blob: 25ba3830ec8bdff39a1f03f8db308618a1dc8afd [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);
Fred Isaman6c75dc02012-04-20 14:47:47 -040045static const struct rpc_call_ops nfs_write_common_ops;
Trond Myklebust788e7a82006-03-20 13:44:27 -050046static const struct rpc_call_ops nfs_commit_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -040047static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
Fred Isamanf453a542012-04-20 14:47:54 -040048static const struct nfs_commit_completion_ops nfs_commit_completion_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
Fred Isaman6c75dc02012-04-20 14:47:47 -040073struct nfs_write_header *nfs_writehdr_alloc(void)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050074{
Mel Gorman192e5012012-07-31 16:45:16 -070075 struct nfs_write_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050076
77 if (p) {
Fred Isamancd841602012-04-20 14:47:44 -040078 struct nfs_pgio_header *hdr = &p->header;
Fred Isamancd841602012-04-20 14:47:44 -040079
Trond Myklebust3feb2d42006-03-20 13:44:37 -050080 memset(p, 0, sizeof(*p));
Fred Isamancd841602012-04-20 14:47:44 -040081 INIT_LIST_HEAD(&hdr->pages);
Fred Isaman6c75dc02012-04-20 14:47:47 -040082 INIT_LIST_HEAD(&hdr->rpc_list);
83 spin_lock_init(&hdr->lock);
84 atomic_set(&hdr->refcnt, 0);
Trond Myklebust9bce0082012-06-05 18:32:03 -040085 hdr->verf = &p->verf;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050086 }
87 return p;
88}
Bryan Schumaker89d77c82012-07-30 16:05:25 -040089EXPORT_SYMBOL_GPL(nfs_writehdr_alloc);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050090
Fred Isaman1763da12012-04-20 14:47:57 -040091static struct nfs_write_data *nfs_writedata_alloc(struct nfs_pgio_header *hdr,
92 unsigned int pagecount)
Fred Isaman6c75dc02012-04-20 14:47:47 -040093{
94 struct nfs_write_data *data, *prealloc;
95
96 prealloc = &container_of(hdr, struct nfs_write_header, header)->rpc_data;
97 if (prealloc->header == NULL)
98 data = prealloc;
99 else
100 data = kzalloc(sizeof(*data), GFP_KERNEL);
101 if (!data)
102 goto out;
103
104 if (nfs_pgarray_set(&data->pages, pagecount)) {
105 data->header = hdr;
106 atomic_inc(&hdr->refcnt);
107 } else {
108 if (data != prealloc)
109 kfree(data);
110 data = NULL;
111 }
112out:
113 return data;
114}
115
Fred Isamancd841602012-04-20 14:47:44 -0400116void nfs_writehdr_free(struct nfs_pgio_header *hdr)
Trond Myklebust3feb2d42006-03-20 13:44:37 -0500117{
Fred Isamancd841602012-04-20 14:47:44 -0400118 struct nfs_write_header *whdr = container_of(hdr, struct nfs_write_header, header);
119 mempool_free(whdr, nfs_wdata_mempool);
Trond Myklebust3feb2d42006-03-20 13:44:37 -0500120}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400121EXPORT_SYMBOL_GPL(nfs_writehdr_free);
Trond Myklebust3feb2d42006-03-20 13:44:37 -0500122
Trond Myklebustdce81292011-07-13 15:59:19 -0400123void nfs_writedata_release(struct nfs_write_data *wdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Fred Isaman6c75dc02012-04-20 14:47:47 -0400125 struct nfs_pgio_header *hdr = wdata->header;
126 struct nfs_write_header *write_header = container_of(hdr, struct nfs_write_header, header);
127
Trond Myklebust383ba712008-02-19 20:04:20 -0500128 put_nfs_open_context(wdata->args.context);
Fred Isaman30dd3742012-04-20 14:47:45 -0400129 if (wdata->pages.pagevec != wdata->pages.page_array)
130 kfree(wdata->pages.pagevec);
Trond Myklebust6db6dd72013-01-04 12:47:04 -0500131 if (wdata == &write_header->rpc_data) {
Fred Isaman6c75dc02012-04-20 14:47:47 -0400132 wdata->header = NULL;
Trond Myklebust6db6dd72013-01-04 12:47:04 -0500133 wdata = NULL;
134 }
Fred Isaman6c75dc02012-04-20 14:47:47 -0400135 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -0400136 hdr->completion_ops->completion(hdr);
Trond Myklebust6db6dd72013-01-04 12:47:04 -0500137 /* Note: we only free the rpc_task after callbacks are done.
138 * See the comment in rpc_free_task() for why
139 */
140 kfree(wdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400142EXPORT_SYMBOL_GPL(nfs_writedata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Trond Myklebust7b159fc2007-07-25 14:09:54 -0400144static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
145{
146 ctx->error = error;
147 smp_wmb();
148 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
149}
150
Mel Gorman29418aa2012-07-31 16:45:10 -0700151static struct nfs_page *
152nfs_page_find_request_locked(struct nfs_inode *nfsi, struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -0500153{
154 struct nfs_page *req = NULL;
155
Mel Gorman29418aa2012-07-31 16:45:10 -0700156 if (PagePrivate(page))
Trond Myklebust277459d2006-12-05 00:35:35 -0500157 req = (struct nfs_page *)page_private(page);
Mel Gorman29418aa2012-07-31 16:45:10 -0700158 else if (unlikely(PageSwapCache(page))) {
159 struct nfs_page *freq, *t;
160
161 /* Linearly search the commit list for the correct req */
162 list_for_each_entry_safe(freq, t, &nfsi->commit_info.list, wb_list) {
163 if (freq->wb_page == page) {
164 req = freq;
165 break;
166 }
167 }
Trond Myklebust277459d2006-12-05 00:35:35 -0500168 }
Mel Gorman29418aa2012-07-31 16:45:10 -0700169
170 if (req)
171 kref_get(&req->wb_kref);
172
Trond Myklebust277459d2006-12-05 00:35:35 -0500173 return req;
174}
175
176static struct nfs_page *nfs_page_find_request(struct page *page)
177{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700178 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500179 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500180
Trond Myklebust587142f2007-07-02 09:57:54 -0400181 spin_lock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700182 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400183 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500184 return req;
185}
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/* Adjust the file length if we're writing beyond the end */
188static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
189{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700190 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400191 loff_t end, i_size;
192 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Trond Myklebusta3d01452008-06-11 12:21:19 -0400194 spin_lock(&inode->i_lock);
195 i_size = i_size_read(inode);
196 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700197 if (i_size > 0 && page_file_index(page) < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400198 goto out;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700199 end = page_file_offset(page) + ((loff_t)offset+count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400201 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400203 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
204out:
205 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
Trond Myklebusta301b772007-02-06 11:07:15 -0800208/* A writeback failed: mark the page as bad, and invalidate the page cache */
209static void nfs_set_pageerror(struct page *page)
210{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700211 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
Trond Myklebusta301b772007-02-06 11:07:15 -0800212}
213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214/* We can set the PG_uptodate flag if we see that a write request
215 * covers the full page.
216 */
217static void nfs_mark_uptodate(struct page *page, unsigned int base, unsigned int count)
218{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 if (PageUptodate(page))
220 return;
221 if (base != 0)
222 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500223 if (count != nfs_page_length(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 return;
Trond Myklebust49a70f22006-12-05 00:35:38 -0500225 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228static int wb_priority(struct writeback_control *wbc)
229{
230 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400231 return FLUSH_HIGHPRI | FLUSH_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100232 if (wbc->for_kupdate || wbc->for_background)
Trond Myklebustb31268a2011-03-21 17:02:00 -0400233 return FLUSH_LOWPRI | FLUSH_COND_STABLE;
234 return FLUSH_COND_STABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
237/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800238 * NFS congestion control
239 */
240
241int nfs_congestion_kb;
242
243#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
244#define NFS_CONGESTION_OFF_THRESH \
245 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
246
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400247static void nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800248{
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400249 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400250 int ret = test_set_page_writeback(page);
251
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400252 WARN_ON_ONCE(ret != 0);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800253
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400254 if (atomic_long_inc_return(&nfss->writeback) >
255 NFS_CONGESTION_ON_THRESH) {
256 set_bdi_congested(&nfss->backing_dev_info,
257 BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800258 }
259}
260
261static void nfs_end_page_writeback(struct page *page)
262{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700263 struct inode *inode = page_file_mapping(page)->host;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800264 struct nfs_server *nfss = NFS_SERVER(inode);
265
266 end_page_writeback(page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700267 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200268 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800269}
270
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400271static struct nfs_page *nfs_find_and_lock_request(struct page *page, bool nonblock)
Trond Myklebuste261f512006-12-05 00:35:41 -0500272{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700273 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500274 struct nfs_page *req;
Trond Myklebuste261f512006-12-05 00:35:41 -0500275 int ret;
276
Trond Myklebust587142f2007-07-02 09:57:54 -0400277 spin_lock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400278 for (;;) {
Mel Gorman29418aa2012-07-31 16:45:10 -0700279 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400280 if (req == NULL)
281 break;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400282 if (nfs_lock_request(req))
Trond Myklebuste261f512006-12-05 00:35:41 -0500283 break;
284 /* Note: If we hold the page lock, as is the case in nfs_writepage,
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400285 * then the call to nfs_lock_request() will always
Trond Myklebuste261f512006-12-05 00:35:41 -0500286 * succeed provided that someone hasn't already marked the
287 * request as dirty (in which case we don't care).
288 */
Trond Myklebust587142f2007-07-02 09:57:54 -0400289 spin_unlock(&inode->i_lock);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400290 if (!nonblock)
291 ret = nfs_wait_on_request(req);
292 else
293 ret = -EAGAIN;
Trond Myklebuste261f512006-12-05 00:35:41 -0500294 nfs_release_request(req);
295 if (ret != 0)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400296 return ERR_PTR(ret);
Trond Myklebust587142f2007-07-02 09:57:54 -0400297 spin_lock(&inode->i_lock);
Trond Myklebuste261f512006-12-05 00:35:41 -0500298 }
Trond Myklebust587142f2007-07-02 09:57:54 -0400299 spin_unlock(&inode->i_lock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400300 return req;
301}
302
303/*
304 * Find an associated nfs write request, and prepare to flush it out
305 * May return an error if the user signalled nfs_wait_on_request().
306 */
307static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400308 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400309{
310 struct nfs_page *req;
311 int ret = 0;
312
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400313 req = nfs_find_and_lock_request(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400314 if (!req)
315 goto out;
316 ret = PTR_ERR(req);
317 if (IS_ERR(req))
318 goto out;
319
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400320 nfs_set_page_writeback(page);
321 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400322
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400323 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400324 if (!nfs_pageio_add_request(pgio, req)) {
325 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400326 ret = pgio->pg_error;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400327 }
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400328out:
329 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500330}
331
Trond Myklebustf758c8852007-07-22 19:27:32 -0400332static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
333{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700334 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400335 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400336
337 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
338 nfs_add_stats(inode, NFSIOS_WRITEPAGES, 1);
339
Mel Gormand56b4dd2012-07-31 16:45:06 -0700340 nfs_pageio_cond_complete(pgio, page_file_index(page));
Wu Fengguang1b430be2010-10-26 14:21:26 -0700341 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400342 if (ret == -EAGAIN) {
343 redirty_page_for_writepage(wbc, page);
344 ret = 0;
345 }
346 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400347}
348
Trond Myklebuste261f512006-12-05 00:35:41 -0500349/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 * Write an mmapped page to the server.
351 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500352static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400354 struct nfs_pageio_descriptor pgio;
Trond Myklebuste261f512006-12-05 00:35:41 -0500355 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200357 nfs_pageio_init_write(&pgio, page->mapping->host, wb_priority(wbc),
358 false, &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400359 err = nfs_do_writepage(page, wbc, &pgio);
360 nfs_pageio_complete(&pgio);
361 if (err < 0)
362 return err;
363 if (pgio.pg_error < 0)
364 return pgio.pg_error;
365 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500366}
367
368int nfs_writepage(struct page *page, struct writeback_control *wbc)
369{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400370 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500371
Trond Myklebustf758c8852007-07-22 19:27:32 -0400372 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400374 return ret;
375}
376
377static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
378{
379 int ret;
380
381 ret = nfs_do_writepage(page, wbc, data);
382 unlock_page(page);
383 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
387{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400389 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400390 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 int err;
392
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400393 /* Stop dirtying of new pages while we sync */
394 err = wait_on_bit_lock(bitlock, NFS_INO_FLUSHING,
395 nfs_wait_bit_killable, TASK_KILLABLE);
396 if (err)
397 goto out_err;
398
Chuck Lever91d5b472006-03-20 13:44:14 -0500399 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
400
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200401 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
402 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400403 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400404 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400405
406 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
407 smp_mb__after_clear_bit();
408 wake_up_bit(bitlock, NFS_INO_FLUSHING);
409
Trond Myklebustf758c8852007-07-22 19:27:32 -0400410 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400411 goto out_err;
412 err = pgio.pg_error;
413 if (err < 0)
414 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400415 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400416out_err:
417 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418}
419
420/*
421 * Insert a write request into an inode
422 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500423static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424{
425 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400426
427 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400428 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400429
430 spin_lock(&inode->i_lock);
Bryan Schumaker011e2a72012-06-20 15:53:43 -0400431 if (!nfsi->npages && NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700432 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700433 /*
434 * Swap-space should not get truncated. Hence no need to plug the race
435 * with invalidate/truncate.
436 */
437 if (likely(!PageSwapCache(req->wb_page))) {
438 set_bit(PG_MAPPED, &req->wb_flags);
439 SetPagePrivate(req->wb_page);
440 set_page_private(req->wb_page, (unsigned long)req);
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 nfsi->npages++;
Trond Myklebustc03b4022007-06-17 13:26:38 -0400443 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400444 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
446
447/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800448 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 */
450static void nfs_inode_remove_request(struct nfs_page *req)
451{
Al Viro3d4ff432011-06-22 18:40:12 -0400452 struct inode *inode = req->wb_context->dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 struct nfs_inode *nfsi = NFS_I(inode);
454
Trond Myklebust587142f2007-07-02 09:57:54 -0400455 spin_lock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700456 if (likely(!PageSwapCache(req->wb_page))) {
457 set_page_private(req->wb_page, 0);
458 ClearPagePrivate(req->wb_page);
459 clear_bit(PG_MAPPED, &req->wb_flags);
460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 nfsi->npages--;
Trond Myklebust4d65c522011-03-25 14:15:11 -0400462 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 nfs_release_request(req);
464}
465
Trond Myklebust61822ab2006-12-05 00:35:42 -0500466static void
Fred6d884e82008-03-19 11:24:38 -0400467nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500468{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500469 __set_page_dirty_nobuffers(req->wb_page);
470}
471
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400472#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400473/**
474 * nfs_request_add_commit_list - add request to a commit list
475 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400476 * @dst: commit list head
477 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400478 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400479 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400480 * number of outstanding requests requiring a commit as well as
481 * the MM page stats.
482 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400483 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400484 * holding the nfs_page lock.
485 */
486void
Fred Isamanea2cf222012-04-20 14:47:53 -0400487nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
488 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400489{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400490 set_bit(PG_CLEAN, &(req)->wb_flags);
Fred Isamanea2cf222012-04-20 14:47:53 -0400491 spin_lock(cinfo->lock);
492 nfs_list_add_request(req, dst);
493 cinfo->mds->ncommit++;
494 spin_unlock(cinfo->lock);
Fred Isaman56f9cd62012-04-20 14:47:56 -0400495 if (!cinfo->dreq) {
496 inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700497 inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
Fred Isaman56f9cd62012-04-20 14:47:56 -0400498 BDI_RECLAIMABLE);
499 __mark_inode_dirty(req->wb_context->dentry->d_inode,
500 I_DIRTY_DATASYNC);
501 }
Trond Myklebust8dd37752012-03-15 17:16:40 -0400502}
503EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
504
505/**
506 * nfs_request_remove_commit_list - Remove request from a commit list
507 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400508 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400509 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400510 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400511 * number of outstanding requests requiring a commit
512 * It does not update the MM page stats.
513 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400514 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400515 */
516void
Fred Isamanea2cf222012-04-20 14:47:53 -0400517nfs_request_remove_commit_list(struct nfs_page *req,
518 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400519{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400520 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
521 return;
522 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400523 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400524}
525EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
526
Fred Isamanea2cf222012-04-20 14:47:53 -0400527static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
528 struct inode *inode)
529{
530 cinfo->lock = &inode->i_lock;
531 cinfo->mds = &NFS_I(inode)->commit_info;
532 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400533 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400534 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400535}
536
537void nfs_init_cinfo(struct nfs_commit_info *cinfo,
538 struct inode *inode,
539 struct nfs_direct_req *dreq)
540{
Fred Isaman1763da12012-04-20 14:47:57 -0400541 if (dreq)
542 nfs_init_cinfo_from_dreq(cinfo, dreq);
543 else
544 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400545}
546EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400547
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548/*
549 * Add a request to the inode's commit list.
550 */
Fred Isaman1763da12012-04-20 14:47:57 -0400551void
Fred Isamanea2cf222012-04-20 14:47:53 -0400552nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
553 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
Fred Isamanea2cf222012-04-20 14:47:53 -0400555 if (pnfs_mark_request_commit(req, lseg, cinfo))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400556 return;
Fred Isamanea2cf222012-04-20 14:47:53 -0400557 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400559
Fred Isamand6d6dc72012-03-08 17:29:35 -0500560static void
561nfs_clear_page_commit(struct page *page)
562{
563 dec_zone_page_state(page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700564 dec_bdi_stat(page_file_mapping(page)->backing_dev_info, BDI_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500565}
566
Trond Myklebust8dd37752012-03-15 17:16:40 -0400567static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400568nfs_clear_request_commit(struct nfs_page *req)
569{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400570 if (test_bit(PG_CLEAN, &req->wb_flags)) {
571 struct inode *inode = req->wb_context->dentry->d_inode;
Fred Isamanea2cf222012-04-20 14:47:53 -0400572 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400573
Fred Isamanea2cf222012-04-20 14:47:53 -0400574 nfs_init_cinfo_from_inode(&cinfo, inode);
575 if (!pnfs_clear_request_commit(req, &cinfo)) {
576 spin_lock(cinfo.lock);
577 nfs_request_remove_commit_list(req, &cinfo);
578 spin_unlock(cinfo.lock);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400579 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500580 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400581 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400582}
583
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400584static inline
585int nfs_write_need_commit(struct nfs_write_data *data)
586{
Fred Isaman465d5242011-03-23 13:27:44 +0000587 if (data->verf.committed == NFS_DATA_SYNC)
Fred Isamancd841602012-04-20 14:47:44 -0400588 return data->header->lseg == NULL;
589 return data->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400590}
591
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400592#else
Bryan Schumaker68cd6fa2012-04-30 14:30:22 -0400593static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
594 struct inode *inode)
595{
596}
597
598void nfs_init_cinfo(struct nfs_commit_info *cinfo,
599 struct inode *inode,
600 struct nfs_direct_req *dreq)
601{
602}
603
Fred Isaman1763da12012-04-20 14:47:57 -0400604void
Fred Isamanea2cf222012-04-20 14:47:53 -0400605nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
606 struct nfs_commit_info *cinfo)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400607{
608}
609
Trond Myklebust8dd37752012-03-15 17:16:40 -0400610static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400611nfs_clear_request_commit(struct nfs_page *req)
612{
Trond Myklebuste468bae2008-06-13 13:25:22 -0400613}
614
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400615static inline
616int nfs_write_need_commit(struct nfs_write_data *data)
617{
618 return 0;
619}
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621#endif
622
Fred Isaman061ae2e2012-04-20 14:47:48 -0400623static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400624{
Fred Isamanea2cf222012-04-20 14:47:53 -0400625 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400626 unsigned long bytes = 0;
627
628 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
629 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400630 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400631 while (!list_empty(&hdr->pages)) {
632 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400633
634 bytes += req->wb_bytes;
635 nfs_list_remove_request(req);
636 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
637 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400638 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400639 nfs_context_set_write_error(req->wb_context, hdr->error);
640 goto remove_req;
641 }
642 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags)) {
643 nfs_mark_request_dirty(req);
644 goto next;
645 }
646 if (test_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags)) {
Trond Myklebust2f2c63b2012-06-08 11:56:09 -0400647 memcpy(&req->wb_verf, &hdr->verf->verifier, sizeof(req->wb_verf));
Fred Isamanea2cf222012-04-20 14:47:53 -0400648 nfs_mark_request_commit(req, hdr->lseg, &cinfo);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400649 goto next;
650 }
651remove_req:
652 nfs_inode_remove_request(req);
653next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400654 nfs_unlock_request(req);
Trond Myklebustd1182b32012-05-09 13:37:43 -0400655 nfs_end_page_writeback(req->wb_page);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400656 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400657 }
658out:
659 hdr->release(hdr);
660}
661
Bryan Schumaker89d77c82012-07-30 16:05:25 -0400662#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Fred Isamanea2cf222012-04-20 14:47:53 -0400663static unsigned long
664nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400665{
Fred Isamanea2cf222012-04-20 14:47:53 -0400666 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400667}
668
Fred Isamanea2cf222012-04-20 14:47:53 -0400669/* cinfo->lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400670int
Fred Isamanea2cf222012-04-20 14:47:53 -0400671nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
672 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500673{
674 struct nfs_page *req, *tmp;
675 int ret = 0;
676
677 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400678 if (!nfs_lock_request(req))
679 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400680 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -0400681 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400682 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400683 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400684 nfs_list_add_request(req, dst);
685 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400686 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400687 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500688 }
689 return ret;
690}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692/*
693 * nfs_scan_commit - Scan an inode for commit requests
694 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400695 * @dst: mds destination list
696 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 *
698 * Moves requests from the inode's 'commit' request list.
699 * The requests are *not* checked to ensure that they form a contiguous set.
700 */
Fred Isaman1763da12012-04-20 14:47:57 -0400701int
Fred Isamanea2cf222012-04-20 14:47:53 -0400702nfs_scan_commit(struct inode *inode, struct list_head *dst,
703 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
Fred Isamand6d6dc72012-03-08 17:29:35 -0500705 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400706
Fred Isamanea2cf222012-04-20 14:47:53 -0400707 spin_lock(cinfo->lock);
708 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400709 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500710
Fred Isamanea2cf222012-04-20 14:47:53 -0400711 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
712 cinfo, max);
713 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500714 }
Fred Isamanea2cf222012-04-20 14:47:53 -0400715 spin_unlock(cinfo->lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800716 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500718
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500719#else
Fred Isamanea2cf222012-04-20 14:47:53 -0400720static unsigned long nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400721{
722 return 0;
723}
724
Fred Isaman1763da12012-04-20 14:47:57 -0400725int nfs_scan_commit(struct inode *inode, struct list_head *dst,
726 struct nfs_commit_info *cinfo)
Trond Myklebustc42de9d2006-03-20 13:44:51 -0500727{
728 return 0;
729}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730#endif
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400733 * Search for an existing write request, and attempt to update
734 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400736 * If the attempt fails, then the existing request is flushed out
737 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 */
Trond Myklebuste7d39062008-06-13 12:12:32 -0400739static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
740 struct page *page,
741 unsigned int offset,
742 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
Trond Myklebuste7d39062008-06-13 12:12:32 -0400744 struct nfs_page *req;
745 unsigned int rqend;
746 unsigned int end;
747 int error;
748
749 if (!PagePrivate(page))
750 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400753 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 for (;;) {
Mel Gorman29418aa2012-07-31 16:45:10 -0700756 req = nfs_page_find_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400757 if (req == NULL)
758 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -0500759
Trond Myklebuste7d39062008-06-13 12:12:32 -0400760 rqend = req->wb_offset + req->wb_bytes;
761 /*
762 * Tell the caller to flush out the request if
763 * the offsets are non-contiguous.
764 * Note: nfs_flush_incompatible() will already
765 * have flushed out requests having wrong owners.
766 */
Trond Myklebuste468bae2008-06-13 13:25:22 -0400767 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -0400768 || end < req->wb_offset)
769 goto out_flushme;
770
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400771 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Trond Myklebuste7d39062008-06-13 12:12:32 -0400774 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -0400775 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400776 error = nfs_wait_on_request(req);
777 nfs_release_request(req);
778 if (error != 0)
779 goto out_err;
780 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 }
782
783 /* Okay, the request matches. Update the region */
784 if (offset < req->wb_offset) {
785 req->wb_offset = offset;
786 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 if (end > rqend)
789 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400790 else
791 req->wb_bytes = rqend - req->wb_offset;
792out_unlock:
793 spin_unlock(&inode->i_lock);
Fred Isamanca138f32012-04-05 15:26:36 -0400794 if (req)
795 nfs_clear_request_commit(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400796 return req;
797out_flushme:
798 spin_unlock(&inode->i_lock);
799 nfs_release_request(req);
800 error = nfs_wb_page(inode, page);
801out_err:
802 return ERR_PTR(error);
803}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Trond Myklebuste7d39062008-06-13 12:12:32 -0400805/*
806 * Try to update an existing write request, or create one if there is none.
807 *
808 * Note: Should always be called with the Page Lock held to prevent races
809 * if we have to add a new request. Also assumes that the caller has
810 * already called nfs_flush_incompatible() if necessary.
811 */
812static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
813 struct page *page, unsigned int offset, unsigned int bytes)
814{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700815 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400816 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -0400817
818 req = nfs_try_to_update_request(inode, page, offset, bytes);
819 if (req != NULL)
820 goto out;
821 req = nfs_create_request(ctx, inode, page, offset, bytes);
822 if (IS_ERR(req))
823 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500824 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -0400825out:
Trond Myklebust61e930a2007-10-18 17:08:05 -0400826 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827}
828
Trond Myklebuste7d39062008-06-13 12:12:32 -0400829static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
830 unsigned int offset, unsigned int count)
831{
832 struct nfs_page *req;
833
834 req = nfs_setup_write_request(ctx, page, offset, count);
835 if (IS_ERR(req))
836 return PTR_ERR(req);
837 /* Update file length */
838 nfs_grow_file(page, offset, count);
839 nfs_mark_uptodate(page, req->wb_pgbase, req->wb_bytes);
Trond Myklebusta6305dd2010-04-09 19:07:08 -0400840 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400841 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400842 return 0;
843}
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845int nfs_flush_incompatible(struct file *file, struct page *page)
846{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400847 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -0400848 struct nfs_lock_context *l_ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -0500850 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 /*
852 * Look for a request corresponding to this page. If there
853 * is one, and it belongs to another file, we flush it out
854 * before we try to copy anything into the page. Do this
855 * due to the lack of an ACCESS-type call in NFSv2.
856 * Also do the same if we find a request from an existing
857 * dropped page.
858 */
Trond Myklebust1a545332006-12-05 00:35:40 -0500859 do {
860 req = nfs_page_find_request(page);
861 if (req == NULL)
862 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -0400863 l_ctx = req->wb_lock_context;
864 do_flush = req->wb_page != page || req->wb_context != ctx;
Trond Myklebust0f1d2602013-09-05 15:52:51 -0400865 if (l_ctx && ctx->dentry->d_inode->i_flock != NULL) {
Trond Myklebust2a369152012-08-13 18:54:45 -0400866 do_flush |= l_ctx->lockowner.l_owner != current->files
867 || l_ctx->lockowner.l_pid != current->tgid;
868 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -0500870 if (!do_flush)
871 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700872 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -0500873 } while (status == 0);
874 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877/*
Andy Adamsondc248262013-08-14 11:59:16 -0400878 * Avoid buffered writes when a open context credential's key would
879 * expire soon.
880 *
881 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
882 *
883 * Return 0 and set a credential flag which triggers the inode to flush
884 * and performs NFS_FILE_SYNC writes if the key will expired within
885 * RPC_KEY_EXPIRE_TIMEO.
886 */
887int
888nfs_key_timeout_notify(struct file *filp, struct inode *inode)
889{
890 struct nfs_open_context *ctx = nfs_file_open_context(filp);
891 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
892
893 return rpcauth_key_timeout_notify(auth, ctx->cred);
894}
895
896/*
897 * Test if the open context credential key is marked to expire soon.
898 */
899bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
900{
901 return rpcauth_cred_key_to_expire(ctx->cred);
902}
903
904/*
Trond Myklebust5d47a352008-02-07 17:24:07 -0500905 * If the page cache is marked as unsafe or invalid, then we can't rely on
906 * the PageUptodate() flag. In this case, we will need to turn off
907 * write optimisations that depend on the page contents being correct.
908 */
Trond Myklebust8d197a52012-04-29 12:50:01 -0400909static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -0500910{
Jeff Laytond529ef82014-01-27 13:46:15 -0500911 struct nfs_inode *nfsi = NFS_I(inode);
912
Trond Myklebust8d197a52012-04-29 12:50:01 -0400913 if (nfs_have_delegated_attributes(inode))
914 goto out;
Jeff Laytond529ef82014-01-27 13:46:15 -0500915 if (nfsi->cache_validity & (NFS_INO_INVALID_DATA|NFS_INO_REVAL_PAGECACHE))
916 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -0500917 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -0500918 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -0400919 return false;
920out:
921 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -0500922}
923
Scott Mayhewc7559662013-07-05 17:33:19 -0400924/* If we know the page is up to date, and we're not using byte range locks (or
925 * if we have the whole file locked for writing), it may be more efficient to
926 * extend the write to cover the entire page in order to avoid fragmentation
927 * inefficiencies.
928 *
Scott Mayhew263b4502014-01-17 15:12:05 -0500929 * If the file is opened for synchronous writes then we can just skip the rest
930 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -0400931 */
932static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
933{
934 if (file->f_flags & O_DSYNC)
935 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -0500936 if (!nfs_write_pageuptodate(page, inode))
937 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -0400938 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
939 return 1;
Scott Mayhew263b4502014-01-17 15:12:05 -0500940 if (inode->i_flock == NULL || (inode->i_flock->fl_start == 0 &&
Scott Mayhewc7559662013-07-05 17:33:19 -0400941 inode->i_flock->fl_end == OFFSET_MAX &&
Scott Mayhew263b4502014-01-17 15:12:05 -0500942 inode->i_flock->fl_type != F_RDLCK))
Scott Mayhewc7559662013-07-05 17:33:19 -0400943 return 1;
944 return 0;
945}
946
Trond Myklebust5d47a352008-02-07 17:24:07 -0500947/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 * Update and possibly write a cached page of an NFS file.
949 *
950 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
951 * things with a page scheduled for an RPC call (e.g. invalidate it).
952 */
953int nfs_updatepage(struct file *file, struct page *page,
954 unsigned int offset, unsigned int count)
955{
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400956 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -0700957 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 int status = 0;
959
Chuck Lever91d5b472006-03-20 13:44:14 -0500960 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
961
Al Viro6de14722013-09-16 10:53:17 -0400962 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
963 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Scott Mayhewc7559662013-07-05 17:33:19 -0400965 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -0500966 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
969
Trond Myklebuste21195a2006-12-05 00:35:39 -0500970 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -0400971 if (status < 0)
972 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -0700973 else
974 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Chuck Lever48186c72008-06-11 17:56:05 -0400976 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 return status;
979}
980
Trond Myklebust3ff75762007-07-14 15:40:00 -0400981static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982{
983 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
984 case FLUSH_HIGHPRI:
985 return RPC_PRIORITY_HIGH;
986 case FLUSH_LOWPRI:
987 return RPC_PRIORITY_LOW;
988 }
989 return RPC_PRIORITY_NORMAL;
990}
991
Fred Isamanc5996c42012-04-20 14:47:41 -0400992int nfs_initiate_write(struct rpc_clnt *clnt,
993 struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +0000994 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -0400995 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Fred Isamancd841602012-04-20 14:47:44 -0400997 struct inode *inode = data->header->inode;
Trond Myklebust3ff75762007-07-14 15:40:00 -0400998 int priority = flush_task_priority(how);
Trond Myklebust07737692007-10-25 18:42:54 -0400999 struct rpc_task *task;
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001000 struct rpc_message msg = {
1001 .rpc_argp = &data->args,
1002 .rpc_resp = &data->res,
Fred Isamancd841602012-04-20 14:47:44 -04001003 .rpc_cred = data->header->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001004 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001005 struct rpc_task_setup task_setup_data = {
Andy Adamsond138d5d2011-03-03 15:13:41 +00001006 .rpc_client = clnt,
Trond Myklebust07737692007-10-25 18:42:54 -04001007 .task = &data->task,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001008 .rpc_message = &msg,
Trond Myklebust84115e12007-07-14 15:39:59 -04001009 .callback_ops = call_ops,
1010 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001011 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001012 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001013 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001014 };
Trond Myklebust2c61be02010-04-09 19:54:50 -04001015 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Andy Adamsond138d5d2011-03-03 15:13:41 +00001017 /* Set up the initial task struct. */
1018 NFS_PROTO(inode)->write_setup(data, &msg);
1019
1020 dprintk("NFS: %5u initiated write call "
Niels de Vos1e8968c2013-12-17 18:20:16 +01001021 "(req %s/%llu, %u bytes @ offset %llu)\n",
Andy Adamsond138d5d2011-03-03 15:13:41 +00001022 data->task.tk_pid,
1023 inode->i_sb->s_id,
Niels de Vos1e8968c2013-12-17 18:20:16 +01001024 (unsigned long long)NFS_FILEID(inode),
Andy Adamsond138d5d2011-03-03 15:13:41 +00001025 data->args.count,
1026 (unsigned long long)data->args.offset);
1027
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001028 nfs4_state_protect_write(NFS_SERVER(inode)->nfs_client,
1029 &task_setup_data.rpc_client, &msg, data);
1030
Andy Adamsond138d5d2011-03-03 15:13:41 +00001031 task = rpc_run_task(&task_setup_data);
1032 if (IS_ERR(task)) {
1033 ret = PTR_ERR(task);
1034 goto out;
1035 }
1036 if (how & FLUSH_SYNC) {
1037 ret = rpc_wait_for_completion_task(task);
1038 if (ret == 0)
1039 ret = task->tk_status;
1040 }
1041 rpc_put_task(task);
1042out:
1043 return ret;
1044}
Fred Isamana69aef12011-03-03 15:13:47 +00001045EXPORT_SYMBOL_GPL(nfs_initiate_write);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001046
1047/*
1048 * Set up the argument/result storage required for the RPC call.
1049 */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001050static void nfs_write_rpcsetup(struct nfs_write_data *data,
Andy Adamsond138d5d2011-03-03 15:13:41 +00001051 unsigned int count, unsigned int offset,
Fred Isamanea2cf222012-04-20 14:47:53 -04001052 int how, struct nfs_commit_info *cinfo)
Andy Adamsond138d5d2011-03-03 15:13:41 +00001053{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001054 struct nfs_page *req = data->header->req;
Andy Adamsond138d5d2011-03-03 15:13:41 +00001055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 /* Set up the RPC argument and reply structs
1057 * NB: take care not to mess about with data->commit et al. */
1058
Fred Isaman6c75dc02012-04-20 14:47:47 -04001059 data->args.fh = NFS_FH(data->header->inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 data->args.offset = req_offset(req) + offset;
Boaz Harrosh2bea0382011-06-16 11:35:46 -04001061 /* pnfs_set_layoutcommit needs this */
1062 data->mds_offset = data->args.offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 data->args.pgbase = req->wb_pgbase + offset;
Fred Isaman30dd3742012-04-20 14:47:45 -04001064 data->args.pages = data->pages.pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 data->args.count = count;
Trond Myklebust383ba712008-02-19 20:04:20 -05001066 data->args.context = get_nfs_open_context(req->wb_context);
Trond Myklebustf11ac8d2010-06-25 16:35:53 -04001067 data->args.lock_context = req->wb_lock_context;
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001068 data->args.stable = NFS_UNSTABLE;
Trond Myklebust87ed5eb2011-07-12 13:42:01 -04001069 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
1070 case 0:
1071 break;
1072 case FLUSH_COND_STABLE:
Fred Isamanea2cf222012-04-20 14:47:53 -04001073 if (nfs_reqs_to_commit(cinfo))
Trond Myklebust87ed5eb2011-07-12 13:42:01 -04001074 break;
1075 default:
1076 data->args.stable = NFS_FILE_SYNC;
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 data->res.fattr = &data->fattr;
1080 data->res.count = count;
1081 data->res.verf = &data->verf;
Trond Myklebust0e574af2005-10-27 22:12:38 -04001082 nfs_fattr_init(&data->fattr);
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001083}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001085static int nfs_do_write(struct nfs_write_data *data,
1086 const struct rpc_call_ops *call_ops,
Trond Myklebust6e4efd52011-07-12 13:42:02 -04001087 int how)
1088{
Fred Isamancd841602012-04-20 14:47:44 -04001089 struct inode *inode = data->header->inode;
Andy Adamson0382b742011-03-03 15:13:45 +00001090
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001091 return nfs_initiate_write(NFS_CLIENT(inode), data, call_ops, how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092}
1093
Trond Myklebust275acaa2011-07-12 13:42:02 -04001094static int nfs_do_multiple_writes(struct list_head *head,
1095 const struct rpc_call_ops *call_ops,
Trond Myklebust275acaa2011-07-12 13:42:02 -04001096 int how)
1097{
1098 struct nfs_write_data *data;
1099 int ret = 0;
1100
1101 while (!list_empty(head)) {
1102 int ret2;
1103
Fred Isaman6c75dc02012-04-20 14:47:47 -04001104 data = list_first_entry(head, struct nfs_write_data, list);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001105 list_del_init(&data->list);
1106
Trond Myklebustdce81292011-07-13 15:59:19 -04001107 ret2 = nfs_do_write(data, call_ops, how);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001108 if (ret == 0)
1109 ret = ret2;
1110 }
1111 return ret;
1112}
1113
Fred6d884e82008-03-19 11:24:38 -04001114/* If a nfs_flush_* function fails, it should remove reqs from @head and
1115 * call this on each, which will prepare them to be retried on next
1116 * writeback using standard nfs.
1117 */
1118static void nfs_redirty_request(struct nfs_page *req)
1119{
1120 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001121 nfs_unlock_request(req);
Trond Myklebustd1182b32012-05-09 13:37:43 -04001122 nfs_end_page_writeback(req->wb_page);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001123 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001124}
1125
Fred Isaman061ae2e2012-04-20 14:47:48 -04001126static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001127{
1128 struct nfs_page *req;
1129
1130 while (!list_empty(head)) {
1131 req = nfs_list_entry(head->next);
1132 nfs_list_remove_request(req);
1133 nfs_redirty_request(req);
1134 }
1135}
1136
Fred Isaman061ae2e2012-04-20 14:47:48 -04001137static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1138 .error_cleanup = nfs_async_write_error,
1139 .completion = nfs_write_completion,
1140};
1141
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001142static void nfs_flush_error(struct nfs_pageio_descriptor *desc,
1143 struct nfs_pgio_header *hdr)
1144{
1145 set_bit(NFS_IOHDR_REDO, &hdr->flags);
1146 while (!list_empty(&hdr->rpc_list)) {
1147 struct nfs_write_data *data = list_first_entry(&hdr->rpc_list,
1148 struct nfs_write_data, list);
1149 list_del(&data->list);
1150 nfs_writedata_release(data);
1151 }
1152 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
1153}
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155/*
1156 * Generate multiple small requests to write out a single
1157 * contiguous dirty area on one page.
1158 */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001159static int nfs_flush_multi(struct nfs_pageio_descriptor *desc,
1160 struct nfs_pgio_header *hdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001162 struct nfs_page *req = hdr->req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 struct page *page = req->wb_page;
1164 struct nfs_write_data *data;
Trond Myklebustd0979712011-07-12 13:42:02 -04001165 size_t wsize = desc->pg_bsize, nbytes;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001166 unsigned int offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 int requests = 0;
Fred Isamanea2cf222012-04-20 14:47:53 -04001168 struct nfs_commit_info cinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169
Fred Isamanea2cf222012-04-20 14:47:53 -04001170 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Trond Myklebustb31268a2011-03-21 17:02:00 -04001172 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
Fred Isamanea2cf222012-04-20 14:47:53 -04001173 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo) ||
Trond Myklebustb31268a2011-03-21 17:02:00 -04001174 desc->pg_count > wsize))
1175 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1176
1177
Trond Myklebust275acaa2011-07-12 13:42:02 -04001178 offset = 0;
Fred Isamanc76069b2011-03-03 15:13:48 +00001179 nbytes = desc->pg_count;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001180 do {
1181 size_t len = min(nbytes, wsize);
1182
Fred Isaman6c75dc02012-04-20 14:47:47 -04001183 data = nfs_writedata_alloc(hdr, 1);
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001184 if (!data) {
1185 nfs_flush_error(desc, hdr);
1186 return -ENOMEM;
1187 }
Fred Isaman30dd3742012-04-20 14:47:45 -04001188 data->pages.pagevec[0] = page;
Fred Isamanea2cf222012-04-20 14:47:53 -04001189 nfs_write_rpcsetup(data, len, offset, desc->pg_ioflags, &cinfo);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001190 list_add(&data->list, &hdr->rpc_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 requests++;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001192 nbytes -= len;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001193 offset += len;
Trond Myklebuste9f7bee2006-09-08 09:48:54 -07001194 } while (nbytes != 0);
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001195 nfs_list_remove_request(req);
1196 nfs_list_add_request(req, &hdr->pages);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001197 desc->pg_rpc_callops = &nfs_write_common_ops;
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001198 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199}
1200
1201/*
1202 * Create an RPC task for the given write request and kick it.
1203 * The page must have been locked by the caller.
1204 *
1205 * It may happen that the page we're passed is not marked dirty.
1206 * This is the case if nfs_updatepage detects a conflicting request
1207 * that has been written but not committed.
1208 */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001209static int nfs_flush_one(struct nfs_pageio_descriptor *desc,
1210 struct nfs_pgio_header *hdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211{
1212 struct nfs_page *req;
1213 struct page **pages;
1214 struct nfs_write_data *data;
Fred Isamanc76069b2011-03-03 15:13:48 +00001215 struct list_head *head = &desc->pg_list;
Fred Isamanea2cf222012-04-20 14:47:53 -04001216 struct nfs_commit_info cinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
Fred Isaman6c75dc02012-04-20 14:47:47 -04001218 data = nfs_writedata_alloc(hdr, nfs_page_array_len(desc->pg_base,
1219 desc->pg_count));
1220 if (!data) {
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001221 nfs_flush_error(desc, hdr);
1222 return -ENOMEM;
Fred Isaman44b83792011-03-03 15:13:44 +00001223 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001224
Fred Isamanea2cf222012-04-20 14:47:53 -04001225 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
Fred Isaman30dd3742012-04-20 14:47:45 -04001226 pages = data->pages.pagevec;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 while (!list_empty(head)) {
1228 req = nfs_list_entry(head->next);
1229 nfs_list_remove_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001230 nfs_list_add_request(req, &hdr->pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 *pages++ = req->wb_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
Trond Myklebustb31268a2011-03-21 17:02:00 -04001234 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
Fred Isamanea2cf222012-04-20 14:47:53 -04001235 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
Trond Myklebustb31268a2011-03-21 17:02:00 -04001236 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
1237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 /* Set up the argument struct */
Fred Isamanea2cf222012-04-20 14:47:53 -04001239 nfs_write_rpcsetup(data, desc->pg_count, 0, desc->pg_ioflags, &cinfo);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001240 list_add(&data->list, &hdr->rpc_list);
1241 desc->pg_rpc_callops = &nfs_write_common_ops;
Trond Myklebust25b11dc2012-05-01 12:07:22 -04001242 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243}
1244
Fred Isaman6c75dc02012-04-20 14:47:47 -04001245int nfs_generic_flush(struct nfs_pageio_descriptor *desc,
1246 struct nfs_pgio_header *hdr)
Trond Myklebustdce81292011-07-13 15:59:19 -04001247{
1248 if (desc->pg_bsize < PAGE_CACHE_SIZE)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001249 return nfs_flush_multi(desc, hdr);
1250 return nfs_flush_one(desc, hdr);
Trond Myklebustdce81292011-07-13 15:59:19 -04001251}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001252EXPORT_SYMBOL_GPL(nfs_generic_flush);
Trond Myklebustdce81292011-07-13 15:59:19 -04001253
1254static int nfs_generic_pg_writepages(struct nfs_pageio_descriptor *desc)
Trond Myklebust1751c362011-06-10 13:30:23 -04001255{
Fred Isaman6c75dc02012-04-20 14:47:47 -04001256 struct nfs_write_header *whdr;
1257 struct nfs_pgio_header *hdr;
Trond Myklebust275acaa2011-07-12 13:42:02 -04001258 int ret;
1259
Fred Isaman6c75dc02012-04-20 14:47:47 -04001260 whdr = nfs_writehdr_alloc();
1261 if (!whdr) {
Trond Myklebust9b5415b2012-04-27 14:31:47 -04001262 desc->pg_completion_ops->error_cleanup(&desc->pg_list);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001263 return -ENOMEM;
1264 }
1265 hdr = &whdr->header;
1266 nfs_pgheader_init(desc, hdr, nfs_writehdr_free);
1267 atomic_inc(&hdr->refcnt);
1268 ret = nfs_generic_flush(desc, hdr);
Trond Myklebust50828d72011-07-12 13:42:02 -04001269 if (ret == 0)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001270 ret = nfs_do_multiple_writes(&hdr->rpc_list,
1271 desc->pg_rpc_callops,
1272 desc->pg_ioflags);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001273 if (atomic_dec_and_test(&hdr->refcnt))
Fred Isaman061ae2e2012-04-20 14:47:48 -04001274 hdr->completion_ops->completion(hdr);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001275 return ret;
Trond Myklebust1751c362011-06-10 13:30:23 -04001276}
Trond Myklebust1751c362011-06-10 13:30:23 -04001277
1278static const struct nfs_pageio_ops nfs_pageio_write_ops = {
1279 .pg_test = nfs_generic_pg_test,
1280 .pg_doio = nfs_generic_pg_writepages,
1281};
1282
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001283void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001284 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001285 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001286{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001287 struct nfs_server *server = NFS_SERVER(inode);
1288 const struct nfs_pageio_ops *pg_ops = &nfs_pageio_write_ops;
1289
1290#ifdef CONFIG_NFS_V4_1
1291 if (server->pnfs_curr_ld && !force_mds)
1292 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1293#endif
1294 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001295}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001296EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001297
Trond Myklebustdce81292011-07-13 15:59:19 -04001298void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1299{
1300 pgio->pg_ops = &nfs_pageio_write_ops;
1301 pgio->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1302}
Trond Myklebust1f945352011-07-13 15:59:57 -04001303EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Andy Adamsondef6ed72009-04-01 09:22:26 -04001306void nfs_write_prepare(struct rpc_task *task, void *calldata)
1307{
1308 struct nfs_write_data *data = calldata;
NeilBrownef1820f2013-09-04 17:04:49 +10001309 int err;
1310 err = NFS_PROTO(data->header->inode)->write_rpc_prepare(task, data);
1311 if (err)
1312 rpc_exit(task, err);
Andy Adamsondef6ed72009-04-01 09:22:26 -04001313}
Andy Adamsondef6ed72009-04-01 09:22:26 -04001314
Fred Isaman0b7c0152012-04-20 14:47:39 -04001315void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1316{
1317 struct nfs_commit_data *data = calldata;
1318
1319 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1320}
1321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322/*
1323 * Handle a write reply that flushes a whole page.
1324 *
1325 * FIXME: There is an inherent race with invalidate_inode_pages and
1326 * writebacks since the page->count is kept > 1 for as long
1327 * as the page has a write request pending.
1328 */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001329static void nfs_writeback_done_common(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001331 struct nfs_write_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001333 nfs_writeback_done(task, data);
1334}
1335
Fred Isaman6c75dc02012-04-20 14:47:47 -04001336static void nfs_writeback_release_common(void *calldata)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001337{
1338 struct nfs_write_data *data = calldata;
Fred Isamancd841602012-04-20 14:47:44 -04001339 struct nfs_pgio_header *hdr = data->header;
Trond Myklebuste2fecb22012-01-06 08:57:46 -05001340 int status = data->task.tk_status;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001341
Fred Isaman6c75dc02012-04-20 14:47:47 -04001342 if ((status >= 0) && nfs_write_need_commit(data)) {
1343 spin_lock(&hdr->lock);
1344 if (test_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags))
1345 ; /* Do nothing */
1346 else if (!test_and_set_bit(NFS_IOHDR_NEED_COMMIT, &hdr->flags))
Trond Myklebust9bce0082012-06-05 18:32:03 -04001347 memcpy(hdr->verf, &data->verf, sizeof(*hdr->verf));
1348 else if (memcmp(hdr->verf, &data->verf, sizeof(*hdr->verf)))
Fred Isaman6c75dc02012-04-20 14:47:47 -04001349 set_bit(NFS_IOHDR_NEED_RESCHED, &hdr->flags);
1350 spin_unlock(&hdr->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
Fred Isamancd841602012-04-20 14:47:44 -04001352 nfs_writedata_release(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
Fred Isaman6c75dc02012-04-20 14:47:47 -04001355static const struct rpc_call_ops nfs_write_common_ops = {
Andy Adamsondef6ed72009-04-01 09:22:26 -04001356 .rpc_call_prepare = nfs_write_prepare,
Fred Isaman6c75dc02012-04-20 14:47:47 -04001357 .rpc_call_done = nfs_writeback_done_common,
1358 .rpc_release = nfs_writeback_release_common,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001359};
1360
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001361/*
1362 * Special version of should_remove_suid() that ignores capabilities.
1363 */
1364static int nfs_should_remove_suid(const struct inode *inode)
1365{
1366 umode_t mode = inode->i_mode;
1367 int kill = 0;
1368
1369 /* suid always must be killed */
1370 if (unlikely(mode & S_ISUID))
1371 kill = ATTR_KILL_SUID;
1372
1373 /*
1374 * sgid without any exec bits is just a mandatory locking mark; leave
1375 * it alone. If some exec bits are set, it's a real sgid; kill it.
1376 */
1377 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1378 kill |= ATTR_KILL_SGID;
1379
1380 if (unlikely(kill && S_ISREG(mode)))
1381 return kill;
1382
1383 return 0;
1384}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386/*
1387 * This function is called when the WRITE call is complete.
1388 */
Fred Isaman13602892011-02-11 15:42:38 +00001389void nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390{
Anna Schumaker3c6b8992014-05-06 09:12:24 -04001391 struct nfs_pgio_args *argp = &data->args;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 struct nfs_writeres *resp = &data->res;
Fred Isamancd841602012-04-20 14:47:44 -04001393 struct inode *inode = data->header->inode;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001394 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Chuck Levera3f565b2007-01-31 12:14:01 -05001396 dprintk("NFS: %5u nfs_writeback_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 task->tk_pid, task->tk_status);
1398
Chuck Leverf551e442006-09-20 14:33:04 -04001399 /*
1400 * ->write_done will attempt to use post-op attributes to detect
1401 * conflicting writes by other clients. A strict interpretation
1402 * of close-to-open would allow us to continue caching even if
1403 * another writer had changed the file, but some applications
1404 * depend on tighter cache coherency when writing.
1405 */
Fred Isamancd841602012-04-20 14:47:44 -04001406 status = NFS_PROTO(inode)->write_done(task, data);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001407 if (status != 0)
Fred Isaman13602892011-02-11 15:42:38 +00001408 return;
Fred Isamancd841602012-04-20 14:47:44 -04001409 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, resp->count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001410
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001411#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (resp->verf->committed < argp->stable && task->tk_status >= 0) {
1413 /* We tried a write call, but the server did not
1414 * commit data to stable storage even though we
1415 * requested it.
1416 * Note: There is a known bug in Tru64 < 5.0 in which
1417 * the server reports NFS_DATA_SYNC, but performs
1418 * NFS_FILE_SYNC. We therefore implement this checking
1419 * as a dprintk() in order to avoid filling syslog.
1420 */
1421 static unsigned long complain;
1422
Fred Isamana69aef12011-03-03 15:13:47 +00001423 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001425 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001427 NFS_SERVER(inode)->nfs_client->cl_hostname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 resp->verf->committed, argp->stable);
1429 complain = jiffies + 300 * HZ;
1430 }
1431 }
1432#endif
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001433 if (task->tk_status < 0) {
Fred Isaman6c75dc02012-04-20 14:47:47 -04001434 nfs_set_pgio_error(data->header, task->tk_status, argp->offset);
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001435 return;
1436 }
1437
1438 /* Deal with the suid/sgid bit corner case */
1439 if (nfs_should_remove_suid(inode))
1440 nfs_mark_for_revalidate(inode);
1441
1442 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 static unsigned long complain;
1444
Fred Isaman6c75dc02012-04-20 14:47:47 -04001445 /* This a short write! */
Fred Isamancd841602012-04-20 14:47:44 -04001446 nfs_inc_stats(inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001449 if (resp->count == 0) {
1450 if (time_before(complain, jiffies)) {
1451 printk(KERN_WARNING
1452 "NFS: Server wrote zero bytes, expected %u.\n",
1453 argp->count);
1454 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001456 nfs_set_pgio_error(data->header, -EIO, argp->offset);
1457 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001458 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001460 /* Was this an NFSv2 write or an NFSv3 stable write? */
1461 if (resp->verf->committed != NFS_UNSTABLE) {
1462 /* Resend from where the server left off */
1463 data->mds_offset += resp->count;
1464 argp->offset += resp->count;
1465 argp->pgbase += resp->count;
1466 argp->count -= resp->count;
1467 } else {
1468 /* Resend as a stable write in order to avoid
1469 * headaches in the case of a server crash.
1470 */
1471 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001473 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475}
1476
1477
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001478#if IS_ENABLED(CONFIG_NFS_V3) || IS_ENABLED(CONFIG_NFS_V4)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001479static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1480{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001481 int ret;
1482
Trond Myklebust71d0a612010-04-22 15:35:57 -04001483 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1484 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001485 if (!may_wait)
1486 return 0;
1487 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1488 NFS_INO_COMMIT,
1489 nfs_wait_bit_killable,
1490 TASK_KILLABLE);
1491 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001492}
1493
Fred Isamanf453a542012-04-20 14:47:54 -04001494static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001495{
1496 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1497 smp_mb__after_clear_bit();
1498 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1499}
1500
Fred Isaman0b7c0152012-04-20 14:47:39 -04001501void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001503 put_nfs_open_context(data->context);
1504 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505}
Fred Isamane0c2b382011-03-23 13:27:53 +00001506EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Fred Isaman0b7c0152012-04-20 14:47:39 -04001508int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001509 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001510 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511{
Trond Myklebust07737692007-10-25 18:42:54 -04001512 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001513 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001514 struct rpc_message msg = {
1515 .rpc_argp = &data->args,
1516 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001517 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001518 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001519 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001520 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001521 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001522 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001523 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001524 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001525 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001526 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001527 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001528 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001529 /* Set up the initial task struct. */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001530 NFS_PROTO(data->inode)->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Chuck Levera3f565b2007-01-31 12:14:01 -05001532 dprintk("NFS: %5u initiated commit call\n", data->task.tk_pid);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001533
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001534 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1535 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1536
Trond Myklebust07737692007-10-25 18:42:54 -04001537 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001538 if (IS_ERR(task))
1539 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001540 if (how & FLUSH_SYNC)
1541 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001542 rpc_put_task(task);
1543 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544}
Fred Isamane0c2b382011-03-23 13:27:53 +00001545EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
1547/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001548 * Set up the argument/result storage required for the RPC call.
1549 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001550void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001551 struct list_head *head,
1552 struct pnfs_layout_segment *lseg,
1553 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001554{
1555 struct nfs_page *first = nfs_list_entry(head->next);
Al Viro3d4ff432011-06-22 18:40:12 -04001556 struct inode *inode = first->wb_context->dentry->d_inode;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001557
1558 /* Set up the RPC argument and reply structs
1559 * NB: take care not to mess about with data->commit et al. */
1560
1561 list_splice_init(head, &data->pages);
1562
1563 data->inode = inode;
1564 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001565 data->lseg = lseg; /* reference transferred */
Fred Isaman9ace33c2011-03-23 13:27:45 +00001566 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001567 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001568 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001569
1570 data->args.fh = NFS_FH(data->inode);
1571 /* Note: we always request a commit of the entire inode */
1572 data->args.offset = 0;
1573 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001574 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001575 data->res.fattr = &data->fattr;
1576 data->res.verf = &data->verf;
1577 nfs_fattr_init(&data->fattr);
1578}
Fred Isamane0c2b382011-03-23 13:27:53 +00001579EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001580
Fred Isamane0c2b382011-03-23 13:27:53 +00001581void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001582 struct pnfs_layout_segment *lseg,
1583 struct nfs_commit_info *cinfo)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001584{
1585 struct nfs_page *req;
1586
1587 while (!list_empty(page_list)) {
1588 req = nfs_list_entry(page_list->next);
1589 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -04001590 nfs_mark_request_commit(req, lseg, cinfo);
Fred Isaman56f9cd62012-04-20 14:47:56 -04001591 if (!cinfo->dreq) {
1592 dec_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001593 dec_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
Fred Isaman56f9cd62012-04-20 14:47:56 -04001594 BDI_RECLAIMABLE);
1595 }
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001596 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001597 }
1598}
Fred Isamane0c2b382011-03-23 13:27:53 +00001599EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001600
Fred Isaman9ace33c2011-03-23 13:27:45 +00001601/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 * Commit dirty pages
1603 */
1604static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001605nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1606 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001608 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001610 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612 if (!data)
1613 goto out_bad;
1614
1615 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001616 nfs_init_commit(data, head, NULL, cinfo);
1617 atomic_inc(&cinfo->mds->rpcs_out);
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001618 return nfs_initiate_commit(NFS_CLIENT(inode), data, data->mds_ops,
1619 how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 out_bad:
Fred Isamanea2cf222012-04-20 14:47:53 -04001621 nfs_retry_commit(head, NULL, cinfo);
Fred Isamanf453a542012-04-20 14:47:54 -04001622 cinfo->completion_ops->error_cleanup(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 return -ENOMEM;
1624}
1625
1626/*
1627 * COMMIT call returned
1628 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001629static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001631 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Chuck Levera3f565b2007-01-31 12:14:01 -05001633 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 task->tk_pid, task->tk_status);
1635
Trond Myklebust788e7a82006-03-20 13:44:27 -05001636 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001637 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001638}
1639
Fred Isamanf453a542012-04-20 14:47:54 -04001640static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001641{
Fred Isaman5917ce82011-03-23 13:27:48 +00001642 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001643 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001644 struct nfs_commit_info cinfo;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 while (!list_empty(&data->pages)) {
1647 req = nfs_list_entry(data->pages.next);
1648 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001649 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Niels de Vos1e8968c2013-12-17 18:20:16 +01001651 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001652 req->wb_context->dentry->d_sb->s_id,
Niels de Vos1e8968c2013-12-17 18:20:16 +01001653 (unsigned long long)NFS_FILEID(req->wb_context->dentry->d_inode),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 req->wb_bytes,
1655 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001656 if (status < 0) {
1657 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001659 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 goto next;
1661 }
1662
1663 /* Okay, COMMIT succeeded, apparently. Check the verifier
1664 * returned by the server against all stored verfs. */
Trond Myklebust2f2c63b2012-06-08 11:56:09 -04001665 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 /* We have a match */
1667 nfs_inode_remove_request(req);
1668 dprintk(" OK\n");
1669 goto next;
1670 }
1671 /* We have a mismatch. Write the page again */
1672 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001673 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001674 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001676 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
Fred Isamanf453a542012-04-20 14:47:54 -04001678 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1679 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1680 nfs_commit_clear_lock(NFS_I(data->inode));
Fred Isaman5917ce82011-03-23 13:27:48 +00001681}
1682
1683static void nfs_commit_release(void *calldata)
1684{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001685 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001686
Fred Isamanf453a542012-04-20 14:47:54 -04001687 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001688 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001690
1691static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001692 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001693 .rpc_call_done = nfs_commit_done,
1694 .rpc_release = nfs_commit_release,
1695};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
Fred Isamanf453a542012-04-20 14:47:54 -04001697static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1698 .completion = nfs_commit_release_pages,
1699 .error_cleanup = nfs_commit_clear_lock,
1700};
1701
Fred Isaman1763da12012-04-20 14:47:57 -04001702int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1703 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001704{
1705 int status;
1706
Fred Isamanea2cf222012-04-20 14:47:53 -04001707 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001708 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001709 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001710 return status;
1711}
1712
Trond Myklebustb608b282010-07-30 15:31:54 -04001713int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001716 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001717 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001718 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Trond Myklebustb8413f92011-03-21 15:37:01 -04001720 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1721 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001722 goto out_mark_dirty;
Fred Isamanea2cf222012-04-20 14:47:53 -04001723 nfs_init_cinfo_from_inode(&cinfo, inode);
1724 res = nfs_scan_commit(inode, &head, &cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001726 int error;
1727
Fred Isamanea2cf222012-04-20 14:47:53 -04001728 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001729 if (error < 0)
1730 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001731 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001732 goto out_mark_dirty;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001733 error = wait_on_bit(&NFS_I(inode)->flags,
1734 NFS_INO_COMMIT,
1735 nfs_wait_bit_killable,
1736 TASK_KILLABLE);
1737 if (error < 0)
1738 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001739 } else
1740 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001741 return res;
1742 /* Note: If we exit without ensuring that the commit is complete,
1743 * we must mark the inode as dirty. Otherwise, future calls to
1744 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1745 * that the data is on the disk.
1746 */
1747out_mark_dirty:
1748 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 return res;
1750}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001751
1752static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1753{
Trond Myklebust420e3642010-02-19 17:00:02 -08001754 struct nfs_inode *nfsi = NFS_I(inode);
1755 int flags = FLUSH_SYNC;
1756 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001757
Jeff Layton3236c3e2011-10-11 09:49:21 -04001758 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001759 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e2011-10-11 09:49:21 -04001760 return ret;
1761
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001762 if (wbc->sync_mode == WB_SYNC_NONE) {
1763 /* Don't commit yet if this is a non-blocking flush and there
1764 * are a lot of outstanding writes for this mapping.
1765 */
Fred Isamanea2cf222012-04-20 14:47:53 -04001766 if (nfsi->commit_info.ncommit <= (nfsi->npages >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001767 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001768
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001769 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001770 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001771 }
1772
Trond Myklebust420e3642010-02-19 17:00:02 -08001773 ret = nfs_commit_inode(inode, flags);
1774 if (ret >= 0) {
1775 if (wbc->sync_mode == WB_SYNC_NONE) {
1776 if (ret < wbc->nr_to_write)
1777 wbc->nr_to_write -= ret;
1778 else
1779 wbc->nr_to_write = 0;
1780 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001781 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001782 }
1783out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001784 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1785 return ret;
1786}
Trond Myklebustc63c7b02007-04-02 19:29:52 -04001787#else
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001788static int nfs_commit_unstable_pages(struct inode *inode, struct writeback_control *wbc)
1789{
1790 return 0;
1791}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792#endif
1793
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001794int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1795{
Bryan Schumakera8d8f022012-06-20 15:53:49 -04001796 return nfs_commit_unstable_pages(inode, wbc);
1797}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001798EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001799
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001800/*
1801 * flush the inode to disk.
1802 */
1803int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001804{
1805 struct writeback_control wbc = {
Trond Myklebust72cb77f2009-03-11 14:10:30 -04001806 .sync_mode = WB_SYNC_ALL,
Trond Myklebust34901f72007-07-25 14:09:54 -04001807 .nr_to_write = LONG_MAX,
Trond Myklebustd7fb1202008-10-06 20:08:56 -04001808 .range_start = 0,
1809 .range_end = LLONG_MAX,
Trond Myklebust34901f72007-07-25 14:09:54 -04001810 };
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001811 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001812
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001813 trace_nfs_writeback_inode_enter(inode);
1814
1815 ret = sync_inode(inode, &wbc);
1816
1817 trace_nfs_writeback_inode_exit(inode, ret);
1818 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001819}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001820EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001821
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001822int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1823{
1824 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001825 int ret = 0;
1826
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001827 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001828 wait_on_page_writeback(page);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001829 req = nfs_page_find_request(page);
1830 if (req == NULL)
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001831 break;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001832 if (nfs_lock_request(req)) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001833 nfs_clear_request_commit(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001834 nfs_inode_remove_request(req);
1835 /*
1836 * In case nfs_inode_remove_request has marked the
1837 * page as being dirty
1838 */
1839 cancel_dirty_page(page, PAGE_CACHE_SIZE);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001840 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001841 break;
1842 }
1843 ret = nfs_wait_on_request(req);
Trond Myklebustc9edda72010-01-26 15:41:34 -05001844 nfs_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001845 if (ret < 0)
Trond Myklebustc9889502010-02-19 17:03:21 -08001846 break;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001847 }
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001848 return ret;
1849}
1850
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001851/*
1852 * Write back all requests on one page - we do this before reading it.
1853 */
1854int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001855{
Mel Gorman29418aa2012-07-31 16:45:10 -07001856 loff_t range_start = page_file_offset(page);
Trond Myklebust1c759502006-10-09 16:18:38 -04001857 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001858 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001859 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001860 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001861 .range_start = range_start,
1862 .range_end = range_end,
1863 };
1864 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001865
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001866 trace_nfs_writeback_page_enter(inode);
1867
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001868 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001869 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001870 if (clear_page_dirty_for_io(page)) {
1871 ret = nfs_writepage_locked(page, &wbc);
1872 if (ret < 0)
1873 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001874 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001875 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001876 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001877 if (!PagePrivate(page))
1878 break;
1879 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001880 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001881 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001882 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001883out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001884 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001885 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001886}
1887
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001888#ifdef CONFIG_MIGRATION
1889int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001890 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001891{
Jeff Layton2da95652011-10-12 10:57:42 -04001892 /*
1893 * If PagePrivate is set, then the page is currently associated with
1894 * an in-progress read or write request. Don't try to migrate it.
1895 *
1896 * FIXME: we could do this in principle, but we'll need a way to ensure
1897 * that we can safely release the inode reference while holding
1898 * the page lock.
1899 */
1900 if (PagePrivate(page))
1901 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001902
David Howells8c209ce2012-12-05 13:34:49 +00001903 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1904 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001905
Mel Gormana6bc32b2012-01-12 17:19:43 -08001906 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001907}
1908#endif
1909
David Howellsf7b422b2006-06-09 09:34:33 -04001910int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911{
1912 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Fred Isamancd841602012-04-20 14:47:44 -04001913 sizeof(struct nfs_write_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001915 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 if (nfs_wdata_cachep == NULL)
1917 return -ENOMEM;
1918
Matthew Dobson93d23412006-03-26 01:37:50 -08001919 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1920 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001922 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Fred Isaman0b7c0152012-04-20 14:47:39 -04001924 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1925 sizeof(struct nfs_commit_data),
1926 0, SLAB_HWCACHE_ALIGN,
1927 NULL);
1928 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001929 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001930
Matthew Dobson93d23412006-03-26 01:37:50 -08001931 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08001932 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001934 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
Peter Zijlstra89a09142007-03-16 13:38:26 -08001936 /*
1937 * NFS congestion size, scale with available memory.
1938 *
1939 * 64MB: 8192k
1940 * 128MB: 11585k
1941 * 256MB: 16384k
1942 * 512MB: 23170k
1943 * 1GB: 32768k
1944 * 2GB: 46340k
1945 * 4GB: 65536k
1946 * 8GB: 92681k
1947 * 16GB: 131072k
1948 *
1949 * This allows larger machines to have larger/more transfers.
1950 * Limit the default to 256M
1951 */
1952 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
1953 if (nfs_congestion_kb > 256*1024)
1954 nfs_congestion_kb = 256*1024;
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04001957
1958out_destroy_commit_cache:
1959 kmem_cache_destroy(nfs_cdata_cachep);
1960out_destroy_write_mempool:
1961 mempool_destroy(nfs_wdata_mempool);
1962out_destroy_write_cache:
1963 kmem_cache_destroy(nfs_wdata_cachep);
1964 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965}
1966
David Brownell266bee82006-06-27 12:59:15 -07001967void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968{
1969 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04001970 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07001972 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973}
1974