blob: 3ee5af4e738efb0f014d5b9dbd9963a179de5dc5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/write.c
3 *
Trond Myklebust7c85d902006-12-13 15:23:48 -05004 * Write file data over NFS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
7 */
8
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/writeback.h>
Peter Zijlstra89a09142007-03-16 13:38:26 -080015#include <linux/swap.h>
Trond Myklebust074cc1d2009-08-10 08:54:13 -040016#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070022#include <linux/backing-dev.h>
Paul Gortmakerafeacc82011-05-26 16:00:52 -040023#include <linux/export.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050028#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050029#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040030#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040031#include "fscache.h"
Fred Isaman94ad1c82011-03-01 01:34:14 +000032#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Trond Myklebustf4ce1292013-08-19 18:59:33 -040034#include "nfstrace.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#define NFSDBG_FACILITY NFSDBG_PAGECACHE
37
38#define MIN_POOL_WRITE (32)
39#define MIN_POOL_COMMIT (4)
40
41/*
42 * Local function declarations
43 */
Fred Isamanf8512ad2008-03-19 11:24:39 -040044static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050045static const struct rpc_call_ops nfs_commit_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -040046static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
Fred Isamanf453a542012-04-20 14:47:54 -040047static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
Anna Schumaker4a0de552014-05-06 09:12:30 -040048static const struct nfs_rw_ops nfs_rw_write_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Christoph Lametere18b8902006-12-06 20:33:20 -080050static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050051static mempool_t *nfs_wdata_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -040052static struct kmem_cache *nfs_cdata_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053static mempool_t *nfs_commit_mempool;
54
Fred Isaman0b7c0152012-04-20 14:47:39 -040055struct nfs_commit_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056{
Mel Gorman192e5012012-07-31 16:45:16 -070057 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
Chuck Lever40859d72005-11-30 18:09:02 -050058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 if (p) {
60 memset(p, 0, sizeof(*p));
61 INIT_LIST_HEAD(&p->pages);
62 }
63 return p;
64}
Fred Isamane0c2b382011-03-23 13:27:53 +000065EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Fred Isaman0b7c0152012-04-20 14:47:39 -040067void nfs_commit_free(struct nfs_commit_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 mempool_free(p, nfs_commit_mempool);
70}
Fred Isamane0c2b382011-03-23 13:27:53 +000071EXPORT_SYMBOL_GPL(nfs_commit_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Anna Schumaker4a0de552014-05-06 09:12:30 -040073static struct nfs_rw_header *nfs_writehdr_alloc(void)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050074{
Anna Schumakerc0752cd2014-05-06 09:12:27 -040075 struct nfs_rw_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050076
Anna Schumaker4a0de552014-05-06 09:12:30 -040077 if (p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050078 memset(p, 0, sizeof(*p));
Trond Myklebust3feb2d42006-03-20 13:44:37 -050079 return p;
80}
81
Anna Schumaker4a0de552014-05-06 09:12:30 -040082static void nfs_writehdr_free(struct nfs_rw_header *whdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -040083{
Fred Isamancd841602012-04-20 14:47:44 -040084 mempool_free(whdr, nfs_wdata_mempool);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050085}
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Trond Myklebust7b159fc2007-07-25 14:09:54 -040087static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
88{
89 ctx->error = error;
90 smp_wmb();
91 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
92}
93
Mel Gorman29418aa2012-07-31 16:45:10 -070094static struct nfs_page *
95nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -050096{
97 struct nfs_page *req = NULL;
98
Mel Gorman29418aa2012-07-31 16:45:10 -070099 if (PagePrivate(page))
Trond Myklebust277459d2006-12-05 00:35:35 -0500100 req = (struct nfs_page *)page_private(page);
Mel Gorman29418aa2012-07-31 16:45:10 -0700101 else if (unlikely(PageSwapCache(page))) {
102 struct nfs_page *freq, *t;
103
104 /* Linearly search the commit list for the correct req */
105 list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) {
106 if (freq->wb_page == page) {
107 req = freq;
108 break;
109 }
110 }
Trond Myklebust277459d2006-12-05 00:35:35 -0500111 }
Mel Gorman29418aa2012-07-31 16:45:10 -0700112
113 if (req)
114 kref_get(&req->wb_kref);
115
Trond Myklebust277459d2006-12-05 00:35:35 -0500116 return req;
117}
118
119static struct nfs_page *nfs_page_find_request(struct page *page)
120{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700121 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500122 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500123
Trond Myklebust587142f2007-07-02 09:57:54 -0400124 spin_lock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700125 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400126 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500127 return req;
128}
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130/* Adjust the file length if we're writing beyond the end */
131static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
132{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700133 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400134 loff_t end, i_size;
135 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136
Trond Myklebusta3d01452008-06-11 12:21:19 -0400137 spin_lock(&inode->i_lock);
138 i_size = i_size_read(inode);
139 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700140 if (i_size > 0 && page_file_index(page) < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400141 goto out;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700142 end = page_file_offset(page) + ((loff_t)offset+count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400144 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400146 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
147out:
148 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Trond Myklebusta301b772007-02-06 11:07:15 -0800151/* A writeback failed: mark the page as bad, and invalidate the page cache */
152static void nfs_set_pageerror(struct page *page)
153{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700154 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
Trond Myklebusta301b772007-02-06 11:07:15 -0800155}
156
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400157/*
158 * nfs_page_group_search_locked
159 * @head - head request of page group
160 * @page_offset - offset into page
161 *
162 * Search page group with head @head to find a request that contains the
163 * page offset @page_offset.
164 *
165 * Returns a pointer to the first matching nfs request, or NULL if no
166 * match is found.
167 *
168 * Must be called with the page group lock held
169 */
170static struct nfs_page *
171nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
172{
173 struct nfs_page *req;
174
175 WARN_ON_ONCE(head != head->wb_head);
176 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
177
178 req = head;
179 do {
180 if (page_offset >= req->wb_pgbase &&
181 page_offset < (req->wb_pgbase + req->wb_bytes))
182 return req;
183
184 req = req->wb_this_page;
185 } while (req != head);
186
187 return NULL;
188}
189
190/*
191 * nfs_page_group_covers_page
192 * @head - head request of page group
193 *
194 * Return true if the page group with head @head covers the whole page,
195 * returns false otherwise
196 */
197static bool nfs_page_group_covers_page(struct nfs_page *req)
198{
199 struct nfs_page *tmp;
200 unsigned int pos = 0;
201 unsigned int len = nfs_page_length(req->wb_page);
202
203 nfs_page_group_lock(req);
204
205 do {
206 tmp = nfs_page_group_search_locked(req->wb_head, pos);
207 if (tmp) {
208 /* no way this should happen */
209 WARN_ON_ONCE(tmp->wb_pgbase != pos);
210 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
211 }
212 } while (tmp && pos < len);
213
214 nfs_page_group_unlock(req);
215 WARN_ON_ONCE(pos > len);
216 return pos == len;
217}
218
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219/* We can set the PG_uptodate flag if we see that a write request
220 * covers the full page.
221 */
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400222static void nfs_mark_uptodate(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223{
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400224 if (PageUptodate(req->wb_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400226 if (!nfs_page_group_covers_page(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400228 SetPageUptodate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231static int wb_priority(struct writeback_control *wbc)
232{
233 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400234 return FLUSH_HIGHPRI | FLUSH_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100235 if (wbc->for_kupdate || wbc->for_background)
Trond Myklebustb31268a2011-03-21 17:02:00 -0400236 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
237 return FLUSH_COND_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800241 * NFS congestion control
242 */
243
244int nfs_congestion_kb;
245
246#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
247#define NFS_CONGESTION_OFF_THRESH \
248 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
249
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400250static void nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800251{
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400252 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400253 int ret = test_set_page_writeback(page);
254
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400255 WARN_ON_ONCE(ret != 0);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800256
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400257 if (atomic_long_inc_return(&nfss->writeback) >
258 NFS_CONGESTION_ON_THRESH) {
259 set_bdi_congested(&nfss->backing_dev_info,
260 BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800261 }
262}
263
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400264static void nfs_end_page_writeback(struct nfs_page *req)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800265{
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400266 struct inode *inode = page_file_mapping(req->wb_page)->host;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800267 struct nfs_server *nfss = NFS_SERVER(inode);
268
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400269 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
270 return;
271
272 end_page_writeback(req->wb_page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700273 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200274 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800275}
276
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400277static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
Trond Myklebuste261f512006-12-05 00:35:41 -0500278{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700279 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500280 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500281 int ret;
282
Trond Myklebust587142f2007-07-02 09:57:54 -0400283 spin_lock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400284 for (;;) {
Mel Gorman29418aa2012-07-31 16:45:10 -0700285 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400286 if (req == NULL)
287 break;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400288 if (nfs_lock_request(req))
Trond Myklebuste261f512006-12-05 00:35:41 -0500289 break;
290 /* Note: If we hold the page lock, as is the case in nfs_writepage,
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400291 * then the call to nfs_lock_request() will always
Trond Myklebuste261f512006-12-05 00:35:41 -0500292 * succeed provided that someone hasn't already marked the
293 * request as dirty (in which case we don't care).
294 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400295 spin_unlock(&inode->i_lock);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400296 if (!nonblock)
297 ret = nfs_wait_on_request(req);
298 else
299 ret = -EAGAIN;
Trond Myklebuste261f512006-12-05 00:35:41 -0500300 nfs_release_request(req);
301 if (ret != 0)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400302 return ERR_PTR(ret);
Trond Myklebust587142f2007-07-02 09:57:54 -0400303 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500304 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400305 spin_unlock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400306 return req;
307}
308
309/*
310 * Find an associated nfs write request, and prepare to flush it out
311 * May return an error if the user signalled nfs_wait_on_request().
312 */
313static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400314 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400315{
316 struct nfs_page *req;
317 int ret = 0;
318
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400319 req = nfs_find_and_lock_request(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400320 if (!req)
321 goto out;
322 ret = PTR_ERR(req);
323 if (IS_ERR(req))
324 goto out;
325
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400326 nfs_set_page_writeback(page);
327 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400328
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400329 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400330 if (!nfs_pageio_add_request(pgio, req)) {
331 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400332 ret = pgio->pg_error;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400333 }
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400334out:
335 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500336}
337
Trond Myklebustf758c8852007-07-22 19:27:32 -0400338static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
339{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700340 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400341 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400342
343 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
344 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
345
Mel Gormand56b4dd2012-07-31 16:45:06 -0700346 nfs_pageio_cond_complete(pgio, page_file_index(page));
Wu Fengguang1b430be2010-10-26 14:21:26 -0700347 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400348 if (ret == -EAGAIN) {
349 redirty_page_for_writepage(wbc, page);
350 ret = 0;
351 }
352 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400353}
354
Trond Myklebuste261f512006-12-05 00:35:41 -0500355/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 * Write an mmapped page to the server.
357 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500358static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400360 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500361 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200363 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
364 false, &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400365 err = nfs_do_writepage(page, wbc, &pgio);
366 nfs_pageio_complete(&pgio);
367 if (err < 0)
368 return err;
369 if (pgio.pg_error < 0)
370 return pgio.pg_error;
371 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500372}
373
374int nfs_writepage(struct page *page, struct writeback_control *wbc)
375{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400376 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500377
Trond Myklebustf758c8852007-07-22 19:27:32 -0400378 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400380 return ret;
381}
382
383static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
384{
385 int ret;
386
387 ret = nfs_do_writepage(page, wbc, data);
388 unlock_page(page);
389 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
393{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400395 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400396 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 int err;
398
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400399 /* Stop dirtying of new pages while we sync */
400 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
401 nfs_wait_bit_killable, TASK_KILLABLE);
402 if (err)
403 goto out_err;
404
Chuck Lever91d5b472006-03-20 13:44:14 -0500405 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
406
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200407 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
408 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400409 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400410 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400411
412 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100413 smp_mb__after_atomic();
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400414 wake_up_bit(bitlock, NFS_INO_FLUSHING);
415
Trond Myklebustf758c8852007-07-22 19:27:32 -0400416 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400417 goto out_err;
418 err = pgio.pg_error;
419 if (err < 0)
420 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400421 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400422out_err:
423 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426/*
427 * Insert a write request into an inode
428 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500429static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
431 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400432
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400433 WARN_ON_ONCE(req->wb_this_page != req);
434
Trond Myklebuste7d39062008-06-13 12:12:32 -0400435 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400436 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400437
438 spin_lock(&inode->i_lock);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400439 if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700440 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700441 /*
442 * Swap-space should not get truncated. Hence no need to plug the race
443 * with invalidate/truncate.
444 */
445 if (likely(!PageSwapCache(req->wb_page))) {
446 set_bit(PG_MAPPED, &req->wb_flags);
447 SetPagePrivate(req->wb_page);
448 set_page_private(req->wb_page, (unsigned long)req);
449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 nfsi->npages++;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400451 set_bit(PG_INODE_REF, &req->wb_flags);
Trond Myklebustc03b4022007-06-17 13:26:38 -0400452 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400453 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454}
455
456/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800457 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 */
459static void nfs_inode_remove_request(struct nfs_page *req)
460{
Al Viro3d4ff432011-06-22 18:40:12 -0400461 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400463 struct nfs_page *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400465 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
466 head = req->wb_head;
467
468 spin_lock(&inode->i_lock);
469 if (likely(!PageSwapCache(head->wb_page))) {
470 set_page_private(head->wb_page, 0);
471 ClearPagePrivate(head->wb_page);
472 clear_bit(PG_MAPPED, &head->wb_flags);
473 }
474 nfsi->npages--;
475 spin_unlock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 nfs_release_request(req);
478}
479
Trond Myklebust61822ab2006-12-05 00:35:42 -0500480static void
Fred6d884e82008-03-19 11:24:38 -0400481nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500482{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500483 __set_page_dirty_nobuffers(req->wb_page);
484}
485
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400486#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400487/**
488 * nfs_request_add_commit_list - add request to a commit list
489 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400490 * @dst: commit list head
491 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400492 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400493 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400494 * number of outstanding requests requiring a commit as well as
495 * the MM page stats.
496 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400497 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400498 * holding the nfs_page lock.
499 */
500void
Fred Isamanea2cf222012-04-20 14:47:53 -0400501nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
502 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400503{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400504 set_bit(PG_CLEAN, &(req)->wb_flags);
Fred Isamanea2cf222012-04-20 14:47:53 -0400505 spin_lock(cinfo->lock);
506 nfs_list_add_request(req, dst);
507 cinfo->mds->ncommit++;
508 spin_unlock(cinfo->lock);
Fred Isaman56f9cd62012-04-20 14:47:56 -0400509 if (!cinfo->dreq) {
510 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700511 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
Fred Isaman56f9cd62012-04-20 14:47:56 -0400512 BDI_RECLAIMABLE);
513 __mark_inode_dirty(req->wb_context->dentry->d_inode,
514 I_DIRTY_DATASYNC);
515 }
Trond Myklebust8dd37752012-03-15 17:16:40 -0400516}
517EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
518
519/**
520 * nfs_request_remove_commit_list - Remove request from a commit list
521 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400522 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400523 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400524 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400525 * number of outstanding requests requiring a commit
526 * It does not update the MM page stats.
527 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400528 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400529 */
530void
Fred Isamanea2cf222012-04-20 14:47:53 -0400531nfs_request_remove_commit_list(struct nfs_page *req,
532 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400533{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400534 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
535 return;
536 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400537 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400538}
539EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
540
Fred Isamanea2cf222012-04-20 14:47:53 -0400541static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
542 struct inode *inode)
543{
544 cinfo->lock = &inode->i_lock;
545 cinfo->mds = &NFS_I(inode)->commit_info;
546 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400547 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400548 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400549}
550
551void nfs_init_cinfo(struct nfs_commit_info *cinfo,
552 struct inode *inode,
553 struct nfs_direct_req *dreq)
554{
Fred Isaman1763da12012-04-20 14:47:57 -0400555 if (dreq)
556 nfs_init_cinfo_from_dreq(cinfo, dreq);
557 else
558 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400559}
560EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * Add a request to the inode's commit list.
564 */
Fred Isaman1763da12012-04-20 14:47:57 -0400565void
Fred Isamanea2cf222012-04-20 14:47:53 -0400566nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
567 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Fred Isamanea2cf222012-04-20 14:47:53 -0400569 if (pnfs_mark_request_commit(req, lseg, cinfo))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400570 return;
Fred Isamanea2cf222012-04-20 14:47:53 -0400571 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400573
Fred Isamand6d6dc72012-03-08 17:29:35 -0500574static void
575nfs_clear_page_commit(struct page *page)
576{
577 dec_zone_page_state(page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700578 dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500579}
580
Trond Myklebust8dd37752012-03-15 17:16:40 -0400581static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400582nfs_clear_request_commit(struct nfs_page *req)
583{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400584 if (test_bit(PG_CLEAN, &req->wb_flags)) {
585 struct inode *inode = req->wb_context->dentry->d_inode;
Fred Isamanea2cf222012-04-20 14:47:53 -0400586 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400587
Fred Isamanea2cf222012-04-20 14:47:53 -0400588 nfs_init_cinfo_from_inode(&cinfo, inode);
589 if (!pnfs_clear_request_commit(req, &cinfo)) {
590 spin_lock(cinfo.lock);
591 nfs_request_remove_commit_list(req, &cinfo);
592 spin_unlock(cinfo.lock);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400593 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500594 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400595 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400596}
597
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400598static inline
Anna Schumaker9c7e1b32014-05-06 09:12:26 -0400599int nfs_write_need_commit(struct nfs_pgio_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400600{
Fred Isaman465d5242011-03-23 13:27:44 +0000601 if (data->verf.committed == NFS_DATA_SYNC)
Fred Isamancd841602012-04-20 14:47:44 -0400602 return data->header->lseg == NULL;
603 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400604}
605
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400606#else
Bryan Schumaker68cd6fa2012-04-30 14:30:22 -0400607static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
608 struct inode *inode)
609{
610}
611
612void nfs_init_cinfo(struct nfs_commit_info *cinfo,
613 struct inode *inode,
614 struct nfs_direct_req *dreq)
615{
616}
617
Fred Isaman1763da12012-04-20 14:47:57 -0400618void
Fred Isamanea2cf222012-04-20 14:47:53 -0400619nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
620 struct nfs_commit_info *cinfo)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400621{
622}
623
Trond Myklebust8dd37752012-03-15 17:16:40 -0400624static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400625nfs_clear_request_commit(struct nfs_page *req)
626{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400627}
628
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400629static inline
Anna Schumaker9c7e1b32014-05-06 09:12:26 -0400630int nfs_write_need_commit(struct nfs_pgio_data *data)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400631{
632 return 0;
633}
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635#endif
636
Fred Isaman061ae2e2012-04-20 14:47:48 -0400637static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400638{
Fred Isamanea2cf222012-04-20 14:47:53 -0400639 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400640 unsigned long bytes = 0;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400641 bool do_destroy;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400642
643 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
644 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400645 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400646 while (!list_empty(&hdr->pages)) {
647 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400648
649 bytes += req->wb_bytes;
650 nfs_list_remove_request(req);
651 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
652 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400653 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400654 nfs_context_set_write_error(req->wb_context, hdr->error);
655 goto remove_req;
656 }
657 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
658 nfs_mark_request_dirty(req);
659 goto next;
660 }
661 if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
Anna Schumakerf79d06f2014-05-06 09:12:28 -0400662 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
Fred Isamanea2cf222012-04-20 14:47:53 -0400663 nfs_mark_request_commit(req, hdr->lseg, &cinfo);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400664 goto next;
665 }
666remove_req:
667 nfs_inode_remove_request(req);
668next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400669 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400670 nfs_end_page_writeback(req);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400671 do_destroy = !test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400672 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400673 }
674out:
675 hdr->release(hdr);
676}
677
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400678#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Anna Schumakerce595152014-05-06 09:12:34 -0400679unsigned long
Fred Isamanea2cf222012-04-20 14:47:53 -0400680nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400681{
Fred Isamanea2cf222012-04-20 14:47:53 -0400682 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400683}
684
Fred Isamanea2cf222012-04-20 14:47:53 -0400685/* cinfo->lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400686int
Fred Isamanea2cf222012-04-20 14:47:53 -0400687nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
688 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500689{
690 struct nfs_page *req, *tmp;
691 int ret = 0;
692
693 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400694 if (!nfs_lock_request(req))
695 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400696 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -0400697 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400698 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400699 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400700 nfs_list_add_request(req, dst);
701 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400702 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400703 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500704 }
705 return ret;
706}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500707
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708/*
709 * nfs_scan_commit - Scan an inode for commit requests
710 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400711 * @dst: mds destination list
712 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 *
714 * Moves requests from the inode's 'commit' request list.
715 * The requests are *not* checked to ensure that they form a contiguous set.
716 */
Fred Isaman1763da12012-04-20 14:47:57 -0400717int
Fred Isamanea2cf222012-04-20 14:47:53 -0400718nfs_scan_commit(struct inode *inode, struct list_head *dst,
719 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720{
Fred Isamand6d6dc72012-03-08 17:29:35 -0500721 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400722
Fred Isamanea2cf222012-04-20 14:47:53 -0400723 spin_lock(cinfo->lock);
724 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400725 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500726
Fred Isamanea2cf222012-04-20 14:47:53 -0400727 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
728 cinfo, max);
729 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500730 }
Fred Isamanea2cf222012-04-20 14:47:53 -0400731 spin_unlock(cinfo->lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800732 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500734
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500735#else
Anna Schumakerce595152014-05-06 09:12:34 -0400736unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400737{
738 return 0;
739}
740
Fred Isaman1763da12012-04-20 14:47:57 -0400741int nfs_scan_commit(struct inode *inode, struct list_head *dst,
742 struct nfs_commit_info *cinfo)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500743{
744 return 0;
745}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746#endif
747
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400749 * Search for an existing write request, and attempt to update
750 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400752 * If the attempt fails, then the existing request is flushed out
753 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400755static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
756 struct page *page,
757 unsigned int offset,
758 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400760 struct nfs_page *req;
761 unsigned int rqend;
762 unsigned int end;
763 int error;
764
765 if (!PagePrivate(page))
766 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400769 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 for (;;) {
Mel Gorman29418aa2012-07-31 16:45:10 -0700772 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400773 if (req == NULL)
774 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500775
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400776 /* should be handled by nfs_flush_incompatible */
777 WARN_ON_ONCE(req->wb_head != req);
778 WARN_ON_ONCE(req->wb_this_page != req);
779
Trond Myklebuste7d39062008-06-13 12:12:32 -0400780 rqend = req->wb_offset + req->wb_bytes;
781 /*
782 * Tell the caller to flush out the request if
783 * the offsets are non-contiguous.
784 * Note: nfs_flush_incompatible() will already
785 * have flushed out requests having wrong owners.
786 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400787 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400788 || end < req->wb_offset)
789 goto out_flushme;
790
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400791 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
Trond Myklebuste7d39062008-06-13 12:12:32 -0400794 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400795 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400796 error = nfs_wait_on_request(req);
797 nfs_release_request(req);
798 if (error != 0)
799 goto out_err;
800 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 }
802
803 /* Okay, the request matches. Update the region */
804 if (offset < req->wb_offset) {
805 req->wb_offset = offset;
806 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 if (end > rqend)
809 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400810 else
811 req->wb_bytes = rqend - req->wb_offset;
812out_unlock:
813 spin_unlock(&inode->i_lock);
Fred Isamanca138f32012-04-05 15:26:36 -0400814 if (req)
815 nfs_clear_request_commit(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400816 return req;
817out_flushme:
818 spin_unlock(&inode->i_lock);
819 nfs_release_request(req);
820 error = nfs_wb_page(inode, page);
821out_err:
822 return ERR_PTR(error);
823}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
Trond Myklebuste7d39062008-06-13 12:12:32 -0400825/*
826 * Try to update an existing write request, or create one if there is none.
827 *
828 * Note: Should always be called with the Page Lock held to prevent races
829 * if we have to add a new request. Also assumes that the caller has
830 * already called nfs_flush_incompatible() if necessary.
831 */
832static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
833 struct page *page, unsigned int offset, unsigned int bytes)
834{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700835 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400836 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400837
838 req = nfs_try_to_update_request(inode, page, offset, bytes);
839 if (req != NULL)
840 goto out;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400841 req = nfs_create_request(ctx, page, NULL, offset, bytes);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400842 if (IS_ERR(req))
843 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500844 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400845out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400846 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847}
848
Trond Myklebuste7d39062008-06-13 12:12:32 -0400849static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
850 unsigned int offset, unsigned int count)
851{
852 struct nfs_page *req;
853
854 req = nfs_setup_write_request(ctx, page, offset, count);
855 if (IS_ERR(req))
856 return PTR_ERR(req);
857 /* Update file length */
858 nfs_grow_file(page, offset, count);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400859 nfs_mark_uptodate(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400860 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400861 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400862 return 0;
863}
864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865int nfs_flush_incompatible(struct file *file, struct page *page)
866{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400867 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -0400868 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500870 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 /*
872 * Look for a request corresponding to this page. If there
873 * is one, and it belongs to another file, we flush it out
874 * before we try to copy anything into the page. Do this
875 * due to the lack of an ACCESS-type call in NFSv2.
876 * Also do the same if we find a request from an existing
877 * dropped page.
878 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500879 do {
880 req = nfs_page_find_request(page);
881 if (req == NULL)
882 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -0400883 l_ctx = req->wb_lock_context;
884 do_flush = req->wb_page != page || req->wb_context != ctx;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400885 /* for now, flush if more than 1 request in page_group */
886 do_flush |= req->wb_this_page != req;
Trond Myklebust0f1d2602013-09-05 15:52:51 -0400887 if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
Trond Myklebust2a369152012-08-13 18:54:45 -0400888 do_flush |= l_ctx->lockowner.l_owner != current->files
889 || l_ctx->lockowner.l_pid != current->tgid;
890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500892 if (!do_flush)
893 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700894 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -0500895 } while (status == 0);
896 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897}
898
899/*
Andy Adamsondc248262013-08-14 11:59:16 -0400900 * Avoid buffered writes when a open context credential's key would
901 * expire soon.
902 *
903 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
904 *
905 * Return 0 and set a credential flag which triggers the inode to flush
906 * and performs NFS_FILE_SYNC writes if the key will expired within
907 * RPC_KEY_EXPIRE_TIMEO.
908 */
909int
910nfs_key_timeout_notify(struct file *filp, struct inode *inode)
911{
912 struct nfs_open_context *ctx = nfs_file_open_context(filp);
913 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
914
915 return rpcauth_key_timeout_notify(auth, ctx->cred);
916}
917
918/*
919 * Test if the open context credential key is marked to expire soon.
920 */
921bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
922{
923 return rpcauth_cred_key_to_expire(ctx->cred);
924}
925
926/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500927 * If the page cache is marked as unsafe or invalid, then we can't rely on
928 * the PageUptodate() flag. In this case, we will need to turn off
929 * write optimisations that depend on the page contents being correct.
930 */
Trond Myklebust8d197a52012-04-29 12:50:01 -0400931static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -0500932{
Jeff Laytond529ef82014-01-27 13:46:15 -0500933 struct nfs_inode *nfsi = NFS_I(inode);
934
Trond Myklebust8d197a52012-04-29 12:50:01 -0400935 if (nfs_have_delegated_attributes(inode))
936 goto out;
Jeff Laytond529ef82014-01-27 13:46:15 -0500937 if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE))
938 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -0500939 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -0500940 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -0400941 return false;
942out:
943 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -0500944}
945
Scott Mayhewc7559662013-07-05 17:33:19 -0400946/* If we know the page is up to date, and we're not using byte range locks (or
947 * if we have the whole file locked for writing), it may be more efficient to
948 * extend the write to cover the entire page in order to avoid fragmentation
949 * inefficiencies.
950 *
Scott Mayhew263b4502014-01-17 15:12:05 -0500951 * If the file is opened for synchronous writes then we can just skip the rest
952 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -0400953 */
954static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
955{
956 if (file->f_flags & O_DSYNC)
957 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -0500958 if (!nfs_write_pageuptodate(page, inode))
959 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -0400960 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
961 return 1;
Scott Mayhew263b4502014-01-17 15:12:05 -0500962 if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
Scott Mayhewc7559662013-07-05 17:33:19 -0400963 inode->i_flock->fl_end == OFFSET_MAX &&
Scott Mayhew263b4502014-01-17 15:12:05 -0500964 inode->i_flock->fl_type != F_RDLCK))
Scott Mayhewc7559662013-07-05 17:33:19 -0400965 return 1;
966 return 0;
967}
968
Trond Myklebust5d47a352008-02-07 17:24:07 -0500969/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * Update and possibly write a cached page of an NFS file.
971 *
972 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
973 * things with a page scheduled for an RPC call (e.g. invalidate it).
974 */
975int nfs_updatepage(struct file *file, struct page *page,
976 unsigned int offset, unsigned int count)
977{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400978 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700979 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 int status = 0;
981
Chuck Lever91d5b472006-03-20 13:44:14 -0500982 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
983
Al Viro6de14722013-09-16 10:53:17 -0400984 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
985 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Scott Mayhewc7559662013-07-05 17:33:19 -0400987 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500988 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 }
991
Trond Myklebuste21195a2006-12-05 00:35:39 -0500992 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400993 if (status < 0)
994 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700995 else
996 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Chuck Lever48186c72008-06-11 17:56:05 -0400998 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 return status;
1001}
1002
Trond Myklebust3ff75762007-07-14 15:40:00 -04001003static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1006 case FLUSH_HIGHPRI:
1007 return RPC_PRIORITY_HIGH;
1008 case FLUSH_LOWPRI:
1009 return RPC_PRIORITY_LOW;
1010 }
1011 return RPC_PRIORITY_NORMAL;
1012}
1013
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001014static void nfs_initiate_write(struct nfs_pgio_data *data, struct rpc_message *msg,
1015 struct rpc_task_setup *task_setup_data, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
Fred Isamancd841602012-04-20 14:47:44 -04001017 struct inode *inode = data->header->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -04001018 int priority = flush_task_priority(how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001020 task_setup_data->priority = priority;
1021 NFS_PROTO(inode)->write_setup(data, msg);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001022
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001023 nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001024 &task_setup_data->rpc_client, msg, data);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001025}
1026
Fred6d884e82008-03-19 11:24:38 -04001027/* If a nfs_flush_* function fails, it should remove reqs from @head and
1028 * call this on each, which will prepare them to be retried on next
1029 * writeback using standard nfs.
1030 */
1031static void nfs_redirty_request(struct nfs_page *req)
1032{
1033 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001034 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -04001035 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001036 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001037}
1038
Fred Isaman061ae2e2012-04-20 14:47:48 -04001039static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001040{
1041 struct nfs_page *req;
1042
1043 while (!list_empty(head)) {
1044 req = nfs_list_entry(head->next);
1045 nfs_list_remove_request(req);
1046 nfs_redirty_request(req);
1047 }
1048}
1049
Fred Isaman061ae2e2012-04-20 14:47:48 -04001050static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1051 .error_cleanup = nfs_async_write_error,
1052 .completion = nfs_write_completion,
1053};
1054
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001055void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001056 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001057 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001058{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001059 struct nfs_server *server = NFS_SERVER(inode);
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001060 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001061
1062#ifdef CONFIG_NFS_V4_1
1063 if (server->pnfs_curr_ld && !force_mds)
1064 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1065#endif
Anna Schumaker4a0de552014-05-06 09:12:30 -04001066 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1067 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001068}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001069EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001070
Trond Myklebustdce81292011-07-13 15:59:19 -04001071void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1072{
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001073 pgio->pg_ops = &nfs_pgio_rw_ops;
Trond Myklebustdce81292011-07-13 15:59:19 -04001074 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1075}
Trond Myklebust1f945352011-07-13 15:59:57 -04001076EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Fred Isaman0b7c0152012-04-20 14:47:39 -04001079void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1080{
1081 struct nfs_commit_data *data = calldata;
1082
1083 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1084}
1085
Anna Schumakera4cdda52014-05-06 09:12:31 -04001086static void nfs_writeback_release_common(struct nfs_pgio_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087{
Fred Isamancd841602012-04-20 14:47:44 -04001088 struct nfs_pgio_header *hdr = data->header;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001089 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001090
Fred Isaman6c75dc02012-04-20 14:47:47 -04001091 if ((status >= 0) && nfs_write_need_commit(data)) {
1092 spin_lock(&hdr->lock);
1093 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
1094 ; /* Do nothing */
1095 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
Anna Schumakerf79d06f2014-05-06 09:12:28 -04001096 memcpy(&hdr->verf, &data->verf, sizeof(hdr->verf));
1097 else if (memcmp(&hdr->verf, &data->verf, sizeof(hdr->verf)))
Fred Isaman6c75dc02012-04-20 14:47:47 -04001098 set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
1099 spin_unlock(&hdr->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 }
1101}
1102
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001103/*
1104 * Special version of should_remove_suid() that ignores capabilities.
1105 */
1106static int nfs_should_remove_suid(const struct inode *inode)
1107{
1108 umode_t mode = inode->i_mode;
1109 int kill = 0;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001110
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001111 /* suid always must be killed */
1112 if (unlikely(mode & S_ISUID))
1113 kill = ATTR_KILL_SUID;
1114
1115 /*
1116 * sgid without any exec bits is just a mandatory locking mark; leave
1117 * it alone. If some exec bits are set, it's a real sgid; kill it.
1118 */
1119 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1120 kill |= ATTR_KILL_SGID;
1121
1122 if (unlikely(kill && S_ISREG(mode)))
1123 return kill;
1124
1125 return 0;
1126}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128/*
1129 * This function is called when the WRITE call is complete.
1130 */
Anna Schumaker0eecb212014-05-06 09:12:32 -04001131static int nfs_writeback_done(struct rpc_task *task, struct nfs_pgio_data *data,
1132 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001134 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Chuck Leverf551e442006-09-20 14:33:04 -04001136 /*
1137 * ->write_done will attempt to use post-op attributes to detect
1138 * conflicting writes by other clients. A strict interpretation
1139 * of close-to-open would allow us to continue caching even if
1140 * another writer had changed the file, but some applications
1141 * depend on tighter cache coherency when writing.
1142 */
Fred Isamancd841602012-04-20 14:47:44 -04001143 status = NFS_PROTO(inode)->write_done(task, data);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001144 if (status != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001145 return status;
1146 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, data->res.count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001147
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001148#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001149 if (data->res.verf->committed < data->args.stable && task->tk_status >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 /* We tried a write call, but the server did not
1151 * commit data to stable storage even though we
1152 * requested it.
1153 * Note: There is a known bug in Tru64 < 5.0 in which
1154 * the server reports NFS_DATA_SYNC, but performs
1155 * NFS_FILE_SYNC. We therefore implement this checking
1156 * as a dprintk() in order to avoid filling syslog.
1157 */
1158 static unsigned long complain;
1159
Fred Isamana69aef12011-03-03 15:13:47 +00001160 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001162 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001164 NFS_SERVER(inode)->nfs_client->cl_hostname,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001165 data->res.verf->committed, data->args.stable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 complain = jiffies + 300 * HZ;
1167 }
1168 }
1169#endif
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001170
1171 /* Deal with the suid/sgid bit corner case */
1172 if (nfs_should_remove_suid(inode))
1173 nfs_mark_for_revalidate(inode);
Anna Schumaker0eecb212014-05-06 09:12:32 -04001174 return 0;
1175}
1176
1177/*
1178 * This function is called when the WRITE call is complete.
1179 */
1180static void nfs_writeback_result(struct rpc_task *task, struct nfs_pgio_data *data)
1181{
1182 struct nfs_pgio_args *argp = &data->args;
1183 struct nfs_pgio_res *resp = &data->res;
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001184
1185 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 static unsigned long complain;
1187
Fred Isaman6c75dc02012-04-20 14:47:47 -04001188 /* This a short write! */
Anna Schumaker0eecb212014-05-06 09:12:32 -04001189 nfs_inc_stats(data->header->inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001192 if (resp->count == 0) {
1193 if (time_before(complain, jiffies)) {
1194 printk(KERN_WARNING
1195 "NFS: Server wrote zero bytes, expected %u.\n",
1196 argp->count);
1197 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001199 nfs_set_pgio_error(data->header, -EIO, argp->offset);
1200 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001201 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001203 /* Was this an NFSv2 write or an NFSv3 stable write? */
1204 if (resp->verf->committed != NFS_UNSTABLE) {
1205 /* Resend from where the server left off */
1206 data->mds_offset += resp->count;
1207 argp->offset += resp->count;
1208 argp->pgbase += resp->count;
1209 argp->count -= resp->count;
1210 } else {
1211 /* Resend as a stable write in order to avoid
1212 * headaches in the case of a server crash.
1213 */
1214 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001216 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
1219
1220
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001221#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001222static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1223{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001224 int ret;
1225
Trond Myklebust71d0a612010-04-22 15:35:57 -04001226 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1227 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001228 if (!may_wait)
1229 return 0;
1230 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1231 NFS_INO_COMMIT,
1232 nfs_wait_bit_killable,
1233 TASK_KILLABLE);
1234 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001235}
1236
Fred Isamanf453a542012-04-20 14:47:54 -04001237static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001238{
1239 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001240 smp_mb__after_atomic();
Trond Myklebust71d0a612010-04-22 15:35:57 -04001241 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1242}
1243
Fred Isaman0b7c0152012-04-20 14:47:39 -04001244void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001246 put_nfs_open_context(data->context);
1247 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
Fred Isamane0c2b382011-03-23 13:27:53 +00001249EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
Fred Isaman0b7c0152012-04-20 14:47:39 -04001251int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001252 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001253 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254{
Trond Myklebust07737692007-10-25 18:42:54 -04001255 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001256 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001257 struct rpc_message msg = {
1258 .rpc_argp = &data->args,
1259 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001260 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001261 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001262 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001263 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001264 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001265 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001266 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001267 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001268 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001269 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001270 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001271 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001272 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001273 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Chuck Levera3f565b2007-01-31 12:14:01 -05001275 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001276
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001277 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1278 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1279
Trond Myklebust07737692007-10-25 18:42:54 -04001280 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001281 if (IS_ERR(task))
1282 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001283 if (how & FLUSH_SYNC)
1284 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001285 rpc_put_task(task);
1286 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
Fred Isamane0c2b382011-03-23 13:27:53 +00001288EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289
1290/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001291 * Set up the argument/result storage required for the RPC call.
1292 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001293void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001294 struct list_head *head,
1295 struct pnfs_layout_segment *lseg,
1296 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001297{
1298 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001299 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001300
1301 /* Set up the RPC argument and reply structs
1302 * NB: take care not to mess about with data->commit et al. */
1303
1304 list_splice_init(head, &data->pages);
1305
1306 data->inode = inode;
1307 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001308 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001309 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001310 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001311 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001312
1313 data->args.fh = NFS_FH(data->inode);
1314 /* Note: we always request a commit of the entire inode */
1315 data->args.offset = 0;
1316 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001317 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001318 data->res.fattr = &data->fattr;
1319 data->res.verf = &data->verf;
1320 nfs_fattr_init(&data->fattr);
1321}
Fred Isamane0c2b382011-03-23 13:27:53 +00001322EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001323
Fred Isamane0c2b382011-03-23 13:27:53 +00001324void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001325 struct pnfs_layout_segment *lseg,
1326 struct nfs_commit_info *cinfo)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001327{
1328 struct nfs_page *req;
1329
1330 while (!list_empty(page_list)) {
1331 req = nfs_list_entry(page_list->next);
1332 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -04001333 nfs_mark_request_commit(req, lseg, cinfo);
Fred Isaman56f9cd62012-04-20 14:47:56 -04001334 if (!cinfo->dreq) {
1335 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001336 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
Fred Isaman56f9cd62012-04-20 14:47:56 -04001337 BDI_RECLAIMABLE);
1338 }
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001339 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001340 }
1341}
Fred Isamane0c2b382011-03-23 13:27:53 +00001342EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001343
Fred Isaman9ace33c2011-03-23 13:27:45 +00001344/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 * Commit dirty pages
1346 */
1347static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001348nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1349 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001351 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001353 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 if (!data)
1356 goto out_bad;
1357
1358 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001359 nfs_init_commit(data, head, NULL, cinfo);
1360 atomic_inc(&cinfo->mds->rpcs_out);
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001361 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
1362 how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 out_bad:
Fred Isamanea2cf222012-04-20 14:47:53 -04001364 nfs_retry_commit(head, NULL, cinfo);
Fred Isamanf453a542012-04-20 14:47:54 -04001365 cinfo->completion_ops->error_cleanup(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 return -ENOMEM;
1367}
1368
1369/*
1370 * COMMIT call returned
1371 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001372static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001374 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Chuck Levera3f565b2007-01-31 12:14:01 -05001376 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 task->tk_pid, task->tk_status);
1378
Trond Myklebust788e7a82006-03-20 13:44:27 -05001379 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001380 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001381}
1382
Fred Isamanf453a542012-04-20 14:47:54 -04001383static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001384{
Fred Isaman5917ce82011-03-23 13:27:48 +00001385 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001386 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001387 struct nfs_commit_info cinfo;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 while (!list_empty(&data->pages)) {
1390 req = nfs_list_entry(data->pages.next);
1391 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001392 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393
Niels de Vos1e8968c2013-12-17 18:20:16 +01001394 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001395 req->wb_context->dentry->d_sb->s_id,
Niels de Vos1e8968c2013-12-17 18:20:16 +01001396 (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 req->wb_bytes,
1398 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001399 if (status < 0) {
1400 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001402 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 goto next;
1404 }
1405
1406 /* Okay, COMMIT succeeded, apparently. Check the verifier
1407 * returned by the server against all stored verfs. */
Trond Myklebust2f2c63b2012-06-08 11:56:09 -04001408 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 /* We have a match */
1410 nfs_inode_remove_request(req);
1411 dprintk(" OK\n");
1412 goto next;
1413 }
1414 /* We have a mismatch. Write the page again */
1415 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001416 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001417 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001419 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 }
Fred Isamanf453a542012-04-20 14:47:54 -04001421 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1422 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1423 nfs_commit_clear_lock(NFS_I(data->inode));
Fred Isaman5917ce82011-03-23 13:27:48 +00001424}
1425
1426static void nfs_commit_release(void *calldata)
1427{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001428 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001429
Fred Isamanf453a542012-04-20 14:47:54 -04001430 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001431 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001433
1434static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001435 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001436 .rpc_call_done = nfs_commit_done,
1437 .rpc_release = nfs_commit_release,
1438};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Fred Isamanf453a542012-04-20 14:47:54 -04001440static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1441 .completion = nfs_commit_release_pages,
1442 .error_cleanup = nfs_commit_clear_lock,
1443};
1444
Fred Isaman1763da12012-04-20 14:47:57 -04001445int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1446 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001447{
1448 int status;
1449
Fred Isamanea2cf222012-04-20 14:47:53 -04001450 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001451 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001452 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001453 return status;
1454}
1455
Trond Myklebustb608b282010-07-30 15:31:54 -04001456int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001459 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001460 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001461 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Trond Myklebustb8413f92011-03-21 15:37:01 -04001463 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1464 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001465 goto out_mark_dirty;
Fred Isamanea2cf222012-04-20 14:47:53 -04001466 nfs_init_cinfo_from_inode(&cinfo, inode);
1467 res = nfs_scan_commit(inode, &head, &cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001469 int error;
1470
Fred Isamanea2cf222012-04-20 14:47:53 -04001471 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001472 if (error < 0)
1473 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001474 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001475 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001476 error = wait_on_bit(&NFS_I(inode)->flags,
1477 NFS_INO_COMMIT,
1478 nfs_wait_bit_killable,
1479 TASK_KILLABLE);
1480 if (error < 0)
1481 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001482 } else
1483 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001484 return res;
1485 /* Note: If we exit without ensuring that the commit is complete,
1486 * we must mark the inode as dirty. Otherwise, future calls to
1487 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1488 * that the data is on the disk.
1489 */
1490out_mark_dirty:
1491 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 return res;
1493}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001494
1495static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1496{
Trond Myklebust420e3642010-02-19 17:00:02 -08001497 struct nfs_inode *nfsi = NFS_I(inode);
1498 int flags = FLUSH_SYNC;
1499 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001500
Jeff Layton3236c3e2011-10-11 09:49:21 -04001501 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001502 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e2011-10-11 09:49:21 -04001503 return ret;
1504
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001505 if (wbc->sync_mode == WB_SYNC_NONE) {
1506 /* Don't commit yet if this is a non-blocking flush and there
1507 * are a lot of outstanding writes for this mapping.
1508 */
Fred Isamanea2cf222012-04-20 14:47:53 -04001509 if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001510 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001511
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001512 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001513 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001514 }
1515
Trond Myklebust420e3642010-02-19 17:00:02 -08001516 ret = nfs_commit_inode(inode, flags);
1517 if (ret >= 0) {
1518 if (wbc->sync_mode == WB_SYNC_NONE) {
1519 if (ret < wbc->nr_to_write)
1520 wbc->nr_to_write -= ret;
1521 else
1522 wbc->nr_to_write = 0;
1523 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001524 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001525 }
1526out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001527 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1528 return ret;
1529}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001530#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001531static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1532{
1533 return 0;
1534}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535#endif
1536
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001537int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1538{
Bryan Schumakera8d8f022012-06-20 15:53:49 -04001539 return nfs_commit_unstable_pages(inode, wbc);
1540}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001541EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001542
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001543/*
1544 * flush the inode to disk.
1545 */
1546int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001547{
1548 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001549 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001550 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001551 .range_start = 0,
1552 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001553 };
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001554 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001555
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001556 trace_nfs_writeback_inode_enter(inode);
1557
1558 ret = sync_inode(inode, &wbc);
1559
1560 trace_nfs_writeback_inode_exit(inode, ret);
1561 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001562}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001563EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001564
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001565int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1566{
1567 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001568 int ret = 0;
1569
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001570 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001571 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001572 req = nfs_page_find_request(page);
1573 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001574 break;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001575 if (nfs_lock_request(req)) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001576 nfs_clear_request_commit(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001577 nfs_inode_remove_request(req);
1578 /*
1579 * In case nfs_inode_remove_request has marked the
1580 * page as being dirty
1581 */
1582 cancel_dirty_page(page, PAGE_CACHE_SIZE);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001583 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001584 break;
1585 }
1586 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001587 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001588 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001589 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001590 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001591 return ret;
1592}
1593
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001594/*
1595 * Write back all requests on one page - we do this before reading it.
1596 */
1597int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001598{
Mel Gorman29418aa2012-07-31 16:45:10 -07001599 loff_t range_start = page_file_offset(page);
Trond Myklebust1c759502006-10-09 16:18:38 -04001600 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001601 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001602 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001603 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001604 .range_start = range_start,
1605 .range_end = range_end,
1606 };
1607 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001608
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001609 trace_nfs_writeback_page_enter(inode);
1610
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001611 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001612 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001613 if (clear_page_dirty_for_io(page)) {
1614 ret = nfs_writepage_locked(page, &wbc);
1615 if (ret < 0)
1616 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001617 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001618 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001619 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001620 if (!PagePrivate(page))
1621 break;
1622 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001623 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001624 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001625 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001626out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001627 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001628 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001629}
1630
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001631#ifdef CONFIG_MIGRATION
1632int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001633 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001634{
Jeff Layton2da95652011-10-12 10:57:42 -04001635 /*
1636 * If PagePrivate is set, then the page is currently associated with
1637 * an in-progress read or write request. Don't try to migrate it.
1638 *
1639 * FIXME: we could do this in principle, but we'll need a way to ensure
1640 * that we can safely release the inode reference while holding
1641 * the page lock.
1642 */
1643 if (PagePrivate(page))
1644 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001645
David Howells8c209ce2012-12-05 13:34:49 +00001646 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1647 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001648
Mel Gormana6bc32b2012-01-12 17:19:43 -08001649 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001650}
1651#endif
1652
David Howellsf7b422b2006-06-09 09:34:33 -04001653int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
1655 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Anna Schumakerc0752cd2014-05-06 09:12:27 -04001656 sizeof(struct nfs_rw_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001658 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 if (nfs_wdata_cachep == NULL)
1660 return -ENOMEM;
1661
Matthew Dobson93d23412006-03-26 01:37:50 -08001662 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1663 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001665 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Fred Isaman0b7c0152012-04-20 14:47:39 -04001667 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1668 sizeof(struct nfs_commit_data),
1669 0, SLAB_HWCACHE_ALIGN,
1670 NULL);
1671 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001672 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001673
Matthew Dobson93d23412006-03-26 01:37:50 -08001674 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08001675 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001677 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
Peter Zijlstra89a09142007-03-16 13:38:26 -08001679 /*
1680 * NFS congestion size, scale with available memory.
1681 *
1682 * 64MB: 8192k
1683 * 128MB: 11585k
1684 * 256MB: 16384k
1685 * 512MB: 23170k
1686 * 1GB: 32768k
1687 * 2GB: 46340k
1688 * 4GB: 65536k
1689 * 8GB: 92681k
1690 * 16GB: 131072k
1691 *
1692 * This allows larger machines to have larger/more transfers.
1693 * Limit the default to 256M
1694 */
1695 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1696 if (nfs_congestion_kb > 256*1024)
1697 nfs_congestion_kb = 256*1024;
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04001700
1701out_destroy_commit_cache:
1702 kmem_cache_destroy(nfs_cdata_cachep);
1703out_destroy_write_mempool:
1704 mempool_destroy(nfs_wdata_mempool);
1705out_destroy_write_cache:
1706 kmem_cache_destroy(nfs_wdata_cachep);
1707 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708}
1709
David Brownell266bee82006-06-27 12:59:15 -07001710void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711{
1712 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04001713 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001715 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716}
1717
Anna Schumaker4a0de552014-05-06 09:12:30 -04001718static const struct nfs_rw_ops nfs_rw_write_ops = {
Anna Schumakera4cdda52014-05-06 09:12:31 -04001719 .rw_mode = FMODE_WRITE,
Anna Schumaker4a0de552014-05-06 09:12:30 -04001720 .rw_alloc_header = nfs_writehdr_alloc,
1721 .rw_free_header = nfs_writehdr_free,
Anna Schumakera4cdda52014-05-06 09:12:31 -04001722 .rw_release = nfs_writeback_release_common,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001723 .rw_done = nfs_writeback_done,
1724 .rw_result = nfs_writeback_result,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001725 .rw_initiate = nfs_initiate_write,
Anna Schumaker4a0de552014-05-06 09:12:30 -04001726};