blob: e4cbc11a74ab56238687c5445989177a7714dd5a [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 Myklebust1ae88b22009-08-12 09:12:30 -0400100static void 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{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400412 struct inode *inode = req->wb_context->path.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{
Trond Myklebust88be9f92007-06-05 10:42:27 -0400441 struct inode *inode = req->wb_context->path.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);
Trond Myklebustb80c3cb2010-04-09 19:07:07 -0400683 nfs_mark_request_dirty(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400684 /* Update file length */
685 nfs_grow_file(page, offset, count);
686 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400687 nfs_mark_request_dirty(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400688 nfs_clear_page_tag_locked(req);
689 return 0;
690}
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692int nfs_flush_incompatible(struct file *file, struct page *page)
693{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400694 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500696 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 /*
698 * Look for a request corresponding to this page. If there
699 * is one, and it belongs to another file, we flush it out
700 * before we try to copy anything into the page. Do this
701 * due to the lack of an ACCESS-type call in NFSv2.
702 * Also do the same if we find a request from an existing
703 * dropped page.
704 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500705 do {
706 req = nfs_page_find_request(page);
707 if (req == NULL)
708 return 0;
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400709 do_flush = req->wb_page != page || req->wb_context != ctx ||
710 req->wb_lock_context->lockowner != current->files ||
711 req->wb_lock_context->pid != current->tgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500713 if (!do_flush)
714 return 0;
715 status = nfs_wb_page(page->mapping->host, page);
716 } while (status == 0);
717 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
720/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500721 * If the page cache is marked as unsafe or invalid, then we can't rely on
722 * the PageUptodate() flag. In this case, we will need to turn off
723 * write optimisations that depend on the page contents being correct.
724 */
725static int nfs_write_pageuptodate(struct page *page, struct inode *inode)
726{
727 return PageUptodate(page) &&
728 !(NFS_I(inode)->cache_validity & (NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_DATA));
729}
730
731/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 * Update and possibly write a cached page of an NFS file.
733 *
734 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
735 * things with a page scheduled for an RPC call (e.g. invalidate it).
736 */
737int nfs_updatepage(struct file *file, struct page *page,
738 unsigned int offset, unsigned int count)
739{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400740 struct nfs_open_context *ctx = nfs_file_open_context(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 int status = 0;
743
Chuck Lever91d5b472006-03-20 13:44:14 -0500744 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
745
Chuck Lever48186c72008-06-11 17:56:05 -0400746 dprintk("NFS: nfs_updatepage(%s/%s %d@%lld)\n",
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800747 file->f_path.dentry->d_parent->d_name.name,
748 file->f_path.dentry->d_name.name, count,
Chuck Lever48186c72008-06-11 17:56:05 -0400749 (long long)(page_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 /* If we're not using byte range locks, and we know the page
Trond Myklebust5d47a352008-02-07 17:24:07 -0500752 * is up to date, it may be more efficient to extend the write
753 * to cover the entire page in order to avoid fragmentation
754 * inefficiencies.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 */
Trond Myklebust5d47a352008-02-07 17:24:07 -0500756 if (nfs_write_pageuptodate(page, inode) &&
757 inode->i_flock == NULL &&
Christoph Hellwig6b2f3d12009-10-27 11:05:28 +0100758 !(file->f_flags & O_DSYNC)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500759 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 }
762
Trond Myklebuste21195a2006-12-05 00:35:39 -0500763 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400764 if (status < 0)
765 nfs_set_pageerror(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 */
849static int nfs_write_rpcsetup(struct nfs_page *req,
850 struct nfs_write_data *data,
851 const struct rpc_call_ops *call_ops,
852 unsigned int count, unsigned int offset,
Fred Isaman5053aa52011-03-03 15:13:43 +0000853 struct pnfs_layout_segment *lseg,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000854 int how)
855{
856 struct inode *inode = req->wb_context->path.dentry->d_inode;
857
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 /* Set up the RPC argument and reply structs
859 * NB: take care not to mess about with data->commit et al. */
860
861 data->req = req;
Trond Myklebust88be9f92007-06-05 10:42:27 -0400862 data->inode = inode = req->wb_context->path.dentry->d_inode;
Andy Adamsond138d5d2011-03-03 15:13:41 +0000863 data->cred = req->wb_context->cred;
Fred Isaman5053aa52011-03-03 15:13:43 +0000864 data->lseg = get_lseg(lseg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
866 data->args.fh = NFS_FH(inode);
867 data->args.offset = req_offset(req) + offset;
868 data->args.pgbase = req->wb_pgbase + offset;
869 data->args.pages = data->pagevec;
870 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -0500871 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -0400872 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400873 data->args.stable = NFS_UNSTABLE;
Trond Myklebustb31268a2011-03-21 17:02:00 -0400874 if (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400875 data->args.stable = NFS_DATA_SYNC;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400876 if (!nfs_need_commit(NFS_I(inode)))
Trond Myklebustbdc7f022007-07-14 15:40:00 -0400877 data->args.stable = NFS_FILE_SYNC;
878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 data->res.fattr = &data->fattr;
881 data->res.count = count;
882 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400883 nfs_fattr_init(&data->fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Andy Adamson0382b742011-03-03 15:13:45 +0000885 if (data->lseg &&
886 (pnfs_try_to_write_data(data, call_ops, how) == PNFS_ATTEMPTED))
887 return 0;
888
Andy Adamsond138d5d2011-03-03 15:13:41 +0000889 return nfs_initiate_write(data, NFS_CLIENT(inode), call_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890}
891
Fred6d884e82008-03-19 11:24:38 -0400892/* If a nfs_flush_* function fails, it should remove reqs from @head and
893 * call this on each, which will prepare them to be retried on next
894 * writeback using standard nfs.
895 */
896static void nfs_redirty_request(struct nfs_page *req)
897{
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400898 struct page *page = req->wb_page;
899
Fred6d884e82008-03-19 11:24:38 -0400900 nfs_mark_request_dirty(req);
Fred6d884e82008-03-19 11:24:38 -0400901 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400902 nfs_end_page_writeback(page);
Fred6d884e82008-03-19 11:24:38 -0400903}
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905/*
906 * Generate multiple small requests to write out a single
907 * contiguous dirty area on one page.
908 */
Fred Isamanc76069b2011-03-03 15:13:48 +0000909static int nfs_flush_multi(struct nfs_pageio_descriptor *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910{
Fred Isamanc76069b2011-03-03 15:13:48 +0000911 struct nfs_page *req = nfs_list_entry(desc->pg_list.next);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 struct page *page = req->wb_page;
913 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000914 size_t wsize = NFS_SERVER(desc->pg_inode)->wsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700915 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 int requests = 0;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400917 int ret = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000918 struct pnfs_layout_segment *lseg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 LIST_HEAD(list);
920
921 nfs_list_remove_request(req);
922
Trond Myklebustb31268a2011-03-21 17:02:00 -0400923 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
924 (desc->pg_moreio || NFS_I(desc->pg_inode)->ncommit ||
925 desc->pg_count > wsize))
926 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
927
928
Fred Isamanc76069b2011-03-03 15:13:48 +0000929 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700930 do {
931 size_t len = min(nbytes, wsize);
932
Trond Myklebust8d5658c2007-04-10 09:26:35 -0400933 data = nfs_writedata_alloc(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (!data)
935 goto out_bad;
936 list_add(&data->pages, &list);
937 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -0700938 nbytes -= len;
939 } while (nbytes != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 atomic_set(&req->wb_complete, requests);
941
Fred Isamanc76069b2011-03-03 15:13:48 +0000942 BUG_ON(desc->pg_lseg);
943 lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +0000946 nbytes = desc->pg_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 do {
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400948 int ret2;
949
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 data = list_entry(list.next, struct nfs_write_data, pages);
951 list_del_init(&data->pages);
952
953 data->pagevec[0] = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400955 if (nbytes < wsize)
956 wsize = nbytes;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400957 ret2 = nfs_write_rpcsetup(req, data, &nfs_write_partial_ops,
Fred Isamanc76069b2011-03-03 15:13:48 +0000958 wsize, offset, lseg, desc->pg_ioflags);
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400959 if (ret == 0)
960 ret = ret2;
Trond Myklebustbcb71bb2007-04-02 18:48:28 -0400961 offset += wsize;
962 nbytes -= wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 } while (nbytes != 0);
964
Fred Isaman44b83792011-03-03 15:13:44 +0000965 put_lseg(lseg);
Fred Isaman36fe4322011-03-03 15:13:49 +0000966 desc->pg_lseg = NULL;
Trond Myklebustdbae4c72008-04-14 14:54:53 -0400967 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
969out_bad:
970 while (!list_empty(&list)) {
971 data = list_entry(list.next, struct nfs_write_data, pages);
972 list_del(&data->pages);
Fred Isaman0da2a4ac2011-01-19 14:18:50 -0500973 nfs_writedata_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 }
Trond Myklebust61822ab2006-12-05 00:35:42 -0500975 nfs_redirty_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return -ENOMEM;
977}
978
979/*
980 * Create an RPC task for the given write request and kick it.
981 * The page must have been locked by the caller.
982 *
983 * It may happen that the page we're passed is not marked dirty.
984 * This is the case if nfs_updatepage detects a conflicting request
985 * that has been written but not committed.
986 */
Fred Isamanc76069b2011-03-03 15:13:48 +0000987static int nfs_flush_one(struct nfs_pageio_descriptor *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988{
989 struct nfs_page *req;
990 struct page **pages;
991 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +0000992 struct list_head *head = &desc->pg_list;
993 struct pnfs_layout_segment *lseg = desc->pg_lseg;
Fred Isaman44b83792011-03-03 15:13:44 +0000994 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Fred Isamanc76069b2011-03-03 15:13:48 +0000996 data = nfs_writedata_alloc(nfs_page_array_len(desc->pg_base,
997 desc->pg_count));
Fred Isaman44b83792011-03-03 15:13:44 +0000998 if (!data) {
999 while (!list_empty(head)) {
1000 req = nfs_list_entry(head->next);
1001 nfs_list_remove_request(req);
1002 nfs_redirty_request(req);
1003 }
1004 ret = -ENOMEM;
1005 goto out;
1006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 pages = data->pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 while (!list_empty(head)) {
1009 req = nfs_list_entry(head->next);
1010 nfs_list_remove_request(req);
1011 nfs_list_add_request(req, &data->pages);
1012 ClearPageError(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 }
1015 req = nfs_list_entry(data->pages.next);
Fred Isaman44b83792011-03-03 15:13:44 +00001016 if ((!lseg) && list_is_singular(&data->pages))
Fred Isamanc76069b2011-03-03 15:13:48 +00001017 lseg = pnfs_update_layout(desc->pg_inode, req->wb_context, IOMODE_RW);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
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 */
Fred Isamanc76069b2011-03-03 15:13:48 +00001024 ret = nfs_write_rpcsetup(req, data, &nfs_write_full_ops, desc->pg_count, 0, lseg, desc->pg_ioflags);
Fred Isaman44b83792011-03-03 15:13:44 +00001025out:
1026 put_lseg(lseg); /* Cleans any gotten in ->pg_test */
Fred Isaman36fe4322011-03-03 15:13:49 +00001027 desc->pg_lseg = NULL;
Fred Isaman44b83792011-03-03 15:13:44 +00001028 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029}
1030
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001031static void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1032 struct inode *inode, int ioflags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033{
Chuck Leverbf4285e2007-12-20 14:54:57 -05001034 size_t wsize = NFS_SERVER(inode)->wsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Fred Isaman44b83792011-03-03 15:13:44 +00001036 pnfs_pageio_init_write(pgio, inode);
Fred Isaman94ad1c82011-03-01 01:34:14 +00001037
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001038 if (wsize < PAGE_CACHE_SIZE)
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001039 nfs_pageio_init(pgio, inode, nfs_flush_multi, wsize, ioflags);
Trond Myklebustbcb71bb2007-04-02 18:48:28 -04001040 else
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001041 nfs_pageio_init(pgio, inode, nfs_flush_one, wsize, ioflags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042}
1043
1044/*
1045 * Handle a write reply that flushed part of a page.
1046 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001047static void nfs_writeback_done_partial(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001049 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
Chuck Lever48186c72008-06-11 17:56:05 -04001051 dprintk("NFS: %5u write(%s/%lld %d@%lld)",
1052 task->tk_pid,
1053 data->req->wb_context->path.dentry->d_inode->i_sb->s_id,
1054 (long long)
1055 NFS_FILEID(data->req->wb_context->path.dentry->d_inode),
1056 data->req->wb_bytes, (long long)req_offset(data->req));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001058 nfs_writeback_done(task, data);
1059}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001060
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001061static void nfs_writeback_release_partial(void *calldata)
1062{
1063 struct nfs_write_data *data = calldata;
1064 struct nfs_page *req = data->req;
1065 struct page *page = req->wb_page;
1066 int status = data->task.tk_status;
1067
1068 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001069 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001070 nfs_context_set_write_error(req->wb_context, status);
1071 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001072 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001075 if (nfs_write_need_commit(data)) {
Trond Myklebust587142f2007-07-02 09:57:54 -04001076 struct inode *inode = page->mapping->host;
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001077
Trond Myklebust587142f2007-07-02 09:57:54 -04001078 spin_lock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001079 if (test_bit(PG_NEED_RESCHED, &req->wb_flags)) {
1080 /* Do nothing we need to resend the writes */
1081 } else if (!test_and_set_bit(PG_NEED_COMMIT, &req->wb_flags)) {
1082 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
1083 dprintk(" defer commit\n");
1084 } else if (memcmp(&req->wb_verf, &data->verf, sizeof(req->wb_verf))) {
1085 set_bit(PG_NEED_RESCHED, &req->wb_flags);
1086 clear_bit(PG_NEED_COMMIT, &req->wb_flags);
1087 dprintk(" server reboot detected\n");
1088 }
Trond Myklebust587142f2007-07-02 09:57:54 -04001089 spin_unlock(&inode->i_lock);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001090 } else
1091 dprintk(" OK\n");
1092
1093out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (atomic_dec_and_test(&req->wb_complete))
Fred Isamana861a1e2011-03-23 13:27:51 +00001095 nfs_writepage_release(req, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001096 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097}
1098
Andy Adamsondef6ed72009-04-01 09:22:26 -04001099#if defined(CONFIG_NFS_V4_1)
1100void nfs_write_prepare(struct rpc_task *task, void *calldata)
1101{
1102 struct nfs_write_data *data = calldata;
Andy Adamsondef6ed72009-04-01 09:22:26 -04001103
Trond Myklebust035168a2010-06-16 09:52:26 -04001104 if (nfs4_setup_sequence(NFS_SERVER(data->inode),
1105 &data->args.seq_args,
Andy Adamsondef6ed72009-04-01 09:22:26 -04001106 &data->res.seq_res, 1, task))
1107 return;
1108 rpc_call_start(task);
1109}
1110#endif /* CONFIG_NFS_V4_1 */
1111
Trond Myklebust788e7a82006-03-20 13:44:27 -05001112static const struct rpc_call_ops nfs_write_partial_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001113#if defined(CONFIG_NFS_V4_1)
1114 .rpc_call_prepare = nfs_write_prepare,
1115#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001116 .rpc_call_done = nfs_writeback_done_partial,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001117 .rpc_release = nfs_writeback_release_partial,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001118};
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120/*
1121 * Handle a write reply that flushes a whole page.
1122 *
1123 * FIXME: There is an inherent race with invalidate_inode_pages and
1124 * writebacks since the page->count is kept > 1 for as long
1125 * as the page has a write request pending.
1126 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001127static void nfs_writeback_done_full(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001129 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001131 nfs_writeback_done(task, data);
1132}
1133
1134static void nfs_writeback_release_full(void *calldata)
1135{
1136 struct nfs_write_data *data = calldata;
1137 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001138
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 /* Update attributes as result of writeback. */
1140 while (!list_empty(&data->pages)) {
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001141 struct nfs_page *req = nfs_list_entry(data->pages.next);
1142 struct page *page = req->wb_page;
1143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 nfs_list_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
Chuck Lever48186c72008-06-11 17:56:05 -04001146 dprintk("NFS: %5u write (%s/%lld %d@%lld)",
1147 data->task.tk_pid,
Trond Myklebust88be9f92007-06-05 10:42:27 -04001148 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1149 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 req->wb_bytes,
1151 (long long)req_offset(req));
1152
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001153 if (status < 0) {
Trond Myklebusta301b772007-02-06 11:07:15 -08001154 nfs_set_pageerror(page);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001155 nfs_context_set_write_error(req->wb_context, status);
1156 dprintk(", error = %d\n", status);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001157 goto remove_request;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001160 if (nfs_write_need_commit(data)) {
1161 memcpy(&req->wb_verf, &data->verf, sizeof(req->wb_verf));
Fred Isamana861a1e2011-03-23 13:27:51 +00001162 nfs_mark_request_commit(req, data->lseg);
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001163 dprintk(" marked for commit\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 goto next;
1165 }
Trond Myklebust8e821ca2007-04-20 16:12:34 -04001166 dprintk(" OK\n");
1167remove_request:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 nfs_inode_remove_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001170 nfs_clear_page_tag_locked(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001171 nfs_end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 }
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001173 nfs_writedata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174}
1175
Trond Myklebust788e7a82006-03-20 13:44:27 -05001176static const struct rpc_call_ops nfs_write_full_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001177#if defined(CONFIG_NFS_V4_1)
1178 .rpc_call_prepare = nfs_write_prepare,
1179#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001180 .rpc_call_done = nfs_writeback_done_full,
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001181 .rpc_release = nfs_writeback_release_full,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001182};
1183
1184
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185/*
1186 * This function is called when the WRITE call is complete.
1187 */
Fred Isaman13602892011-02-11 15:42:38 +00001188void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 struct nfs_writeargs *argp = &data->args;
1191 struct nfs_writeres *resp = &data->res;
Andy Adamsoneedc0202009-04-01 09:22:41 -04001192 struct nfs_server *server = NFS_SERVER(data->inode);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001193 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Chuck Levera3f565b2007-01-31 12:14:01 -05001195 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 task->tk_pid, task->tk_status);
1197
Chuck Leverf551e442006-09-20 14:33:04 -04001198 /*
1199 * ->write_done will attempt to use post-op attributes to detect
1200 * conflicting writes by other clients. A strict interpretation
1201 * of close-to-open would allow us to continue caching even if
1202 * another writer had changed the file, but some applications
1203 * depend on tighter cache coherency when writing.
1204 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001205 status = NFS_PROTO(data->inode)->write_done(task, data);
1206 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001207 return;
Chuck Lever91d5b472006-03-20 13:44:14 -05001208 nfs_add_stats(data->inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
1211 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1212 /* We tried a write call, but the server did not
1213 * commit data to stable storage even though we
1214 * requested it.
1215 * Note: There is a known bug in Tru64 < 5.0 in which
1216 * the server reports NFS_DATA_SYNC, but performs
1217 * NFS_FILE_SYNC. We therefore implement this checking
1218 * as a dprintk() in order to avoid filling syslog.
1219 */
1220 static unsigned long complain;
1221
Fred Isamana69aef12011-03-03 15:13:47 +00001222 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001224 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 " (committed = %d) != (stable = %d)\n",
Andy Adamsoneedc0202009-04-01 09:22:41 -04001226 server->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 resp->verf->committed, argp->stable);
1228 complain = jiffies + 300 * HZ;
1229 }
1230 }
1231#endif
1232 /* Is this a short write? */
1233 if (task->tk_status >= 0 && resp->count < argp->count) {
1234 static unsigned long complain;
1235
Chuck Lever91d5b472006-03-20 13:44:14 -05001236 nfs_inc_stats(data->inode, NFSIOS_SHORTWRITE);
1237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 /* Has the server at least made some progress? */
1239 if (resp->count != 0) {
1240 /* Was this an NFSv2 write or an NFSv3 stable write? */
1241 if (resp->verf->committed != NFS_UNSTABLE) {
1242 /* Resend from where the server left off */
Fred Isamana69aef12011-03-03 15:13:47 +00001243 data->mds_offset += resp->count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 argp->offset += resp->count;
1245 argp->pgbase += resp->count;
1246 argp->count -= resp->count;
1247 } else {
1248 /* Resend as a stable write in order to avoid
1249 * headaches in the case of a server crash.
1250 */
1251 argp->stable = NFS_FILE_SYNC;
1252 }
Trond Myklebust0110ee12009-12-07 09:00:24 -05001253 nfs_restart_rpc(task, server->nfs_client);
Fred Isaman13602892011-02-11 15:42:38 +00001254 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
1256 if (time_before(complain, jiffies)) {
1257 printk(KERN_WARNING
1258 "NFS: Server wrote zero bytes, expected %u.\n",
1259 argp->count);
1260 complain = jiffies + 300 * HZ;
1261 }
1262 /* Can't do anything about it except throw an error. */
1263 task->tk_status = -EIO;
1264 }
Fred Isaman13602892011-02-11 15:42:38 +00001265 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266}
1267
1268
1269#if defined(CONFIG_NFS_V3) || defined(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001270static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1271{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001272 int ret;
1273
Trond Myklebust71d0a612010-04-22 15:35:57 -04001274 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1275 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001276 if (!may_wait)
1277 return 0;
1278 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1279 NFS_INO_COMMIT,
1280 nfs_wait_bit_killable,
1281 TASK_KILLABLE);
1282 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001283}
1284
Fred Isamane0c2b382011-03-23 13:27:53 +00001285void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001286{
1287 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1288 smp_mb__after_clear_bit();
1289 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1290}
Fred Isamane0c2b382011-03-23 13:27:53 +00001291EXPORT_SYMBOL_GPL(nfs_commit_clear_lock);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001292
Fred Isamane0c2b382011-03-23 13:27:53 +00001293void nfs_commitdata_release(void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
Trond Myklebust383ba712008-02-19 20:04:20 -05001295 struct nfs_write_data *wdata = data;
1296
Fred Isaman988b6dc2011-03-23 13:27:52 +00001297 put_lseg(wdata->lseg);
Trond Myklebust383ba712008-02-19 20:04:20 -05001298 put_nfs_open_context(wdata->args.context);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 nfs_commit_free(wdata);
1300}
Fred Isamane0c2b382011-03-23 13:27:53 +00001301EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Fred Isamane0c2b382011-03-23 13:27:53 +00001303int nfs_initiate_commit(struct nfs_write_data *data, struct rpc_clnt *clnt,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001304 const struct rpc_call_ops *call_ops,
1305 int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Trond Myklebust07737692007-10-25 18:42:54 -04001307 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001308 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001309 struct rpc_message msg = {
1310 .rpc_argp = &data->args,
1311 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001312 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001313 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001314 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001315 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001316 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001317 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001318 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001319 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001320 .workqueue = nfsiod_workqueue,
Trond Myklebust2c61be02010-04-09 19:54:50 -04001321 .flags = RPC_TASK_ASYNC,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001322 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001323 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001324 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001325 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Chuck Levera3f565b2007-01-31 12:14:01 -05001327 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001328
Trond Myklebust07737692007-10-25 18:42:54 -04001329 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001330 if (IS_ERR(task))
1331 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001332 if (how & FLUSH_SYNC)
1333 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001334 rpc_put_task(task);
1335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
Fred Isamane0c2b382011-03-23 13:27:53 +00001337EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338
1339/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001340 * Set up the argument/result storage required for the RPC call.
1341 */
Fred Isamane0c2b382011-03-23 13:27:53 +00001342void nfs_init_commit(struct nfs_write_data *data,
Fred Isaman988b6dc2011-03-23 13:27:52 +00001343 struct list_head *head,
1344 struct pnfs_layout_segment *lseg)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001345{
1346 struct nfs_page *first = nfs_list_entry(head->next);
1347 struct inode *inode = first->wb_context->path.dentry->d_inode;
1348
1349 /* Set up the RPC argument and reply structs
1350 * NB: take care not to mess about with data->commit et al. */
1351
1352 list_splice_init(head, &data->pages);
1353
1354 data->inode = inode;
1355 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001356 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001357 data->mds_ops = &nfs_commit_ops;
1358
1359 data->args.fh = NFS_FH(data->inode);
1360 /* Note: we always request a commit of the entire inode */
1361 data->args.offset = 0;
1362 data->args.count = 0;
1363 data->args.context = get_nfs_open_context(first->wb_context);
1364 data->res.count = 0;
1365 data->res.fattr = &data->fattr;
1366 data->res.verf = &data->verf;
1367 nfs_fattr_init(&data->fattr);
1368}
Fred Isamane0c2b382011-03-23 13:27:53 +00001369EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001370
Fred Isamane0c2b382011-03-23 13:27:53 +00001371void nfs_retry_commit(struct list_head *page_list,
Fred Isamana861a1e2011-03-23 13:27:51 +00001372 struct pnfs_layout_segment *lseg)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001373{
1374 struct nfs_page *req;
1375
1376 while (!list_empty(page_list)) {
1377 req = nfs_list_entry(page_list->next);
1378 nfs_list_remove_request(req);
Fred Isamana861a1e2011-03-23 13:27:51 +00001379 nfs_mark_request_commit(req, lseg);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001380 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
1381 dec_bdi_stat(req->wb_page->mapping->backing_dev_info,
1382 BDI_RECLAIMABLE);
1383 nfs_clear_page_tag_locked(req);
1384 }
1385}
Fred Isamane0c2b382011-03-23 13:27:53 +00001386EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001387
Fred Isaman9ace33c2011-03-23 13:27:45 +00001388/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 * Commit dirty pages
1390 */
1391static int
Chuck Lever40859d72005-11-30 18:09:02 -05001392nfs_commit_list(struct inode *inode, struct list_head *head, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 struct nfs_write_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001396 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
1398 if (!data)
1399 goto out_bad;
1400
1401 /* Set up the argument struct */
Fred Isaman988b6dc2011-03-23 13:27:52 +00001402 nfs_init_commit(data, head, NULL);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001403 return nfs_initiate_commit(data, NFS_CLIENT(inode), data->mds_ops, how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 out_bad:
Fred Isamana861a1e2011-03-23 13:27:51 +00001405 nfs_retry_commit(head, NULL);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001406 nfs_commit_clear_lock(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 return -ENOMEM;
1408}
1409
1410/*
1411 * COMMIT call returned
1412 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001413static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414{
Trond Myklebust963d8fe2006-01-03 09:55:04 +01001415 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Chuck Levera3f565b2007-01-31 12:14:01 -05001417 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 task->tk_pid, task->tk_status);
1419
Trond Myklebust788e7a82006-03-20 13:44:27 -05001420 /* Call the NFS version-specific code */
1421 if (NFS_PROTO(data->inode)->commit_done(task, data) != 0)
1422 return;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001423}
1424
Fred Isamane0c2b382011-03-23 13:27:53 +00001425void nfs_commit_release_pages(struct nfs_write_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001426{
Fred Isaman5917ce82011-03-23 13:27:48 +00001427 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001428 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 while (!list_empty(&data->pages)) {
1431 req = nfs_list_entry(data->pages.next);
1432 nfs_list_remove_request(req);
Trond Myklebuste468bae2008-06-13 13:25:22 -04001433 nfs_clear_request_commit(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Chuck Lever48186c72008-06-11 17:56:05 -04001435 dprintk("NFS: commit (%s/%lld %d@%lld)",
Trond Myklebust88be9f92007-06-05 10:42:27 -04001436 req->wb_context->path.dentry->d_inode->i_sb->s_id,
1437 (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 req->wb_bytes,
1439 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001440 if (status < 0) {
1441 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001443 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 goto next;
1445 }
1446
1447 /* Okay, COMMIT succeeded, apparently. Check the verifier
1448 * returned by the server against all stored verfs. */
1449 if (!memcmp(req->wb_verf.verifier, data->verf.verifier, sizeof(data->verf.verifier))) {
1450 /* We have a match */
1451 nfs_inode_remove_request(req);
1452 dprintk(" OK\n");
1453 goto next;
1454 }
1455 /* We have a mismatch. Write the page again */
1456 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001457 nfs_mark_request_dirty(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 next:
Trond Myklebust9fd367f2007-06-17 15:10:24 -04001459 nfs_clear_page_tag_locked(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 }
Fred Isaman5917ce82011-03-23 13:27:48 +00001461}
Fred Isamane0c2b382011-03-23 13:27:53 +00001462EXPORT_SYMBOL_GPL(nfs_commit_release_pages);
Fred Isaman5917ce82011-03-23 13:27:48 +00001463
1464static void nfs_commit_release(void *calldata)
1465{
1466 struct nfs_write_data *data = calldata;
1467
1468 nfs_commit_release_pages(data);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001469 nfs_commit_clear_lock(NFS_I(data->inode));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001470 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001472
1473static const struct rpc_call_ops nfs_commit_ops = {
Andy Adamson21d9a852009-04-01 09:22:27 -04001474#if defined(CONFIG_NFS_V4_1)
1475 .rpc_call_prepare = nfs_write_prepare,
1476#endif /* CONFIG_NFS_V4_1 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001477 .rpc_call_done = nfs_commit_done,
1478 .rpc_release = nfs_commit_release,
1479};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Trond Myklebustb608b282010-07-30 15:31:54 -04001481int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 LIST_HEAD(head);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001484 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001485 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Trond Myklebustb8413f92011-03-21 15:37:01 -04001487 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1488 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001489 goto out_mark_dirty;
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001490 res = nfs_scan_commit(inode, &head, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001492 int error;
1493
1494 error = pnfs_commit_list(inode, &head, how);
1495 if (error == PNFS_NOT_ATTEMPTED)
1496 error = nfs_commit_list(inode, &head, how);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001497 if (error < 0)
1498 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001499 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001500 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001501 error = wait_on_bit(&NFS_I(inode)->flags,
1502 NFS_INO_COMMIT,
1503 nfs_wait_bit_killable,
1504 TASK_KILLABLE);
1505 if (error < 0)
1506 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001507 } else
1508 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001509 return res;
1510 /* Note: If we exit without ensuring that the commit is complete,
1511 * we must mark the inode as dirty. Otherwise, future calls to
1512 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1513 * that the data is on the disk.
1514 */
1515out_mark_dirty:
1516 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 return res;
1518}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001519
1520static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1521{
Trond Myklebust420e3642010-02-19 17:00:02 -08001522 struct nfs_inode *nfsi = NFS_I(inode);
1523 int flags = FLUSH_SYNC;
1524 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001525
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001526 if (wbc->sync_mode == WB_SYNC_NONE) {
1527 /* Don't commit yet if this is a non-blocking flush and there
1528 * are a lot of outstanding writes for this mapping.
1529 */
1530 if (nfsi->ncommit <= (nfsi->npages >> 1))
1531 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001532
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001533 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001534 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001535 }
1536
Trond Myklebust420e3642010-02-19 17:00:02 -08001537 ret = nfs_commit_inode(inode, flags);
1538 if (ret >= 0) {
1539 if (wbc->sync_mode == WB_SYNC_NONE) {
1540 if (ret < wbc->nr_to_write)
1541 wbc->nr_to_write -= ret;
1542 else
1543 wbc->nr_to_write = 0;
1544 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001545 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001546 }
1547out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001548 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1549 return ret;
1550}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001551#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001552static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1553{
1554 return 0;
1555}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556#endif
1557
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001558int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1559{
Andy Adamson863a3c62011-03-23 13:27:54 +00001560 int ret;
1561
1562 ret = nfs_commit_unstable_pages(inode, wbc);
1563 if (ret >= 0 && test_bit(NFS_INO_LAYOUTCOMMIT, &NFS_I(inode)->flags)) {
Andy Adamsonef311532011-03-12 02:58:10 -05001564 int status;
1565 bool sync = true;
Andy Adamson863a3c62011-03-23 13:27:54 +00001566
Andy Adamsonef311532011-03-12 02:58:10 -05001567 if (wbc->sync_mode == WB_SYNC_NONE || wbc->nonblocking ||
1568 wbc->for_background)
1569 sync = false;
Andy Adamson863a3c62011-03-23 13:27:54 +00001570
1571 status = pnfs_layoutcommit_inode(inode, sync);
1572 if (status < 0)
1573 return status;
1574 }
1575 return ret;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001576}
1577
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001578/*
1579 * flush the inode to disk.
1580 */
1581int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001582{
1583 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001584 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001585 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001586 .range_start = 0,
1587 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001588 };
Trond Myklebust34901f72007-07-25 14:09:54 -04001589
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001590 return sync_inode(inode, &wbc);
Trond Myklebust1c759502006-10-09 16:18:38 -04001591}
1592
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001593int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1594{
1595 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001596 int ret = 0;
1597
1598 BUG_ON(!PageLocked(page));
1599 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001600 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001601 req = nfs_page_find_request(page);
1602 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001603 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001604 if (nfs_lock_request_dontget(req)) {
1605 nfs_inode_remove_request(req);
1606 /*
1607 * In case nfs_inode_remove_request has marked the
1608 * page as being dirty
1609 */
1610 cancel_dirty_page(page, PAGE_CACHE_SIZE);
1611 nfs_unlock_request(req);
1612 break;
1613 }
1614 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001615 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001616 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001617 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001618 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001619 return ret;
1620}
1621
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001622/*
1623 * Write back all requests on one page - we do this before reading it.
1624 */
1625int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001626{
1627 loff_t range_start = page_offset(page);
1628 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001629 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001630 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001631 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001632 .range_start = range_start,
1633 .range_end = range_end,
1634 };
1635 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001636
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001637 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001638 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001639 if (clear_page_dirty_for_io(page)) {
1640 ret = nfs_writepage_locked(page, &wbc);
1641 if (ret < 0)
1642 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001643 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001644 }
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001645 if (!PagePrivate(page))
1646 break;
1647 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001648 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001649 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001650 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001651 return 0;
1652out_error:
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001653 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001654}
1655
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001656#ifdef CONFIG_MIGRATION
1657int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1658 struct page *page)
1659{
1660 struct nfs_page *req;
1661 int ret;
1662
Trond Myklebust7549ad52010-02-08 09:32:34 -05001663 nfs_fscache_release_page(page, GFP_KERNEL);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001664
Trond Myklebustcfb506e2010-07-30 15:31:57 -04001665 req = nfs_find_and_lock_request(page, false);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001666 ret = PTR_ERR(req);
1667 if (IS_ERR(req))
1668 goto out;
1669
1670 ret = migrate_page(mapping, newpage, page);
1671 if (!req)
1672 goto out;
1673 if (ret)
1674 goto out_unlock;
1675 page_cache_get(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001676 spin_lock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001677 req->wb_page = newpage;
1678 SetPagePrivate(newpage);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001679 set_page_private(newpage, (unsigned long)req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001680 ClearPagePrivate(page);
1681 set_page_private(page, 0);
Trond Myklebust190f38e2009-12-10 09:05:55 -05001682 spin_unlock(&mapping->host->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001683 page_cache_release(page);
1684out_unlock:
1685 nfs_clear_page_tag_locked(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001686out:
1687 return ret;
1688}
1689#endif
1690
David Howellsf7b422b2006-06-09 09:34:33 -04001691int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692{
1693 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1694 sizeof(struct nfs_write_data),
1695 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001696 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (nfs_wdata_cachep == NULL)
1698 return -ENOMEM;
1699
Matthew Dobson93d23412006-03-26 01:37:50 -08001700 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1701 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 if (nfs_wdata_mempool == NULL)
1703 return -ENOMEM;
1704
Matthew Dobson93d23412006-03-26 01:37:50 -08001705 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1706 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 if (nfs_commit_mempool == NULL)
1708 return -ENOMEM;
1709
Peter Zijlstra89a09142007-03-16 13:38:26 -08001710 /*
1711 * NFS congestion size, scale with available memory.
1712 *
1713 * 64MB: 8192k
1714 * 128MB: 11585k
1715 * 256MB: 16384k
1716 * 512MB: 23170k
1717 * 1GB: 32768k
1718 * 2GB: 46340k
1719 * 4GB: 65536k
1720 * 8GB: 92681k
1721 * 16GB: 131072k
1722 *
1723 * This allows larger machines to have larger/more transfers.
1724 * Limit the default to 256M
1725 */
1726 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1727 if (nfs_congestion_kb > 256*1024)
1728 nfs_congestion_kb = 256*1024;
1729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 return 0;
1731}
1732
David Brownell266bee82006-06-27 12:59:15 -07001733void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
1735 mempool_destroy(nfs_commit_mempool);
1736 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001737 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738}
1739