tree_tagged(&nfsi->nfs_page_tree, NFS_PAGE_TAG_COMMIT);
524}
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526/*
527 * nfs_scan_commit - Scan an inode for commit requests
528 * @inode: NFS inode to scan
529 * @dst: destination list
530 * @idx_start: lower bound of page->index to scan.
531 * @npages: idx_start + npages sets the upper bound to scan.
532 *
533 * Moves requests from the inode's 'commit' request list.
534 * The requests are *not* checked to ensure that they form a contiguous set.
535 */
536static int
Trond Myklebustca52fec2007-04-17 17:22:13 -0400537nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
539 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebustff778d02010-02-19 16:53:39 -0800540 int ret;
Trond Myklebust3da28eb2005-06-22 17:16:31 +0000541
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400542 if (!nfs_need_commit(nfsi))
543 return 0;
544
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000545 spin_lock(&inode->i_lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800546 ret = nfs_scan_list(nfsi, dst, idx_start, npages, NFS_PAGE_TAG_COMMIT);
547 if (ret > 0)
548 nfsi->ncommit -= ret;
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000549 spin_unlock(&inode->i_lock);
550
Trond Myklebust2928db12010-02-19 17:03:18 -0800551 if (nfs_need_commit(NFS_I(inode)))
552 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Dave Chinner0d88f6e2011-04-12 19:18:08 +1000553
Trond Myklebustff778d02010-02-19 16:53:39 -0800554 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500556#else
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400557static inline int nfs_need_commit(struct nfs_inode *nfsi)
558{
559 return 0;
560}
561
Trond Myklebustca52fec2007-04-17 17:22:13 -0400562static inline int nfs_scan_commit(struct inode *inode, struct list_head *dst, pgoff_t idx_start, unsigned int npages)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500563{
564 return 0;
565}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566#endif
567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400569 * Search for an existing write request, and attempt to update
570 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400572 * If the attempt fails, then the existing request is flushed out
573 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400575static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
576 struct page *page,
577 unsigned int offset,
578 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400580 struct nfs_page *req;
581 unsigned int rqend;
582 unsigned int end;
583 int error;
584
585 if (!PagePrivate(page))
586 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400589 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 for (;;) {
Trond Myklebust277459d2006-12-05 00:35:35 -0500592 req = nfs_page_find_request_locked(page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400593 if (req == NULL)
594 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500595
Trond Myklebuste7d39062008-06-13 12:12:32 -0400596 rqend = req->wb_offset + req->wb_bytes;
597 /*
598 * Tell the caller to flush out the request if
599 * the offsets are non-contiguous.
600 * Note: nfs_flush_incompatible() will already
601 * have flushed out requests having wrong owners.
602 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400603 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400604 || end < req->wb_offset)
605 goto out_flushme;
606
607 if (nfs_set_page_tag_locked(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
Trond Myklebuste7d39062008-06-13 12:12:32 -0400610 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400611 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400612 error = nfs_wait_on_request(req);
613 nfs_release_request(req);
614 if (error != 0)
615 goto out_err;
616 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Trond Myklebustff778d02010-02-19 16:53:39 -0800619 if (nfs_clear_request_commit(req) &&
Fred Isamana861a1e2011-03-23 13:27:51 +0000620 radix_tree_tag_clear(&NFS_I(inode)->nfs_page_tree,
621 req->wb_index, NFS_PAGE_TAG_COMMIT) != NULL) {
Trond Myklebustff778d02010-02-19 16:53:39 -0800622 NFS_I(inode)->ncommit--;
Fred Isamana861a1e2011-03-23 13:27:51 +0000623 pnfs_clear_request_commit(req);
624 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 /* Okay, the request matches. Update the region */
627 if (offset < req->wb_offset) {
628 req->wb_offset = offset;
629 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (end > rqend)
632 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400633 else
634 req->wb_bytes = rqend - req->wb_offset;
635out_unlock:
636 spin_unlock(&inode->i_lock);
637 return req;
638out_flushme:
639 spin_unlock(&inode->i_lock);
640 nfs_release_request(req);
641 error = nfs_wb_page(inode, page);
642out_err:
643 return ERR_PTR(error);
644}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Trond Myklebuste7d39062008-06-13 12:12:32 -0400646/*
647 * Try to update an existing write request, or create one if there is none.
648 *
649 * Note: Should always be called with the Page Lock held to prevent races
650 * if we have to add a new request. Also assumes that the caller has
651 * already called nfs_flush_incompatible() if necessary.
652 */
653static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
654 struct page *page, unsigned int offset, unsigned int bytes)
655{
656 struct inode *inode = page->mapping->host;
657 struct nfs_page *req;
658 int error;
659
660 req = nfs_try_to_update_request(inode, page, offset, bytes);
661 if (req != NULL)
662 goto out;
663 req = nfs_create_request(ctx, inode, page, offset, bytes);
664 if (IS_ERR(req))
665 goto out;
666 error = nfs_inode_add_request(inode, req);
667 if (error != 0) {
668 nfs_release_request(req);
669 req = ERR_PTR(error);
670 }
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400671out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400672 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673}
674
Trond Myklebuste7d39062008-06-13 12:12:32 -0400675static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
676 unsigned int offset, unsigned int count)
677{
678 struct nfs_page *req;
679
680 req = nfs_setup_write_request(ctx, page, offset, count);
681 if (IS_ERR(req))
682 return PTR_ERR(req);
683 /* Update file length */
684 nfs_grow_file(page, offset, count);
685 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400686 nfs_mark_request_dirty(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400687 nfs_clear_page_tag_locked(req);
688 return 0;
689}
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691int nfs_flush_incompatible(struct file *file, struct page *page)
692{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400693 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500695 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 /*
697 * Look for a request corresponding to this page. If there
698 * is one, and it belongs to another file, we flush it out
699 * before we try to copy anything into the page. Do this
700 * due to the lack of an ACCESS-type call in NFSv2.
701 * Also do the same if we find a request from an existing
702 * dropped page.
703 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500704 do {
705 req = nfs_page_find_request(page);
706 if (req == NULL)
707 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400708 do_flush = req->wb_page != page || req->wb_context != ctx ||
709 req->wb_lock_context->lockowner != current->files ||
710 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500712 if (!do_flush)
713 return 0;
714 status = nfs_wb_page(page->mapping->host, page);
715 } while (status == 0);
716 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
719/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500720 * If the page cache is marked as unsafe or invalid, then we can't rely on
721 * the PageUptodate() flag. In this case, we will need to turn off
722 * write optimisations that depend on the page contents being correct.
723 */
724static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
725{
726 return PageUptodate(page) &&
727 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
728}
729
730/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 * Update and possibly write a cached page of an NFS file.
732 *
733 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
734 * things with a page scheduled for an RPC call (e.g. invalidate it).
735 */
736int nfs_updatepage(struct file *file, struct page *page,
737 unsigned int offset, unsigned int count)
738{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400739 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 int status = 0;
742
Chuck Lever91d5b472006-03-20 13:44:14 -0500743 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
744
Chuck Lever48186c72008-06-11 17:56:05 -0400745 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800746 file->f_path.dentry->d_parent->d_name.name,
747 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400748 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500751 * is up to date, it may be more efficient to extend the write
752 * to cover the entire page in order to avoid fragmentation
753 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500755 if (nfs_write_pageuptodate(page, inode) &&
756 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100757 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500758 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
761
Trond Myklebuste21195a2006-12-05 00:35:39 -0500762 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400763 if (status < 0)
764 nfs_set_pageerror(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Chuck Lever48186c72008-06-11 17:56:05 -0400766 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 return status;
769}
770
Fred Isamana861a1e2011-03-23 13:27:51 +0000771static void nfs_writepage_release(struct nfs_page *req,
772 struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400774 struct page *page = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Fred Isamana861a1e2011-03-23 13:27:51 +0000776 if (PageError(req->wb_page) || !nfs_reschedule_unstable_write(req, data))
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400777 nfs_inode_remove_request(req);
Trond Myklebust9fd367f2007-06-17 15:10:24 -0400778 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400779 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780}
781
Trond Myklebust3ff75762007-07-14 15:40:00 -0400782static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783{
784 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
785 case FLUSH_HIGHPRI:
786 return RPC_PRIORITY_HIGH;
787 case FLUSH_LOWPRI:
788 return RPC_PRIORITY_LOW;
789 }
790 return RPC_PRIORITY_NORMAL;
791}
792
Fred Isamana69aef12011-03-03 15:13:47 +0000793int nfs_initiate_write(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000794 struct rpc_clnt *clnt,
795 const struct rpc_call_ops *call_ops,
796 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
Andy Adamsond138d5d2011-03-03 15:13:41 +0000798 struct inode *inode = data->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400799 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400800 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400801 struct rpc_message msg = {
802 .rpc_argp = &data->args,
803 .rpc_resp = &data->res,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000804 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400805 };
Trond Myklebust84115e12007-07-14 15:39:59 -0400806 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +0000807 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -0400808 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400809 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -0400810 .callback_ops = call_ops,
811 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -0500812 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -0400813 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -0400814 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -0400815 };
Trond Myklebust2c61be02010-04-09 19:54:50 -0400816 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Andy Adamsond138d5d2011-03-03 15:13:41 +0000818 /* Set up the initial task struct. */
819 NFS_PROTO(inode)->write_setup(data, &msg);
820
821 dprintk("NFS: %5u initiated write call "
822 "(req %s/%lld, %u bytes @ offset %llu)\n",
823 data->task.tk_pid,
824 inode->i_sb->s_id,
825 (long long)NFS_FILEID(inode),
826 data->args.count,
827 (unsigned long long)data->args.offset);
828
829 task = rpc_run_task(&task_setup_data);
830 if (IS_ERR(task)) {
831 ret = PTR_ERR(task);
832 goto out;
833 }
834 if (how & FLUSH_SYNC) {
835 ret = rpc_wait_for_completion_task(task);
836 if (ret == 0)
837 ret = task->tk_status;
838 }
839 rpc_put_task(task);
840out:
841 return ret;
842}
Fred Isamana69aef12011-03-03 15:13:47 +0000843EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +0000844
845/*
846 * Set up the argument/result storage required for the RPC call.
847 */
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400848static void nfs_write_rpcsetup(struct nfs_page *req,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000849 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000850 unsigned int count, unsigned int offset,
851 int how)
852{
Al Viro3d4ff432011-06-22 18:40:12 -0400853 struct inode *inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 /* Set up the RPC argument and reply structs
856 * NB: take care not to mess about with data->commit et al. */
857
858 data->req = req;
Al Viro3d4ff432011-06-22 18:40:12 -0400859 data->inode = inode = req->wb_context->dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000860 data->cred = req->wb_context->cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
862 data->args.fh = NFS_FH(inode);
863 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -0400864 /* pnfs_set_layoutcommit needs this */
865 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 data->args.pgbase = req->wb_pgbase + offset;
867 data->args.pages = data->pagevec;
868 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500869 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400870 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400871 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -0400872 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
873 case 0:
874 break;
875 case FLUSH_COND_STABLE:
876 if (nfs_need_commit(NFS_I(inode)))
877 break;
878 default:
879 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 data->res.fattr = &data->fattr;
883 data->res.count = count;
884 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400885 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400886}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400888static int nfs_do_write(struct nfs_write_data *data,
889 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400890 int how)
891{
Stephen Rothwell5f00bcb2011-07-25 13:59:46 -0400892 struct inode *inode = data->args.context->dentry->d_inode;
Andy Adamson0382b742011-03-03 15:13:45 +0000893
Andy Adamsond138d5d2011-03-03 15:13:41 +0000894 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
896
Trond Myklebust275acaa2011-07-12 13:42:02 -0400897static int nfs_do_multiple_writes(struct list_head *head,
898 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -0400899 int how)
900{
901 struct nfs_write_data *data;
902 int ret = 0;
903
904 while (!list_empty(head)) {
905 int ret2;
906
907 data = list_entry(head->next, struct nfs_write_data, list);
908 list_del_init(&data->list);
909
Trond Myklebustdce81292011-07-13 15:59:19 -0400910 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400911 if (ret == 0)
912 ret = ret2;
913 }
914 return ret;
915}
916
Fred6d884e82008-03-19 11:24:38 -0400917/* If a nfs_flush_* function fails, it should remove reqs from @head and
918 * call this on each, which will prepare them to be retried on next
919 * writeback using standard nfs.
920 */
921static void nfs_redirty_request(struct nfs_page *req)
922{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400923 struct page *page = req->wb_page;
924
Fred6d884e82008-03-19 11:24:38 -0400925 nfs_mark_request_dirty(req);
Fred6d884e82008-03-19 11:24:38 -0400926 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400927 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930/*
931 * Generate multiple small requests to write out a single
932 * contiguous dirty area on one page.
933 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400934static int nfs_flush_multi(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
Fred Isamanc76069b2011-03-03 15:13:48 +0000936 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 struct page *page = req->wb_page;
938 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -0400939 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700940 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400942 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 nfs_list_remove_request(req);
945
Trond Myklebustb31268a2011-03-21 17:02:00 -0400946 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
947 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
948 desc->pg_count > wsize))
949 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
950
951
Trond Myklebust275acaa2011-07-12 13:42:02 -0400952 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000953 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700954 do {
955 size_t len = min(nbytes, wsize);
956
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400957 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 if (!data)
959 goto out_bad;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400960 data->pagevec[0] = page;
Trond Myklebustf13c3622011-09-12 11:47:53 -0400961 nfs_write_rpcsetup(req, data, len, offset, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -0400962 list_add(&data->list, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700964 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -0400965 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700966 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 atomic_set(&req->wb_complete, requests);
Trond Myklebust50828d72011-07-12 13:42:02 -0400968 desc->pg_rpc_callops = &nfs_write_partial_ops;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400969 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971out_bad:
Trond Myklebust275acaa2011-07-12 13:42:02 -0400972 while (!list_empty(res)) {
973 data = list_entry(res->next, struct nfs_write_data, list);
Trond Myklebust6e4efd52011-07-12 13:42:02 -0400974 list_del(&data->list);
Fred Isaman0da2a4ac2011-01-19 14:18:50 -0500975 nfs_writedata_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500977 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return -ENOMEM;
979}
980
981/*
982 * Create an RPC task for the given write request and kick it.
983 * The page must have been locked by the caller.
984 *
985 * It may happen that the page we're passed is not marked dirty.
986 * This is the case if nfs_updatepage detects a conflicting request
987 * that has been written but not committed.
988 */
Trond Myklebust275acaa2011-07-12 13:42:02 -0400989static int nfs_flush_one(struct nfs_pageio_descriptor *desc, struct list_head *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990{
991 struct nfs_page *req;
992 struct page **pages;
993 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000994 struct list_head *head = &desc->pg_list;
Peng Tao3b609182011-07-15 03:33:42 -0400995 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
Fred Isamanc76069b2011-03-03 15:13:48 +0000997 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
998 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +0000999 if (!data) {
1000 while (!list_empty(head)) {
1001 req = nfs_list_entry(head->next);
1002 nfs_list_remove_request(req);
1003 nfs_redirty_request(req);
1004 }
1005 ret = -ENOMEM;
1006 goto out;
1007 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 while (!list_empty(head)) {
1010 req = nfs_list_entry(head->next);
1011 nfs_list_remove_request(req);
1012 nfs_list_add_request(req, &data->pages);
1013 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
1016 req = nfs_list_entry(data->pages.next);
1017
Trond Myklebustb31268a2011-03-21 17:02:00 -04001018 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
1019 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit))
1020 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1021
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /* Set up the argument struct */
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001023 nfs_write_rpcsetup(req, data, desc->pg_count, 0, desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001024 list_add(&data->list, res);
Trond Myklebust50828d72011-07-12 13:42:02 -04001025 desc->pg_rpc_callops = &nfs_write_full_ops;
Fred Isaman44b83792011-03-03 15:13:44 +00001026out:
Fred Isaman44b83792011-03-03 15:13:44 +00001027 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028}
1029
Trond Myklebustdce81292011-07-13 15:59:19 -04001030int nfs_generic_flush(struct nfs_pageio_descriptor *desc, struct list_head *head)
1031{
1032 if (desc->pg_bsize < PAGE_CACHE_SIZE)
1033 return nfs_flush_multi(desc, head);
1034 return nfs_flush_one(desc, head);
1035}
1036
1037static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001038{
Trond Myklebust275acaa2011-07-12 13:42:02 -04001039 LIST_HEAD(head);
1040 int ret;
1041
Trond Myklebustdce81292011-07-13 15:59:19 -04001042 ret = nfs_generic_flush(desc, &head);
Trond Myklebust50828d72011-07-12 13:42:02 -04001043 if (ret == 0)
1044 ret = nfs_do_multiple_writes(&head, desc->pg_rpc_callops,
Trond Myklebustdce81292011-07-13 15:59:19 -04001045 desc->pg_ioflags);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001046 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001047}
Trond Myklebust1751c362011-06-10 13:30:23 -04001048
1049static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1050 .pg_test = nfs_generic_pg_test,
1051 .pg_doio = nfs_generic_pg_writepages,
1052};
1053
Trond Myklebust1f945352011-07-13 15:59:57 -04001054static void nfs_pageio_init_write_mds(struct nfs_pageio_descriptor *pgio,
Trond Myklebust1751c362011-06-10 13:30:23 -04001055 struct inode *inode, int ioflags)
1056{
1057 nfs_pageio_init(pgio, inode, &nfs_pageio_write_ops,
1058 NFS_SERVER(inode)->wsize, ioflags);
1059}
1060
Trond Myklebustdce81292011-07-13 15:59:19 -04001061void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1062{
1063 pgio->pg_ops = &nfs_pageio_write_ops;
1064 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1065}
Trond Myklebust1f945352011-07-13 15:59:57 -04001066EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001067
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001068static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1069 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
Trond Myklebust1751c362011-06-10 13:30:23 -04001071 if (!pnfs_pageio_init_write(pgio, inode, ioflags))
1072 nfs_pageio_init_write_mds(pgio, inode, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073}
1074
1075/*
1076 * Handle a write reply that flushed part of a page.
1077 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001078static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001080 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Chuck Lever48186c72008-06-11 17:56:05 -04001082 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1083 task->tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001084 data->req->wb_context->dentry->d_inode->i_sb->s_id,
Chuck Lever48186c72008-06-11 17:56:05 -04001085 (long long)
Al Viro3d4ff432011-06-22 18:40:12 -04001086 NFS_FILEID(data->req->wb_context->dentry->d_inode),
Chuck Lever48186c72008-06-11 17:56:05 -04001087 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001089 nfs_writeback_done(task, data);
1090}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001091
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001092static void nfs_writeback_release_partial(void *calldata)
1093{
1094 struct nfs_write_data *data = calldata;
1095 struct nfs_page *req = data->req;
1096 struct page *page = req->wb_page;
1097 int status = data->task.tk_status;
1098
1099 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001100 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001101 nfs_context_set_write_error(req->wb_context, status);
1102 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001103 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001106 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001107 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001108
Trond Myklebust587142f2007-07-02 09:57:54 -04001109 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001110 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1111 /* Do nothing we need to resend the writes */
1112 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1113 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1114 dprintk(" defer commit\n");
1115 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1116 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1117 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1118 dprintk(" server reboot detected\n");
1119 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001120 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001121 } else
1122 dprintk(" OK\n");
1123
1124out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001126 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001127 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128}
1129
Andy Adamsondef6ed72009-04-01 09:22:26 -04001130#if defined(CONFIG_NFS_V4_1)
1131void nfs_write_prepare(struct rpc_task *task, void *calldata)
1132{
1133 struct nfs_write_data *data = calldata;
Andy Adamsondef6ed72009-04-01 09:22:26 -04001134
Trond Myklebust035168a2010-06-16 09:52:26 -04001135 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
1136 &data->args.seq_args,
Andy Adamsondef6ed72009-04-01 09:22:26 -04001137 &data->res.seq_res, 1, task))
1138 return;
1139 rpc_call_start(task);
1140}
1141#endif /* CONFIG_NFS_V4_1 */
1142
Trond Myklebust788e7a82006-03-20 13:44:27 -05001143static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001144#if defined(CONFIG_NFS_V4_1)
1145 .rpc_call_prepare = nfs_write_prepare,
1146#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001147 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001148 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001149};
1150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151/*
1152 * Handle a write reply that flushes a whole page.
1153 *
1154 * FIXME: There is an inherent race with invalidate_inode_pages and
1155 * writebacks since the page->count is kept > 1 for as long
1156 * as the page has a write request pending.
1157 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001158static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001160 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001162 nfs_writeback_done(task, data);
1163}
1164
1165static void nfs_writeback_release_full(void *calldata)
1166{
1167 struct nfs_write_data *data = calldata;
1168 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /* Update attributes as result of writeback. */
1171 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001172 struct nfs_page *req = nfs_list_entry(data->pages.next);
1173 struct page *page = req->wb_page;
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Chuck Lever48186c72008-06-11 17:56:05 -04001177 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1178 data->task.tk_pid,
Al Viro3d4ff432011-06-22 18:40:12 -04001179 req->wb_context->dentry->d_inode->i_sb->s_id,
1180 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 req->wb_bytes,
1182 (long long)req_offset(req));
1183
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001184 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001185 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001186 nfs_context_set_write_error(req->wb_context, status);
1187 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001188 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001191 if (nfs_write_need_commit(data)) {
1192 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001193 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001194 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 goto next;
1196 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001197 dprintk(" OK\n");
1198remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001201 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001202 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001204 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
Trond Myklebust788e7a82006-03-20 13:44:27 -05001207static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001208#if defined(CONFIG_NFS_V4_1)
1209 .rpc_call_prepare = nfs_write_prepare,
1210#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001211 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001212 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001213};
1214
1215
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216/*
1217 * This function is called when the WRITE call is complete.
1218 */
Fred Isaman13602892011-02-11 15:42:38 +00001219void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 struct nfs_writeargs *argp = &data->args;
1222 struct nfs_writeres *resp = &data->res;
Andy Adamsoneedc0202009-04-01 09:22:41 -04001223 struct nfs_server *server = NFS_SERVER(data->inode);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001224 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Chuck Levera3f565b2007-01-31 12:14:01 -05001226 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 task->tk_pid, task->tk_status);
1228
Chuck Leverf551e442006-09-20 14:33:04 -04001229 /*
1230 * ->write_done will attempt to use post-op attributes to detect
1231 * conflicting writes by other clients. A strict interpretation
1232 * of close-to-open would allow us to continue caching even if
1233 * another writer had changed the file, but some applications
1234 * depend on tighter cache coherency when writing.
1235 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001236 status = NFS_PROTO(data->inode)->write_done(task, data);
1237 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001238 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001239 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1242 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1243 /* We tried a write call, but the server did not
1244 * commit data to stable storage even though we
1245 * requested it.
1246 * Note: There is a known bug in Tru64 < 5.0 in which
1247 * the server reports NFS_DATA_SYNC, but performs
1248 * NFS_FILE_SYNC. We therefore implement this checking
1249 * as a dprintk() in order to avoid filling syslog.
1250 */
1251 static unsigned long complain;
1252
Fred Isamana69aef12011-03-03 15:13:47 +00001253 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001255 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 " (committed = %d) != (stable = %d)\n",
Andy Adamsoneedc0202009-04-01 09:22:41 -04001257 server->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 resp->verf->committed, argp->stable);
1259 complain = jiffies + 300 * HZ;
1260 }
1261 }
1262#endif
1263 /* Is this a short write? */
1264 if (task->tk_status >= 0 && resp->count < argp->count) {
1265 static unsigned long complain;
1266
Chuck Lever91d5b472006-03-20 13:44:14 -05001267 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 /* Has the server at least made some progress? */
1270 if (resp->count != 0) {
1271 /* Was this an NFSv2 write or an NFSv3 stable write? */
1272 if (resp->verf->committed != NFS_UNSTABLE) {
1273 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001274 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 argp->offset += resp->count;
1276 argp->pgbase += resp->count;
1277 argp->count -= resp->count;
1278 } else {
1279 /* Resend as a stable write in order to avoid
1280 * headaches in the case of a server crash.
1281 */
1282 argp->stable = NFS_FILE_SYNC;
1283 }
Trond Myklebust0110ee12009-12-07 09:00:24 -05001284 nfs_restart_rpc(task, server->nfs_client);
Fred Isaman13602892011-02-11 15:42:38 +00001285 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
1287 if (time_before(complain, jiffies)) {
1288 printk(KERN_WARNING
1289 "NFS: Server wrote zero bytes, expected %u.\n",
1290 argp->count);
1291 complain = jiffies + 300 * HZ;
1292 }
1293 /* Can't do anything about it except throw an error. */
1294 task->tk_status = -EIO;
1295 }
Fred Isaman13602892011-02-11 15:42:38 +00001296 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297}
1298
1299
1300#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001301static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1302{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001303 int ret;
1304
Trond Myklebust71d0a612010-04-22 15:35:57 -04001305 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1306 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001307 if (!may_wait)
1308 return 0;
1309 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1310 NFS_INO_COMMIT,
1311 nfs_wait_bit_killable,
1312 TASK_KILLABLE);
1313 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001314}
1315
Fred Isamane0c2b382011-03-23 13:27:53 +00001316void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001317{
1318 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1319 smp_mb__after_clear_bit();
1320 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1321}
Fred Isamane0c2b382011-03-23 13:27:53 +00001322EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001323
Fred Isamane0c2b382011-03-23 13:27:53 +00001324void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
Trond Myklebust383ba712008-02-19 20:04:20 -05001326 struct nfs_write_data *wdata = data;
1327
Fred Isaman988b6dc2011-03-23 13:27:52 +00001328 put_lseg(wdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -05001329 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 nfs_commit_free(wdata);
1331}
Fred Isamane0c2b382011-03-23 13:27:53 +00001332EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Fred Isamane0c2b382011-03-23 13:27:53 +00001334int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001335 const struct rpc_call_ops *call_ops,
1336 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
Trond Myklebust07737692007-10-25 18:42:54 -04001338 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001339 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001340 struct rpc_message msg = {
1341 .rpc_argp = &data->args,
1342 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001343 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001344 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001345 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001346 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001347 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001348 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001349 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001350 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001351 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001352 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001353 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001354 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001355 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001356 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Chuck Levera3f565b2007-01-31 12:14:01 -05001358 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001359
Trond Myklebust07737692007-10-25 18:42:54 -04001360 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001361 if (IS_ERR(task))
1362 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001363 if (how & FLUSH_SYNC)
1364 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001365 rpc_put_task(task);
1366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367}
Fred Isamane0c2b382011-03-23 13:27:53 +00001368EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369
1370/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001371 * Set up the argument/result storage required for the RPC call.
1372 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001373void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001374 struct list_head *head,
1375 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001376{
1377 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001378 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001379
1380 /* Set up the RPC argument and reply structs
1381 * NB: take care not to mess about with data->commit et al. */
1382
1383 list_splice_init(head, &data->pages);
1384
1385 data->inode = inode;
1386 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001387 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001388 data->mds_ops = &nfs_commit_ops;
1389
1390 data->args.fh = NFS_FH(data->inode);
1391 /* Note: we always request a commit of the entire inode */
1392 data->args.offset = 0;
1393 data->args.count = 0;
1394 data->args.context = get_nfs_open_context(first->wb_context);
1395 data->res.count = 0;
1396 data->res.fattr = &data->fattr;
1397 data->res.verf = &data->verf;
1398 nfs_fattr_init(&data->fattr);
1399}
Fred Isamane0c2b382011-03-23 13:27:53 +00001400EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001401
Fred Isamane0c2b382011-03-23 13:27:53 +00001402void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001403 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001404{
1405 struct nfs_page *req;
1406
1407 while (!list_empty(page_list)) {
1408 req = nfs_list_entry(page_list->next);
1409 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001410 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001411 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1412 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1413 BDI_RECLAIMABLE);
1414 nfs_clear_page_tag_locked(req);
1415 }
1416}
Fred Isamane0c2b382011-03-23 13:27:53 +00001417EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001418
Fred Isaman9ace33c2011-03-23 13:27:45 +00001419/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 * Commit dirty pages
1421 */
1422static int
Chuck Lever40859d72005-11-30 18:09:02 -05001423nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001427 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429 if (!data)
1430 goto out_bad;
1431
1432 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001433 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001434 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001436 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001437 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 return -ENOMEM;
1439}
1440
1441/*
1442 * COMMIT call returned
1443 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001444static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001446 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Chuck Levera3f565b2007-01-31 12:14:01 -05001448 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 task->tk_pid, task->tk_status);
1450
Trond Myklebust788e7a82006-03-20 13:44:27 -05001451 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001452 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001453}
1454
Fred Isamane0c2b382011-03-23 13:27:53 +00001455void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001456{
Fred Isaman5917ce82011-03-23 13:27:48 +00001457 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001458 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 while (!list_empty(&data->pages)) {
1461 req = nfs_list_entry(data->pages.next);
1462 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -04001463 nfs_clear_request_commit(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
Chuck Lever48186c72008-06-11 17:56:05 -04001465 dprintk("NFS: commit (%s/%lld %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001466 req->wb_context->dentry->d_sb->s_id,
1467 (long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 req->wb_bytes,
1469 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001470 if (status < 0) {
1471 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001473 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 goto next;
1475 }
1476
1477 /* Okay, COMMIT succeeded, apparently. Check the verifier
1478 * returned by the server against all stored verfs. */
1479 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1480 /* We have a match */
1481 nfs_inode_remove_request(req);
1482 dprintk(" OK\n");
1483 goto next;
1484 }
1485 /* We have a mismatch. Write the page again */
1486 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001487 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001489 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001491}
Fred Isamane0c2b382011-03-23 13:27:53 +00001492EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001493
1494static void nfs_commit_release(void *calldata)
1495{
1496 struct nfs_write_data *data = calldata;
1497
1498 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001499 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001500 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001502
1503static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001504#if defined(CONFIG_NFS_V4_1)
1505 .rpc_call_prepare = nfs_write_prepare,
1506#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001507 .rpc_call_done = nfs_commit_done,
1508 .rpc_release = nfs_commit_release,
1509};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Trond Myklebustb608b282010-07-30 15:31:54 -04001511int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001514 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001515 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Trond Myklebustb8413f92011-03-21 15:37:01 -04001517 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1518 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001519 goto out_mark_dirty;
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001520 res = nfs_scan_commit(inode, &head, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001522 int error;
1523
1524 error = pnfs_commit_list(inode, &head, how);
1525 if (error == PNFS_NOT_ATTEMPTED)
1526 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001527 if (error < 0)
1528 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001529 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001530 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001531 error = wait_on_bit(&NFS_I(inode)->flags,
1532 NFS_INO_COMMIT,
1533 nfs_wait_bit_killable,
1534 TASK_KILLABLE);
1535 if (error < 0)
1536 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001537 } else
1538 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001539 return res;
1540 /* Note: If we exit without ensuring that the commit is complete,
1541 * we must mark the inode as dirty. Otherwise, future calls to
1542 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1543 * that the data is on the disk.
1544 */
1545out_mark_dirty:
1546 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 return res;
1548}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001549
1550static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1551{
Trond Myklebust420e3642010-02-19 17:00:02 -08001552 struct nfs_inode *nfsi = NFS_I(inode);
1553 int flags = FLUSH_SYNC;
1554 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001555
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001556 if (wbc->sync_mode == WB_SYNC_NONE) {
1557 /* Don't commit yet if this is a non-blocking flush and there
1558 * are a lot of outstanding writes for this mapping.
1559 */
1560 if (nfsi->ncommit <= (nfsi->npages >> 1))
1561 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001562
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001563 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001564 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001565 }
1566
Trond Myklebust420e3642010-02-19 17:00:02 -08001567 ret = nfs_commit_inode(inode, flags);
1568 if (ret >= 0) {
1569 if (wbc->sync_mode == WB_SYNC_NONE) {
1570 if (ret < wbc->nr_to_write)
1571 wbc->nr_to_write -= ret;
1572 else
1573 wbc->nr_to_write = 0;
1574 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001575 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001576 }
1577out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001578 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1579 return ret;
1580}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001581#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001582static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1583{
1584 return 0;
1585}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586#endif
1587
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001588int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1589{
Andy Adamson863a3c62011-03-23 13:27:54 +00001590 int ret;
1591
1592 ret = nfs_commit_unstable_pages(inode, wbc);
1593 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001594 int status;
1595 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001596
Wu Fengguang846d5a02011-05-05 21:10:38 -06001597 if (wbc->sync_mode == WB_SYNC_NONE)
Andy Adamsonef311532011-03-12 02:58:10 -05001598 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001599
1600 status = pnfs_layoutcommit_inode(inode, sync);
1601 if (status < 0)
1602 return status;
1603 }
1604 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001605}
1606
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001607/*
1608 * flush the inode to disk.
1609 */
1610int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001611{
1612 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001613 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001614 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001615 .range_start = 0,
1616 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001617 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001618
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001619 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001620}
1621
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001622int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1623{
1624 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001625 int ret = 0;
1626
1627 BUG_ON(!PageLocked(page));
1628 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001629 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001630 req = nfs_page_find_request(page);
1631 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001632 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001633 if (nfs_lock_request_dontget(req)) {
1634 nfs_inode_remove_request(req);
1635 /*
1636 * In case nfs_inode_remove_request has marked the
1637 * page as being dirty
1638 */
1639 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1640 nfs_unlock_request(req);
1641 break;
1642 }
1643 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001644 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001645 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001646 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001647 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001648 return ret;
1649}
1650
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001651/*
1652 * Write back all requests on one page - we do this before reading it.
1653 */
1654int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001655{
1656 loff_t range_start = page_offset(page);
1657 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001658 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001659 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001660 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001661 .range_start = range_start,
1662 .range_end = range_end,
1663 };
1664 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001665
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001666 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001667 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001668 if (clear_page_dirty_for_io(page)) {
1669 ret = nfs_writepage_locked(page, &wbc);
1670 if (ret < 0)
1671 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001672 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001673 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001674 if (!PagePrivate(page))
1675 break;
1676 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001677 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001678 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001679 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001680 return 0;
1681out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001682 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001683}
1684
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001685#ifdef CONFIG_MIGRATION
1686int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1687 struct page *page)
1688{
1689 struct nfs_page *req;
1690 int ret;
1691
Trond Myklebust7549ad52010-02-08 09:32:34 -05001692 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001693
Trond Myklebustcfb506e2010-07-30 15:31:57 -04001694 req = nfs_find_and_lock_request(page, false);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001695 ret = PTR_ERR(req);
1696 if (IS_ERR(req))
1697 goto out;
1698
1699 ret = migrate_page(mapping, newpage, page);
1700 if (!req)
1701 goto out;
1702 if (ret)
1703 goto out_unlock;
1704 page_cache_get(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001705 spin_lock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001706 req->wb_page = newpage;
1707 SetPagePrivate(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001708 set_page_private(newpage, (unsigned long)req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001709 ClearPagePrivate(page);
1710 set_page_private(page, 0);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001711 spin_unlock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001712 page_cache_release(page);
1713out_unlock:
1714 nfs_clear_page_tag_locked(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001715out:
1716 return ret;
1717}
1718#endif
1719
David Howellsf7b422b2006-06-09 09:34:33 -04001720int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721{
1722 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1723 sizeof(struct nfs_write_data),
1724 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001725 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (nfs_wdata_cachep == NULL)
1727 return -ENOMEM;
1728
Matthew Dobson93d23412006-03-26 01:37:50 -08001729 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1730 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 if (nfs_wdata_mempool == NULL)
1732 return -ENOMEM;
1733
Matthew Dobson93d23412006-03-26 01:37:50 -08001734 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1735 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 if (nfs_commit_mempool == NULL)
1737 return -ENOMEM;
1738
Peter Zijlstra89a09142007-03-16 13:38:26 -08001739 /*
1740 * NFS congestion size, scale with available memory.
1741 *
1742 * 64MB: 8192k
1743 * 128MB: 11585k
1744 * 256MB: 16384k
1745 * 512MB: 23170k
1746 * 1GB: 32768k
1747 * 2GB: 46340k
1748 * 4GB: 65536k
1749 * 8GB: 92681k
1750 * 16GB: 131072k
1751 *
1752 * This allows larger machines to have larger/more transfers.
1753 * Limit the default to 256M
1754 */
1755 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1756 if (nfs_congestion_kb > 256*1024)
1757 nfs_congestion_kb = 256*1024;
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 return 0;
1760}
1761
David Brownell266bee82006-06-27 12:59:15 -07001762void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763{
1764 mempool_destroy(nfs_commit_mempool);
1765 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001766 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767}
1768