blob: 5f4fd53e5764884391569d010661fe1639b35053 [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>
Trond Myklebustaf7cf052015-09-29 20:34:05 -040024#include <linux/freezer.h>
25#include <linux/wait.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070026
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050030#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050031#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040032#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040033#include "fscache.h"
Fred Isaman94ad1c82011-03-01 01:34:14 +000034#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Trond Myklebustf4ce1292013-08-19 18:59:33 -040036#include "nfstrace.h"
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#define NFSDBG_FACILITY NFSDBG_PAGECACHE
39
40#define MIN_POOL_WRITE (32)
41#define MIN_POOL_COMMIT (4)
42
43/*
44 * Local function declarations
45 */
Fred Isamanf8512ad2008-03-19 11:24:39 -040046static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050047static const struct rpc_call_ops nfs_commit_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -040048static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
Fred Isamanf453a542012-04-20 14:47:54 -040049static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
Anna Schumaker4a0de552014-05-06 09:12:30 -040050static const struct nfs_rw_ops nfs_rw_write_ops;
Weston Andros Adamsond4581382014-07-11 10:20:48 -040051static void nfs_clear_request_commit(struct nfs_page *req);
Weston Andros Adamson02d14262014-07-17 20:42:17 -040052static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
53 struct inode *inode);
Trond Myklebust3a3908c2014-09-08 22:21:00 -070054static struct nfs_page *
55nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
56 struct page *page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
Christoph Lametere18b8902006-12-06 20:33:20 -080058static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050059static mempool_t *nfs_wdata_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -040060static struct kmem_cache *nfs_cdata_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061static mempool_t *nfs_commit_mempool;
62
Fred Isaman0b7c0152012-04-20 14:47:39 -040063struct nfs_commit_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064{
Mel Gorman192e5012012-07-31 16:45:16 -070065 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
Chuck Lever40859d72005-11-30 18:09:02 -050066
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 if (p) {
68 memset(p, 0, sizeof(*p));
69 INIT_LIST_HEAD(&p->pages);
70 }
71 return p;
72}
Fred Isamane0c2b382011-03-23 13:27:53 +000073EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Fred Isaman0b7c0152012-04-20 14:47:39 -040075void nfs_commit_free(struct nfs_commit_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
77 mempool_free(p, nfs_commit_mempool);
78}
Fred Isamane0c2b382011-03-23 13:27:53 +000079EXPORT_SYMBOL_GPL(nfs_commit_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040081static struct nfs_pgio_header *nfs_writehdr_alloc(void)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050082{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040083 struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050084
Anna Schumaker4a0de552014-05-06 09:12:30 -040085 if (p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050086 memset(p, 0, sizeof(*p));
Trond Myklebust3feb2d42006-03-20 13:44:37 -050087 return p;
88}
89
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040090static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -040091{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040092 mempool_free(hdr, nfs_wdata_mempool);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050093}
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Trond Myklebust7b159fc2007-07-25 14:09:54 -040095static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
96{
97 ctx->error = error;
98 smp_wmb();
99 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
100}
101
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400102/*
103 * nfs_page_find_head_request_locked - find head request associated with @page
104 *
105 * must be called while holding the inode lock.
106 *
107 * returns matching head request with reference held, or NULL if not found.
108 */
Mel Gorman29418aa2012-07-31 16:45:10 -0700109static struct nfs_page *
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400110nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -0500111{
112 struct nfs_page *req = NULL;
113
Mel Gorman29418aa2012-07-31 16:45:10 -0700114 if (PagePrivate(page))
Trond Myklebust277459d2006-12-05 00:35:35 -0500115 req = (struct nfs_page *)page_private(page);
Weston Andros Adamson02d14262014-07-17 20:42:17 -0400116 else if (unlikely(PageSwapCache(page)))
117 req = nfs_page_search_commits_for_head_request_locked(nfsi,
118 page);
Mel Gorman29418aa2012-07-31 16:45:10 -0700119
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400120 if (req) {
121 WARN_ON_ONCE(req->wb_head != req);
Mel Gorman29418aa2012-07-31 16:45:10 -0700122 kref_get(&req->wb_kref);
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400123 }
Mel Gorman29418aa2012-07-31 16:45:10 -0700124
Trond Myklebust277459d2006-12-05 00:35:35 -0500125 return req;
126}
127
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400128/*
129 * nfs_page_find_head_request - find head request associated with @page
130 *
131 * returns matching head request with reference held, or NULL if not found.
132 */
133static struct nfs_page *nfs_page_find_head_request(struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -0500134{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700135 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500136 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500137
Trond Myklebust587142f2007-07-02 09:57:54 -0400138 spin_lock(&inode->i_lock);
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400139 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400140 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500141 return req;
142}
143
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144/* Adjust the file length if we're writing beyond the end */
145static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
146{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700147 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400148 loff_t end, i_size;
149 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Trond Myklebusta3d01452008-06-11 12:21:19 -0400151 spin_lock(&inode->i_lock);
152 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300153 end_index = (i_size - 1) >> PAGE_SHIFT;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700154 if (i_size > 0 && page_file_index(page) < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400155 goto out;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700156 end = page_file_offset(page) + ((loff_t)offset+count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400158 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400160 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
161out:
162 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
Trond Myklebusta301b772007-02-06 11:07:15 -0800165/* A writeback failed: mark the page as bad, and invalidate the page cache */
166static void nfs_set_pageerror(struct page *page)
167{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700168 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
Trond Myklebusta301b772007-02-06 11:07:15 -0800169}
170
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400171/*
172 * nfs_page_group_search_locked
173 * @head - head request of page group
174 * @page_offset - offset into page
175 *
176 * Search page group with head @head to find a request that contains the
177 * page offset @page_offset.
178 *
179 * Returns a pointer to the first matching nfs request, or NULL if no
180 * match is found.
181 *
182 * Must be called with the page group lock held
183 */
184static struct nfs_page *
185nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
186{
187 struct nfs_page *req;
188
189 WARN_ON_ONCE(head != head->wb_head);
190 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
191
192 req = head;
193 do {
194 if (page_offset >= req->wb_pgbase &&
195 page_offset < (req->wb_pgbase + req->wb_bytes))
196 return req;
197
198 req = req->wb_this_page;
199 } while (req != head);
200
201 return NULL;
202}
203
204/*
205 * nfs_page_group_covers_page
206 * @head - head request of page group
207 *
208 * Return true if the page group with head @head covers the whole page,
209 * returns false otherwise
210 */
211static bool nfs_page_group_covers_page(struct nfs_page *req)
212{
213 struct nfs_page *tmp;
214 unsigned int pos = 0;
215 unsigned int len = nfs_page_length(req->wb_page);
216
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400217 nfs_page_group_lock(req, false);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400218
219 do {
220 tmp = nfs_page_group_search_locked(req->wb_head, pos);
221 if (tmp) {
222 /* no way this should happen */
223 WARN_ON_ONCE(tmp->wb_pgbase != pos);
224 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
225 }
226 } while (tmp && pos < len);
227
228 nfs_page_group_unlock(req);
229 WARN_ON_ONCE(pos > len);
230 return pos == len;
231}
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233/* We can set the PG_uptodate flag if we see that a write request
234 * covers the full page.
235 */
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400236static void nfs_mark_uptodate(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400238 if (PageUptodate(req->wb_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400240 if (!nfs_page_group_covers_page(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400242 SetPageUptodate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
244
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245static int wb_priority(struct writeback_control *wbc)
246{
NeilBrowne87b4c72014-09-18 16:09:27 +1000247 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (wbc->for_reclaim)
Trond Myklebust494f74a2015-12-28 13:34:59 -0500249 return FLUSH_HIGHPRI | FLUSH_COND_STABLE;
NeilBrowne87b4c72014-09-18 16:09:27 +1000250 if (wbc->sync_mode == WB_SYNC_ALL)
251 ret = FLUSH_COND_STABLE;
NeilBrowne87b4c72014-09-18 16:09:27 +1000252 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
255/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800256 * NFS congestion control
257 */
258
259int nfs_congestion_kb;
260
261#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
262#define NFS_CONGESTION_OFF_THRESH \
263 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
264
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400265static void nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800266{
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400267 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400268 int ret = test_set_page_writeback(page);
269
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400270 WARN_ON_ONCE(ret != 0);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800271
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400272 if (atomic_long_inc_return(&nfss->writeback) >
273 NFS_CONGESTION_ON_THRESH) {
274 set_bdi_congested(&nfss->backing_dev_info,
275 BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800276 }
277}
278
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400279static void nfs_end_page_writeback(struct nfs_page *req)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800280{
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400281 struct inode *inode = page_file_mapping(req->wb_page)->host;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800282 struct nfs_server *nfss = NFS_SERVER(inode);
283
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400284 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
285 return;
286
287 end_page_writeback(req->wb_page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700288 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200289 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800290}
291
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400292
293/* nfs_page_group_clear_bits
294 * @req - an nfs request
295 * clears all page group related bits from @req
296 */
297static void
298nfs_page_group_clear_bits(struct nfs_page *req)
Trond Myklebuste261f512006-12-05 00:35:41 -0500299{
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400300 clear_bit(PG_TEARDOWN, &req->wb_flags);
301 clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
302 clear_bit(PG_UPTODATE, &req->wb_flags);
303 clear_bit(PG_WB_END, &req->wb_flags);
304 clear_bit(PG_REMOVE, &req->wb_flags);
305}
306
307
308/*
309 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
310 *
311 * this is a helper function for nfs_lock_and_join_requests
312 *
313 * @inode - inode associated with request page group, must be holding inode lock
314 * @head - head request of page group, must be holding head lock
315 * @req - request that couldn't lock and needs to wait on the req bit lock
316 * @nonblock - if true, don't actually wait
317 *
318 * NOTE: this must be called holding page_group bit lock and inode spin lock
319 * and BOTH will be released before returning.
320 *
321 * returns 0 on success, < 0 on error.
322 */
323static int
324nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
325 struct nfs_page *req, bool nonblock)
326 __releases(&inode->i_lock)
327{
328 struct nfs_page *tmp;
Trond Myklebuste261f512006-12-05 00:35:41 -0500329 int ret;
330
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400331 /* relinquish all the locks successfully grabbed this run */
332 for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
333 nfs_unlock_request(tmp);
334
335 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
336
337 /* grab a ref on the request that will be waited on */
338 kref_get(&req->wb_kref);
339
340 nfs_page_group_unlock(head);
Trond Myklebust587142f2007-07-02 09:57:54 -0400341 spin_unlock(&inode->i_lock);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400342
343 /* release ref from nfs_page_find_head_request_locked */
344 nfs_release_request(head);
345
346 if (!nonblock)
347 ret = nfs_wait_on_request(req);
348 else
349 ret = -EAGAIN;
350 nfs_release_request(req);
351
352 return ret;
353}
354
355/*
356 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
357 *
358 * @destroy_list - request list (using wb_this_page) terminated by @old_head
359 * @old_head - the old head of the list
360 *
361 * All subrequests must be locked and removed from all lists, so at this point
362 * they are only "active" in this function, and possibly in nfs_wait_on_request
363 * with a reference held by some other context.
364 */
365static void
366nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
367 struct nfs_page *old_head)
368{
369 while (destroy_list) {
370 struct nfs_page *subreq = destroy_list;
371
372 destroy_list = (subreq->wb_this_page == old_head) ?
373 NULL : subreq->wb_this_page;
374
375 WARN_ON_ONCE(old_head != subreq->wb_head);
376
377 /* make sure old group is not used */
378 subreq->wb_head = subreq;
379 subreq->wb_this_page = subreq;
380
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400381 /* subreq is now totally disconnected from page group or any
382 * write / commit lists. last chance to wake any waiters */
383 nfs_unlock_request(subreq);
384
385 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
386 /* release ref on old head request */
387 nfs_release_request(old_head);
388
389 nfs_page_group_clear_bits(subreq);
390
391 /* release the PG_INODE_REF reference */
392 if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
393 nfs_release_request(subreq);
394 else
395 WARN_ON_ONCE(1);
396 } else {
397 WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
398 /* zombie requests have already released the last
399 * reference and were waiting on the rest of the
400 * group to complete. Since it's no longer part of a
401 * group, simply free the request */
402 nfs_page_group_clear_bits(subreq);
403 nfs_free_request(subreq);
404 }
405 }
406}
407
408/*
409 * nfs_lock_and_join_requests - join all subreqs to the head req and return
410 * a locked reference, cancelling any pending
411 * operations for this page.
412 *
413 * @page - the page used to lookup the "page group" of nfs_page structures
414 * @nonblock - if true, don't block waiting for request locks
415 *
416 * This function joins all sub requests to the head request by first
417 * locking all requests in the group, cancelling any pending operations
418 * and finally updating the head request to cover the whole range covered by
419 * the (former) group. All subrequests are removed from any write or commit
420 * lists, unlinked from the group and destroyed.
421 *
422 * Returns a locked, referenced pointer to the head request - which after
423 * this call is guaranteed to be the only request associated with the page.
424 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
425 * error was encountered.
426 */
427static struct nfs_page *
428nfs_lock_and_join_requests(struct page *page, bool nonblock)
429{
430 struct inode *inode = page_file_mapping(page)->host;
431 struct nfs_page *head, *subreq;
432 struct nfs_page *destroy_list = NULL;
433 unsigned int total_bytes;
434 int ret;
435
436try_again:
437 total_bytes = 0;
438
439 WARN_ON_ONCE(destroy_list);
440
441 spin_lock(&inode->i_lock);
442
443 /*
444 * A reference is taken only on the head request which acts as a
445 * reference to the whole page group - the group will not be destroyed
446 * until the head reference is released.
447 */
448 head = nfs_page_find_head_request_locked(NFS_I(inode), page);
449
450 if (!head) {
451 spin_unlock(&inode->i_lock);
452 return NULL;
453 }
454
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400455 /* holding inode lock, so always make a non-blocking call to try the
456 * page group lock */
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400457 ret = nfs_page_group_lock(head, true);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400458 if (ret < 0) {
459 spin_unlock(&inode->i_lock);
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400460
461 if (!nonblock && ret == -EAGAIN) {
462 nfs_page_group_lock_wait(head);
463 nfs_release_request(head);
464 goto try_again;
465 }
466
Weston Andros Adamson94970012014-08-08 11:00:56 -0400467 nfs_release_request(head);
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400468 return ERR_PTR(ret);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400469 }
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400470
471 /* lock each request in the page group */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400472 subreq = head;
473 do {
474 /*
475 * Subrequests are always contiguous, non overlapping
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400476 * and in order - but may be repeated (mirrored writes).
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400477 */
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400478 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
479 /* keep track of how many bytes this group covers */
480 total_bytes += subreq->wb_bytes;
481 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
482 ((subreq->wb_offset + subreq->wb_bytes) >
483 (head->wb_offset + total_bytes)))) {
484 nfs_page_group_unlock(head);
485 spin_unlock(&inode->i_lock);
486 return ERR_PTR(-EIO);
487 }
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400488
489 if (!nfs_lock_request(subreq)) {
490 /* releases page group bit lock and
491 * inode spin lock and all references */
492 ret = nfs_unroll_locks_and_wait(inode, head,
493 subreq, nonblock);
494
495 if (ret == 0)
496 goto try_again;
497
498 return ERR_PTR(ret);
499 }
500
501 subreq = subreq->wb_this_page;
502 } while (subreq != head);
503
504 /* Now that all requests are locked, make sure they aren't on any list.
505 * Commit list removal accounting is done after locks are dropped */
506 subreq = head;
507 do {
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400508 nfs_clear_request_commit(subreq);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400509 subreq = subreq->wb_this_page;
510 } while (subreq != head);
511
512 /* unlink subrequests from head, destroy them later */
513 if (head->wb_this_page != head) {
514 /* destroy list will be terminated by head */
515 destroy_list = head->wb_this_page;
516 head->wb_this_page = head;
517
518 /* change head request to cover whole range that
519 * the former page group covered */
520 head->wb_bytes = total_bytes;
521 }
522
523 /*
524 * prepare head request to be added to new pgio descriptor
525 */
526 nfs_page_group_clear_bits(head);
527
528 /*
529 * some part of the group was still on the inode list - otherwise
530 * the group wouldn't be involved in async write.
531 * grab a reference for the head request, iff it needs one.
532 */
533 if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
534 kref_get(&head->wb_kref);
535
536 nfs_page_group_unlock(head);
537
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400538 /* drop lock to clean uprequests on destroy list */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400539 spin_unlock(&inode->i_lock);
540
541 nfs_destroy_unlinked_subrequests(destroy_list, head);
542
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400543 /* still holds ref on head from nfs_page_find_head_request_locked
544 * and still has lock on head from lock loop */
545 return head;
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400546}
547
Peng Tao0bcbf032015-12-05 15:57:31 +0800548static void nfs_write_error_remove_page(struct nfs_page *req)
549{
550 nfs_unlock_request(req);
551 nfs_end_page_writeback(req);
552 nfs_release_request(req);
553 generic_error_remove_page(page_file_mapping(req->wb_page),
554 req->wb_page);
555}
556
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400557/*
558 * Find an associated nfs write request, and prepare to flush it out
559 * May return an error if the user signalled nfs_wait_on_request().
560 */
561static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Peng Taod6c843b2015-12-05 16:20:43 +0800562 struct page *page, bool nonblock,
563 bool launder)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400564{
565 struct nfs_page *req;
566 int ret = 0;
567
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400568 req = nfs_lock_and_join_requests(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400569 if (!req)
570 goto out;
571 ret = PTR_ERR(req);
572 if (IS_ERR(req))
573 goto out;
574
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400575 nfs_set_page_writeback(page);
576 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400577
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400578 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400579 if (!nfs_pageio_add_request(pgio, req)) {
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400580 ret = pgio->pg_error;
Peng Tao0bcbf032015-12-05 15:57:31 +0800581 /*
Peng Taod6c843b2015-12-05 16:20:43 +0800582 * Remove the problematic req upon fatal errors
583 * in launder case, while other dirty pages can
584 * still be around until they get flushed.
Peng Tao0bcbf032015-12-05 15:57:31 +0800585 */
586 if (nfs_error_is_fatal(ret)) {
587 nfs_context_set_write_error(req->wb_context, ret);
Peng Taod6c843b2015-12-05 16:20:43 +0800588 if (launder) {
589 nfs_write_error_remove_page(req);
590 goto out;
591 }
Peng Tao0bcbf032015-12-05 15:57:31 +0800592 }
Peng Taod6c843b2015-12-05 16:20:43 +0800593 nfs_redirty_request(req);
594 ret = -EAGAIN;
Trond Myklebust40f90272015-10-01 11:36:38 -0400595 } else
596 nfs_add_stats(page_file_mapping(page)->host,
597 NFSIOS_WRITEPAGES, 1);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400598out:
599 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500600}
601
Peng Taod6c843b2015-12-05 16:20:43 +0800602static int nfs_do_writepage(struct page *page, struct writeback_control *wbc,
603 struct nfs_pageio_descriptor *pgio, bool launder)
Trond Myklebustf758c8852007-07-22 19:27:32 -0400604{
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400605 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400606
Mel Gormand56b4dd2012-07-31 16:45:06 -0700607 nfs_pageio_cond_complete(pgio, page_file_index(page));
Peng Taod6c843b2015-12-05 16:20:43 +0800608 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE,
609 launder);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400610 if (ret == -EAGAIN) {
611 redirty_page_for_writepage(wbc, page);
612 ret = 0;
613 }
614 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400615}
616
Trond Myklebuste261f512006-12-05 00:35:41 -0500617/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 * Write an mmapped page to the server.
619 */
Peng Taod6c843b2015-12-05 16:20:43 +0800620static int nfs_writepage_locked(struct page *page,
621 struct writeback_control *wbc,
622 bool launder)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400624 struct nfs_pageio_descriptor pgio;
Trond Myklebust40f90272015-10-01 11:36:38 -0400625 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500626 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Trond Myklebust40f90272015-10-01 11:36:38 -0400628 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
629 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200630 false, &nfs_async_write_completion_ops);
Peng Taod6c843b2015-12-05 16:20:43 +0800631 err = nfs_do_writepage(page, wbc, &pgio, launder);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400632 nfs_pageio_complete(&pgio);
633 if (err < 0)
634 return err;
635 if (pgio.pg_error < 0)
636 return pgio.pg_error;
637 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500638}
639
640int nfs_writepage(struct page *page, struct writeback_control *wbc)
641{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400642 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500643
Peng Taod6c843b2015-12-05 16:20:43 +0800644 ret = nfs_writepage_locked(page, wbc, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400646 return ret;
647}
648
649static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
650{
651 int ret;
652
Peng Taod6c843b2015-12-05 16:20:43 +0800653 ret = nfs_do_writepage(page, wbc, data, false);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400654 unlock_page(page);
655 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656}
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
659{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400661 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400662 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 int err;
664
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400665 /* Stop dirtying of new pages while we sync */
NeilBrown74316202014-07-07 15:16:04 +1000666 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400667 nfs_wait_bit_killable, TASK_KILLABLE);
668 if (err)
669 goto out_err;
670
Chuck Lever91d5b472006-03-20 13:44:14 -0500671 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
672
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200673 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
674 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400675 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400676 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400677
678 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100679 smp_mb__after_atomic();
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400680 wake_up_bit(bitlock, NFS_INO_FLUSHING);
681
Trond Myklebustf758c8852007-07-22 19:27:32 -0400682 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400683 goto out_err;
684 err = pgio.pg_error;
685 if (err < 0)
686 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400687 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400688out_err:
689 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
692/*
693 * Insert a write request into an inode
694 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500695static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
697 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400698
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400699 WARN_ON_ONCE(req->wb_this_page != req);
700
Trond Myklebuste7d39062008-06-13 12:12:32 -0400701 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400702 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400703
704 spin_lock(&inode->i_lock);
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500705 if (!nfsi->nrequests &&
706 NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700707 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700708 /*
709 * Swap-space should not get truncated. Hence no need to plug the race
710 * with invalidate/truncate.
711 */
712 if (likely(!PageSwapCache(req->wb_page))) {
713 set_bit(PG_MAPPED, &req->wb_flags);
714 SetPagePrivate(req->wb_page);
715 set_page_private(req->wb_page, (unsigned long)req);
716 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500717 nfsi->nrequests++;
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400718 /* this a head request for a page group - mark it as having an
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500719 * extra reference so sub groups can follow suit.
720 * This flag also informs pgio layer when to bump nrequests when
721 * adding subrequests. */
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400722 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
Trond Myklebustc03b4022007-06-17 13:26:38 -0400723 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400724 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725}
726
727/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800728 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 */
730static void nfs_inode_remove_request(struct nfs_page *req)
731{
David Howells2b0143b2015-03-17 22:25:59 +0000732 struct inode *inode = d_inode(req->wb_context->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400734 struct nfs_page *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400736 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
737 head = req->wb_head;
738
739 spin_lock(&inode->i_lock);
740 if (likely(!PageSwapCache(head->wb_page))) {
741 set_page_private(head->wb_page, 0);
742 ClearPagePrivate(head->wb_page);
NeilBrown95905442014-09-24 11:28:32 +1000743 smp_mb__after_atomic();
744 wake_up_page(head->wb_page, PG_private);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400745 clear_bit(PG_MAPPED, &head->wb_flags);
746 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500747 nfsi->nrequests--;
748 spin_unlock(&inode->i_lock);
749 } else {
750 spin_lock(&inode->i_lock);
751 nfsi->nrequests--;
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400752 spin_unlock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700753 }
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400754
755 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
756 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757}
758
Trond Myklebust61822ab2006-12-05 00:35:42 -0500759static void
Fred6d884e82008-03-19 11:24:38 -0400760nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500761{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500762 __set_page_dirty_nobuffers(req->wb_page);
763}
764
Trond Myklebust3a3908c2014-09-08 22:21:00 -0700765/*
766 * nfs_page_search_commits_for_head_request_locked
767 *
768 * Search through commit lists on @inode for the head request for @page.
769 * Must be called while holding the inode (which is cinfo) lock.
770 *
771 * Returns the head request if found, or NULL if not found.
772 */
773static struct nfs_page *
774nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
775 struct page *page)
776{
777 struct nfs_page *freq, *t;
778 struct nfs_commit_info cinfo;
779 struct inode *inode = &nfsi->vfs_inode;
780
781 nfs_init_cinfo_from_inode(&cinfo, inode);
782
783 /* search through pnfs commit lists */
784 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
785 if (freq)
786 return freq->wb_head;
787
788 /* Linearly search the commit list for the correct request */
789 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
790 if (freq->wb_page == page)
791 return freq->wb_head;
792 }
793
794 return NULL;
795}
796
Trond Myklebust8dd37752012-03-15 17:16:40 -0400797/**
Trond Myklebust86d80f92015-07-31 16:24:30 -0400798 * nfs_request_add_commit_list_locked - add request to a commit list
799 * @req: pointer to a struct nfs_page
800 * @dst: commit list head
801 * @cinfo: holds list lock and accounting info
802 *
803 * This sets the PG_CLEAN bit, updates the cinfo count of
804 * number of outstanding requests requiring a commit as well as
805 * the MM page stats.
806 *
807 * The caller must hold the cinfo->lock, and the nfs_page lock.
808 */
809void
810nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
811 struct nfs_commit_info *cinfo)
812{
813 set_bit(PG_CLEAN, &req->wb_flags);
814 nfs_list_add_request(req, dst);
815 cinfo->mds->ncommit++;
816}
817EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
818
819/**
Trond Myklebust8dd37752012-03-15 17:16:40 -0400820 * nfs_request_add_commit_list - add request to a commit list
821 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400822 * @dst: commit list head
823 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400824 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400825 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400826 * number of outstanding requests requiring a commit as well as
827 * the MM page stats.
828 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400829 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400830 * holding the nfs_page lock.
831 */
832void
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500833nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400834{
Fred Isamanea2cf222012-04-20 14:47:53 -0400835 spin_lock(cinfo->lock);
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500836 nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo);
Fred Isamanea2cf222012-04-20 14:47:53 -0400837 spin_unlock(cinfo->lock);
Trond Myklebust86d80f92015-07-31 16:24:30 -0400838 nfs_mark_page_unstable(req->wb_page, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400839}
840EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
841
842/**
843 * nfs_request_remove_commit_list - Remove request from a commit list
844 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400845 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400846 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400847 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400848 * number of outstanding requests requiring a commit
849 * It does not update the MM page stats.
850 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400851 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400852 */
853void
Fred Isamanea2cf222012-04-20 14:47:53 -0400854nfs_request_remove_commit_list(struct nfs_page *req,
855 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400856{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400857 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
858 return;
859 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400860 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400861}
862EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
863
Fred Isamanea2cf222012-04-20 14:47:53 -0400864static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
865 struct inode *inode)
866{
867 cinfo->lock = &inode->i_lock;
868 cinfo->mds = &NFS_I(inode)->commit_info;
869 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400870 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400871 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400872}
873
874void nfs_init_cinfo(struct nfs_commit_info *cinfo,
875 struct inode *inode,
876 struct nfs_direct_req *dreq)
877{
Fred Isaman1763da12012-04-20 14:47:57 -0400878 if (dreq)
879 nfs_init_cinfo_from_dreq(cinfo, dreq);
880 else
881 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400882}
883EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
886 * Add a request to the inode's commit list.
887 */
Fred Isaman1763da12012-04-20 14:47:57 -0400888void
Fred Isamanea2cf222012-04-20 14:47:53 -0400889nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400890 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891{
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400892 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400893 return;
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500894 nfs_request_add_commit_list(req, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400896
Fred Isamand6d6dc72012-03-08 17:29:35 -0500897static void
898nfs_clear_page_commit(struct page *page)
899{
900 dec_zone_page_state(page, NR_UNSTABLE_NFS);
Tejun Heo93f78d82015-05-22 17:13:27 -0400901 dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
902 WB_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500903}
904
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400905/* Called holding inode (/cinfo) lock */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400906static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400907nfs_clear_request_commit(struct nfs_page *req)
908{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400909 if (test_bit(PG_CLEAN, &req->wb_flags)) {
David Howells2b0143b2015-03-17 22:25:59 +0000910 struct inode *inode = d_inode(req->wb_context->dentry);
Fred Isamanea2cf222012-04-20 14:47:53 -0400911 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400912
Fred Isamanea2cf222012-04-20 14:47:53 -0400913 nfs_init_cinfo_from_inode(&cinfo, inode);
914 if (!pnfs_clear_request_commit(req, &cinfo)) {
Fred Isamanea2cf222012-04-20 14:47:53 -0400915 nfs_request_remove_commit_list(req, &cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400916 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500917 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400918 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400919}
920
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400921int nfs_write_need_commit(struct nfs_pgio_header *hdr)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400922{
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400923 if (hdr->verf.committed == NFS_DATA_SYNC)
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400924 return hdr->lseg == NULL;
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400925 return hdr->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400926}
927
Fred Isaman061ae2e2012-04-20 14:47:48 -0400928static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400929{
Fred Isamanea2cf222012-04-20 14:47:53 -0400930 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400931 unsigned long bytes = 0;
932
933 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
934 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400935 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400936 while (!list_empty(&hdr->pages)) {
937 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400938
939 bytes += req->wb_bytes;
940 nfs_list_remove_request(req);
941 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
942 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400943 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400944 nfs_context_set_write_error(req->wb_context, hdr->error);
945 goto remove_req;
946 }
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400947 if (nfs_write_need_commit(hdr)) {
Anna Schumakerf79d06f2014-05-06 09:12:28 -0400948 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400949 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -0400950 hdr->pgio_mirror_idx);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400951 goto next;
952 }
953remove_req:
954 nfs_inode_remove_request(req);
955next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400956 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400957 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400958 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400959 }
960out:
961 hdr->release(hdr);
962}
963
Anna Schumakerce595152014-05-06 09:12:34 -0400964unsigned long
Fred Isamanea2cf222012-04-20 14:47:53 -0400965nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400966{
Fred Isamanea2cf222012-04-20 14:47:53 -0400967 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400968}
969
Fred Isamanea2cf222012-04-20 14:47:53 -0400970/* cinfo->lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400971int
Fred Isamanea2cf222012-04-20 14:47:53 -0400972nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
973 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500974{
975 struct nfs_page *req, *tmp;
976 int ret = 0;
977
978 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400979 if (!nfs_lock_request(req))
980 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400981 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -0400982 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400983 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400984 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400985 nfs_list_add_request(req, dst);
986 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400987 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400988 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500989 }
990 return ret;
991}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500992
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993/*
994 * nfs_scan_commit - Scan an inode for commit requests
995 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400996 * @dst: mds destination list
997 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 *
999 * Moves requests from the inode's 'commit' request list.
1000 * The requests are *not* checked to ensure that they form a contiguous set.
1001 */
Fred Isaman1763da12012-04-20 14:47:57 -04001002int
Fred Isamanea2cf222012-04-20 14:47:53 -04001003nfs_scan_commit(struct inode *inode, struct list_head *dst,
1004 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
Fred Isamand6d6dc72012-03-08 17:29:35 -05001006 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -04001007
Fred Isamanea2cf222012-04-20 14:47:53 -04001008 spin_lock(cinfo->lock);
1009 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001010 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001011
Fred Isamanea2cf222012-04-20 14:47:53 -04001012 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1013 cinfo, max);
1014 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001015 }
Fred Isamanea2cf222012-04-20 14:47:53 -04001016 spin_unlock(cinfo->lock);
Trond Myklebustff778d02010-02-19 16:53:39 -08001017 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018}
Fred Isamand6d6dc72012-03-08 17:29:35 -05001019
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020/*
Trond Myklebuste7d39062008-06-13 12:12:32 -04001021 * Search for an existing write request, and attempt to update
1022 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 *
Trond Myklebuste7d39062008-06-13 12:12:32 -04001024 * If the attempt fails, then the existing request is flushed out
1025 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 */
Trond Myklebuste7d39062008-06-13 12:12:32 -04001027static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1028 struct page *page,
1029 unsigned int offset,
1030 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
Trond Myklebuste7d39062008-06-13 12:12:32 -04001032 struct nfs_page *req;
1033 unsigned int rqend;
1034 unsigned int end;
1035 int error;
1036
1037 if (!PagePrivate(page))
1038 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001041 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 for (;;) {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001044 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001045 if (req == NULL)
1046 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -05001047
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001048 /* should be handled by nfs_flush_incompatible */
1049 WARN_ON_ONCE(req->wb_head != req);
1050 WARN_ON_ONCE(req->wb_this_page != req);
1051
Trond Myklebuste7d39062008-06-13 12:12:32 -04001052 rqend = req->wb_offset + req->wb_bytes;
1053 /*
1054 * Tell the caller to flush out the request if
1055 * the offsets are non-contiguous.
1056 * Note: nfs_flush_incompatible() will already
1057 * have flushed out requests having wrong owners.
1058 */
Trond Myklebuste468bae2008-06-13 13:25:22 -04001059 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -04001060 || end < req->wb_offset)
1061 goto out_flushme;
1062
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001063 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Trond Myklebuste7d39062008-06-13 12:12:32 -04001066 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -04001067 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001068 error = nfs_wait_on_request(req);
1069 nfs_release_request(req);
1070 if (error != 0)
1071 goto out_err;
1072 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074
1075 /* Okay, the request matches. Update the region */
1076 if (offset < req->wb_offset) {
1077 req->wb_offset = offset;
1078 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 if (end > rqend)
1081 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001082 else
1083 req->wb_bytes = rqend - req->wb_offset;
1084out_unlock:
Fred Isamanca138f32012-04-05 15:26:36 -04001085 if (req)
1086 nfs_clear_request_commit(req);
Weston Andros Adamson411a99a2014-07-17 20:42:19 -04001087 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001088 return req;
1089out_flushme:
1090 spin_unlock(&inode->i_lock);
1091 nfs_release_request(req);
1092 error = nfs_wb_page(inode, page);
1093out_err:
1094 return ERR_PTR(error);
1095}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
Trond Myklebuste7d39062008-06-13 12:12:32 -04001097/*
1098 * Try to update an existing write request, or create one if there is none.
1099 *
1100 * Note: Should always be called with the Page Lock held to prevent races
1101 * if we have to add a new request. Also assumes that the caller has
1102 * already called nfs_flush_incompatible() if necessary.
1103 */
1104static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1105 struct page *page, unsigned int offset, unsigned int bytes)
1106{
Mel Gormand56b4dd2012-07-31 16:45:06 -07001107 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001108 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001109
1110 req = nfs_try_to_update_request(inode, page, offset, bytes);
1111 if (req != NULL)
1112 goto out;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001113 req = nfs_create_request(ctx, page, NULL, offset, bytes);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001114 if (IS_ERR(req))
1115 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001116 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -04001117out:
Trond Myklebust61e930a2007-10-18 17:08:05 -04001118 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119}
1120
Trond Myklebuste7d39062008-06-13 12:12:32 -04001121static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1122 unsigned int offset, unsigned int count)
1123{
1124 struct nfs_page *req;
1125
1126 req = nfs_setup_write_request(ctx, page, offset, count);
1127 if (IS_ERR(req))
1128 return PTR_ERR(req);
1129 /* Update file length */
1130 nfs_grow_file(page, offset, count);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -04001131 nfs_mark_uptodate(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001132 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001133 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001134 return 0;
1135}
1136
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137int nfs_flush_incompatible(struct file *file, struct page *page)
1138{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001139 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -04001140 struct nfs_lock_context *l_ctx;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001141 struct file_lock_context *flctx = file_inode(file)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -05001143 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 /*
1145 * Look for a request corresponding to this page. If there
1146 * is one, and it belongs to another file, we flush it out
1147 * before we try to copy anything into the page. Do this
1148 * due to the lack of an ACCESS-type call in NFSv2.
1149 * Also do the same if we find a request from an existing
1150 * dropped page.
1151 */
Trond Myklebust1a545332006-12-05 00:35:40 -05001152 do {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001153 req = nfs_page_find_head_request(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001154 if (req == NULL)
1155 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -04001156 l_ctx = req->wb_lock_context;
Trond Myklebust138a2932015-10-01 17:17:06 -04001157 do_flush = req->wb_page != page ||
1158 !nfs_match_open_context(req->wb_context, ctx);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001159 /* for now, flush if more than 1 request in page_group */
1160 do_flush |= req->wb_this_page != req;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001161 if (l_ctx && flctx &&
1162 !(list_empty_careful(&flctx->flc_posix) &&
1163 list_empty_careful(&flctx->flc_flock))) {
Jeff Layton5263e312015-01-16 15:05:55 -05001164 do_flush |= l_ctx->lockowner.l_owner != current->files
1165 || l_ctx->lockowner.l_pid != current->tgid;
1166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -05001168 if (!do_flush)
1169 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -07001170 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001171 } while (status == 0);
1172 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175/*
Andy Adamsondc248262013-08-14 11:59:16 -04001176 * Avoid buffered writes when a open context credential's key would
1177 * expire soon.
1178 *
1179 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1180 *
1181 * Return 0 and set a credential flag which triggers the inode to flush
1182 * and performs NFS_FILE_SYNC writes if the key will expired within
1183 * RPC_KEY_EXPIRE_TIMEO.
1184 */
1185int
1186nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1187{
1188 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1189 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1190
1191 return rpcauth_key_timeout_notify(auth, ctx->cred);
1192}
1193
1194/*
1195 * Test if the open context credential key is marked to expire soon.
1196 */
1197bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1198{
1199 return rpcauth_cred_key_to_expire(ctx->cred);
1200}
1201
1202/*
Trond Myklebust5d47a352008-02-07 17:24:07 -05001203 * If the page cache is marked as unsafe or invalid, then we can't rely on
1204 * the PageUptodate() flag. In this case, we will need to turn off
1205 * write optimisations that depend on the page contents being correct.
1206 */
Trond Myklebust8d197a52012-04-29 12:50:01 -04001207static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -05001208{
Jeff Laytond529ef82014-01-27 13:46:15 -05001209 struct nfs_inode *nfsi = NFS_I(inode);
1210
Trond Myklebust8d197a52012-04-29 12:50:01 -04001211 if (nfs_have_delegated_attributes(inode))
1212 goto out;
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001213 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
Jeff Laytond529ef82014-01-27 13:46:15 -05001214 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -05001215 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -05001216 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -04001217 return false;
1218out:
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001219 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1220 return false;
Trond Myklebust8d197a52012-04-29 12:50:01 -04001221 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -05001222}
1223
Jeff Layton5263e312015-01-16 15:05:55 -05001224static bool
1225is_whole_file_wrlock(struct file_lock *fl)
1226{
1227 return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1228 fl->fl_type == F_WRLCK;
1229}
1230
Scott Mayhewc7559662013-07-05 17:33:19 -04001231/* If we know the page is up to date, and we're not using byte range locks (or
1232 * if we have the whole file locked for writing), it may be more efficient to
1233 * extend the write to cover the entire page in order to avoid fragmentation
1234 * inefficiencies.
1235 *
Scott Mayhew263b4502014-01-17 15:12:05 -05001236 * If the file is opened for synchronous writes then we can just skip the rest
1237 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -04001238 */
1239static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1240{
Jeff Layton5263e312015-01-16 15:05:55 -05001241 int ret;
1242 struct file_lock_context *flctx = inode->i_flctx;
1243 struct file_lock *fl;
1244
Scott Mayhewc7559662013-07-05 17:33:19 -04001245 if (file->f_flags & O_DSYNC)
1246 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -05001247 if (!nfs_write_pageuptodate(page, inode))
1248 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -04001249 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1250 return 1;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001251 if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1252 list_empty_careful(&flctx->flc_posix)))
Trond Myklebust8fa45922015-10-01 18:38:27 -04001253 return 1;
Jeff Layton5263e312015-01-16 15:05:55 -05001254
1255 /* Check to see if there are whole file write locks */
Jeff Layton5263e312015-01-16 15:05:55 -05001256 ret = 0;
Jeff Layton6109c852015-01-16 15:05:57 -05001257 spin_lock(&flctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001258 if (!list_empty(&flctx->flc_posix)) {
1259 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1260 fl_list);
1261 if (is_whole_file_wrlock(fl))
1262 ret = 1;
1263 } else if (!list_empty(&flctx->flc_flock)) {
Jeff Layton5263e312015-01-16 15:05:55 -05001264 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1265 fl_list);
1266 if (fl->fl_type == F_WRLCK)
1267 ret = 1;
1268 }
Jeff Layton6109c852015-01-16 15:05:57 -05001269 spin_unlock(&flctx->flc_lock);
Jeff Layton5263e312015-01-16 15:05:55 -05001270 return ret;
Scott Mayhewc7559662013-07-05 17:33:19 -04001271}
1272
Trond Myklebust5d47a352008-02-07 17:24:07 -05001273/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 * Update and possibly write a cached page of an NFS file.
1275 *
1276 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1277 * things with a page scheduled for an RPC call (e.g. invalidate it).
1278 */
1279int nfs_updatepage(struct file *file, struct page *page,
1280 unsigned int offset, unsigned int count)
1281{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001282 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001283 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 int status = 0;
1285
Chuck Lever91d5b472006-03-20 13:44:14 -05001286 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1287
Al Viro6de14722013-09-16 10:53:17 -04001288 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1289 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Scott Mayhewc7559662013-07-05 17:33:19 -04001291 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -05001292 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295
Trond Myklebuste21195a2006-12-05 00:35:39 -05001296 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -04001297 if (status < 0)
1298 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -07001299 else
1300 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
Chuck Lever48186c72008-06-11 17:56:05 -04001302 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 return status;
1305}
1306
Trond Myklebust3ff75762007-07-14 15:40:00 -04001307static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
1309 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1310 case FLUSH_HIGHPRI:
1311 return RPC_PRIORITY_HIGH;
1312 case FLUSH_LOWPRI:
1313 return RPC_PRIORITY_LOW;
1314 }
1315 return RPC_PRIORITY_NORMAL;
1316}
1317
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001318static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1319 struct rpc_message *msg,
Tom Haynesabde71f2014-06-09 13:12:20 -07001320 const struct nfs_rpc_ops *rpc_ops,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001321 struct rpc_task_setup *task_setup_data, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
Trond Myklebust3ff75762007-07-14 15:40:00 -04001323 int priority = flush_task_priority(how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001325 task_setup_data->priority = priority;
Tom Haynesabde71f2014-06-09 13:12:20 -07001326 rpc_ops->write_setup(hdr, msg);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001327
Tom Haynesabde71f2014-06-09 13:12:20 -07001328 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001329 &task_setup_data->rpc_client, msg, hdr);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001330}
1331
Fred6d884e82008-03-19 11:24:38 -04001332/* If a nfs_flush_* function fails, it should remove reqs from @head and
1333 * call this on each, which will prepare them to be retried on next
1334 * writeback using standard nfs.
1335 */
1336static void nfs_redirty_request(struct nfs_page *req)
1337{
1338 nfs_mark_request_dirty(req);
Trond Myklebustc7070112015-06-17 19:56:22 -04001339 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001340 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -04001341 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001342 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001343}
1344
Fred Isaman061ae2e2012-04-20 14:47:48 -04001345static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001346{
1347 struct nfs_page *req;
1348
1349 while (!list_empty(head)) {
1350 req = nfs_list_entry(head->next);
1351 nfs_list_remove_request(req);
1352 nfs_redirty_request(req);
1353 }
1354}
1355
Trond Myklebustdc602dd2015-12-31 11:44:06 -05001356static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
1357{
1358 nfs_async_write_error(&hdr->pages);
1359}
1360
Fred Isaman061ae2e2012-04-20 14:47:48 -04001361static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1362 .error_cleanup = nfs_async_write_error,
1363 .completion = nfs_write_completion,
Trond Myklebustdc602dd2015-12-31 11:44:06 -05001364 .reschedule_io = nfs_async_write_reschedule_io,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001365};
1366
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001367void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001368 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001369 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001370{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001371 struct nfs_server *server = NFS_SERVER(inode);
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001372 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001373
1374#ifdef CONFIG_NFS_V4_1
1375 if (server->pnfs_curr_ld && !force_mds)
1376 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1377#endif
Anna Schumaker4a0de552014-05-06 09:12:30 -04001378 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1379 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001380}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001381EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001382
Trond Myklebustdce81292011-07-13 15:59:19 -04001383void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1384{
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001385 struct nfs_pgio_mirror *mirror;
1386
Kinglong Mee6f29b9b2015-09-20 23:03:28 +08001387 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1388 pgio->pg_ops->pg_cleanup(pgio);
1389
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001390 pgio->pg_ops = &nfs_pgio_rw_ops;
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001391
1392 nfs_pageio_stop_mirroring(pgio);
1393
1394 mirror = &pgio->pg_mirrors[0];
1395 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
Trond Myklebustdce81292011-07-13 15:59:19 -04001396}
Trond Myklebust1f945352011-07-13 15:59:57 -04001397EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001398
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Fred Isaman0b7c0152012-04-20 14:47:39 -04001400void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1401{
1402 struct nfs_commit_data *data = calldata;
1403
1404 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1405}
1406
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001407/*
1408 * Special version of should_remove_suid() that ignores capabilities.
1409 */
1410static int nfs_should_remove_suid(const struct inode *inode)
1411{
1412 umode_t mode = inode->i_mode;
1413 int kill = 0;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001414
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001415 /* suid always must be killed */
1416 if (unlikely(mode & S_ISUID))
1417 kill = ATTR_KILL_SUID;
1418
1419 /*
1420 * sgid without any exec bits is just a mandatory locking mark; leave
1421 * it alone. If some exec bits are set, it's a real sgid; kill it.
1422 */
1423 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1424 kill |= ATTR_KILL_SGID;
1425
1426 if (unlikely(kill && S_ISREG(mode)))
1427 return kill;
1428
1429 return 0;
1430}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001431
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001432static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1433 struct nfs_fattr *fattr)
1434{
1435 struct nfs_pgio_args *argp = &hdr->args;
1436 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001437 u64 size = argp->offset + resp->count;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001438
1439 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001440 fattr->size = size;
1441 if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1442 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001443 return;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001444 }
1445 if (size != fattr->size)
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001446 return;
1447 /* Set attribute barrier */
1448 nfs_fattr_set_barrier(fattr);
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001449 /* ...and update size */
1450 fattr->valid |= NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001451}
1452
1453void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1454{
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001455 struct nfs_fattr *fattr = &hdr->fattr;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001456 struct inode *inode = hdr->inode;
1457
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001458 spin_lock(&inode->i_lock);
1459 nfs_writeback_check_extend(hdr, fattr);
1460 nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1461 spin_unlock(&inode->i_lock);
1462}
1463EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465/*
1466 * This function is called when the WRITE call is complete.
1467 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001468static int nfs_writeback_done(struct rpc_task *task,
1469 struct nfs_pgio_header *hdr,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001470 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001472 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
Chuck Leverf551e442006-09-20 14:33:04 -04001474 /*
1475 * ->write_done will attempt to use post-op attributes to detect
1476 * conflicting writes by other clients. A strict interpretation
1477 * of close-to-open would allow us to continue caching even if
1478 * another writer had changed the file, but some applications
1479 * depend on tighter cache coherency when writing.
1480 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001481 status = NFS_PROTO(inode)->write_done(task, hdr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001482 if (status != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001483 return status;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001484 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001485
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001486 if (hdr->res.verf->committed < hdr->args.stable &&
1487 task->tk_status >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 /* We tried a write call, but the server did not
1489 * commit data to stable storage even though we
1490 * requested it.
1491 * Note: There is a known bug in Tru64 < 5.0 in which
1492 * the server reports NFS_DATA_SYNC, but performs
1493 * NFS_FILE_SYNC. We therefore implement this checking
1494 * as a dprintk() in order to avoid filling syslog.
1495 */
1496 static unsigned long complain;
1497
Fred Isamana69aef12011-03-03 15:13:47 +00001498 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001500 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001502 NFS_SERVER(inode)->nfs_client->cl_hostname,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001503 hdr->res.verf->committed, hdr->args.stable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 complain = jiffies + 300 * HZ;
1505 }
1506 }
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001507
1508 /* Deal with the suid/sgid bit corner case */
1509 if (nfs_should_remove_suid(inode))
1510 nfs_mark_for_revalidate(inode);
Anna Schumaker0eecb212014-05-06 09:12:32 -04001511 return 0;
1512}
1513
1514/*
1515 * This function is called when the WRITE call is complete.
1516 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001517static void nfs_writeback_result(struct rpc_task *task,
1518 struct nfs_pgio_header *hdr)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001519{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001520 struct nfs_pgio_args *argp = &hdr->args;
1521 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001522
1523 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 static unsigned long complain;
1525
Fred Isaman6c75dc02012-04-20 14:47:47 -04001526 /* This a short write! */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001527 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001530 if (resp->count == 0) {
1531 if (time_before(complain, jiffies)) {
1532 printk(KERN_WARNING
1533 "NFS: Server wrote zero bytes, expected %u.\n",
1534 argp->count);
1535 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 }
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001537 nfs_set_pgio_error(hdr, -EIO, argp->offset);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001538 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001539 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 }
Kinglong Meef8417b482015-10-16 17:23:29 +08001541
1542 /* For non rpc-based layout drivers, retry-through-MDS */
1543 if (!task->tk_ops) {
1544 hdr->pnfs_error = -EAGAIN;
1545 return;
1546 }
1547
Fred Isaman6c75dc02012-04-20 14:47:47 -04001548 /* Was this an NFSv2 write or an NFSv3 stable write? */
1549 if (resp->verf->committed != NFS_UNSTABLE) {
1550 /* Resend from where the server left off */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001551 hdr->mds_offset += resp->count;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001552 argp->offset += resp->count;
1553 argp->pgbase += resp->count;
1554 argp->count -= resp->count;
1555 } else {
1556 /* Resend as a stable write in order to avoid
1557 * headaches in the case of a server crash.
1558 */
1559 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001561 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001565static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001566{
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001567 return wait_on_atomic_t(&cinfo->rpcs_out,
1568 nfs_wait_atomic_killable, TASK_KILLABLE);
1569}
1570
1571static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
1572{
1573 atomic_inc(&cinfo->rpcs_out);
1574}
1575
1576static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
1577{
1578 if (atomic_dec_and_test(&cinfo->rpcs_out))
1579 wake_up_atomic_t(&cinfo->rpcs_out);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001580}
1581
Fred Isaman0b7c0152012-04-20 14:47:39 -04001582void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001584 put_nfs_open_context(data->context);
1585 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
Fred Isamane0c2b382011-03-23 13:27:53 +00001587EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
Fred Isaman0b7c0152012-04-20 14:47:39 -04001589int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Peng Taoc36aae92014-06-09 07:10:14 +08001590 const struct nfs_rpc_ops *nfs_ops,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001591 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001592 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593{
Trond Myklebust07737692007-10-25 18:42:54 -04001594 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001595 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001596 struct rpc_message msg = {
1597 .rpc_argp = &data->args,
1598 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001599 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001600 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001601 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001602 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001603 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001604 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001605 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001606 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001607 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001608 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001609 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001610 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001611 /* Set up the initial task struct. */
Peng Taoc36aae92014-06-09 07:10:14 +08001612 nfs_ops->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
Kinglong Meeb4839eb2015-07-01 12:00:13 +08001614 dprintk("NFS: initiated commit call\n");
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001615
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001616 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1617 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1618
Trond Myklebust07737692007-10-25 18:42:54 -04001619 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001620 if (IS_ERR(task))
1621 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001622 if (how & FLUSH_SYNC)
1623 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001624 rpc_put_task(task);
1625 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626}
Fred Isamane0c2b382011-03-23 13:27:53 +00001627EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
Peng Tao378520b2014-08-07 10:15:02 +08001629static loff_t nfs_get_lwb(struct list_head *head)
1630{
1631 loff_t lwb = 0;
1632 struct nfs_page *req;
1633
1634 list_for_each_entry(req, head, wb_list)
1635 if (lwb < (req_offset(req) + req->wb_bytes))
1636 lwb = req_offset(req) + req->wb_bytes;
1637
1638 return lwb;
1639}
1640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001642 * Set up the argument/result storage required for the RPC call.
1643 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001644void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001645 struct list_head *head,
1646 struct pnfs_layout_segment *lseg,
1647 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001648{
1649 struct nfs_page *first = nfs_list_entry(head->next);
David Howells2b0143b2015-03-17 22:25:59 +00001650 struct inode *inode = d_inode(first->wb_context->dentry);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001651
1652 /* Set up the RPC argument and reply structs
1653 * NB: take care not to mess about with data->commit et al. */
1654
1655 list_splice_init(head, &data->pages);
1656
1657 data->inode = inode;
1658 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001659 data->lseg = lseg; /* reference transferred */
Peng Tao378520b2014-08-07 10:15:02 +08001660 /* only set lwb for pnfs commit */
1661 if (lseg)
1662 data->lwb = nfs_get_lwb(&data->pages);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001663 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001664 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001665 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001666
1667 data->args.fh = NFS_FH(data->inode);
1668 /* Note: we always request a commit of the entire inode */
1669 data->args.offset = 0;
1670 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001671 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001672 data->res.fattr = &data->fattr;
1673 data->res.verf = &data->verf;
1674 nfs_fattr_init(&data->fattr);
1675}
Fred Isamane0c2b382011-03-23 13:27:53 +00001676EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001677
Fred Isamane0c2b382011-03-23 13:27:53 +00001678void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001679 struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001680 struct nfs_commit_info *cinfo,
1681 u32 ds_commit_idx)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001682{
1683 struct nfs_page *req;
1684
1685 while (!list_empty(page_list)) {
1686 req = nfs_list_entry(page_list->next);
1687 nfs_list_remove_request(req);
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001688 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
Tom Haynes487b9b82015-02-13 13:19:54 -08001689 if (!cinfo->dreq)
1690 nfs_clear_page_commit(req->wb_page);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001691 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001692 }
1693}
Fred Isamane0c2b382011-03-23 13:27:53 +00001694EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001695
Trond Myklebustb20135d2015-12-31 09:28:06 -05001696static void
1697nfs_commit_resched_write(struct nfs_commit_info *cinfo,
1698 struct nfs_page *req)
1699{
1700 __set_page_dirty_nobuffers(req->wb_page);
1701}
1702
Fred Isaman9ace33c2011-03-23 13:27:45 +00001703/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 * Commit dirty pages
1705 */
1706static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001707nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1708 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001710 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001712 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713
1714 if (!data)
1715 goto out_bad;
1716
1717 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001718 nfs_init_commit(data, head, NULL, cinfo);
1719 atomic_inc(&cinfo->mds->rpcs_out);
Peng Taoc36aae92014-06-09 07:10:14 +08001720 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1721 data->mds_ops, how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 out_bad:
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001723 nfs_retry_commit(head, NULL, cinfo, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 return -ENOMEM;
1725}
1726
1727/*
1728 * COMMIT call returned
1729 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001730static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001732 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
Chuck Levera3f565b2007-01-31 12:14:01 -05001734 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 task->tk_pid, task->tk_status);
1736
Trond Myklebust788e7a82006-03-20 13:44:27 -05001737 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001738 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001739}
1740
Fred Isamanf453a542012-04-20 14:47:54 -04001741static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001742{
Fred Isaman5917ce82011-03-23 13:27:48 +00001743 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001744 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001745 struct nfs_commit_info cinfo;
NeilBrown353db792014-09-24 11:28:32 +10001746 struct nfs_server *nfss;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 while (!list_empty(&data->pages)) {
1749 req = nfs_list_entry(data->pages.next);
1750 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001751 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
Niels de Vos1e8968c2013-12-17 18:20:16 +01001753 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001754 req->wb_context->dentry->d_sb->s_id,
David Howells2b0143b2015-03-17 22:25:59 +00001755 (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 req->wb_bytes,
1757 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001758 if (status < 0) {
1759 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001761 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 goto next;
1763 }
1764
1765 /* Okay, COMMIT succeeded, apparently. Check the verifier
1766 * returned by the server against all stored verfs. */
Trond Myklebust2f2c63b2012-06-08 11:56:09 -04001767 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 /* We have a match */
1769 nfs_inode_remove_request(req);
1770 dprintk(" OK\n");
1771 goto next;
1772 }
1773 /* We have a mismatch. Write the page again */
1774 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001775 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001776 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001778 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 }
NeilBrown353db792014-09-24 11:28:32 +10001780 nfss = NFS_SERVER(data->inode);
1781 if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1782 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1783
Fred Isamanf453a542012-04-20 14:47:54 -04001784 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001785 nfs_commit_end(cinfo.mds);
Fred Isaman5917ce82011-03-23 13:27:48 +00001786}
1787
1788static void nfs_commit_release(void *calldata)
1789{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001790 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001791
Fred Isamanf453a542012-04-20 14:47:54 -04001792 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001793 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001795
1796static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001797 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001798 .rpc_call_done = nfs_commit_done,
1799 .rpc_release = nfs_commit_release,
1800};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Fred Isamanf453a542012-04-20 14:47:54 -04001802static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1803 .completion = nfs_commit_release_pages,
Trond Myklebustb20135d2015-12-31 09:28:06 -05001804 .resched_write = nfs_commit_resched_write,
Fred Isamanf453a542012-04-20 14:47:54 -04001805};
1806
Fred Isaman1763da12012-04-20 14:47:57 -04001807int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1808 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001809{
1810 int status;
1811
Fred Isamanea2cf222012-04-20 14:47:53 -04001812 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001813 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001814 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001815 return status;
1816}
1817
Trond Myklebustb608b282010-07-30 15:31:54 -04001818int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001821 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001822 int may_wait = how & FLUSH_SYNC;
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001823 int error = 0;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001824 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Fred Isamanea2cf222012-04-20 14:47:53 -04001826 nfs_init_cinfo_from_inode(&cinfo, inode);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001827 nfs_commit_begin(cinfo.mds);
Fred Isamanea2cf222012-04-20 14:47:53 -04001828 res = nfs_scan_commit(inode, &head, &cinfo);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001829 if (res)
Fred Isamanea2cf222012-04-20 14:47:53 -04001830 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001831 nfs_commit_end(cinfo.mds);
1832 if (error < 0)
1833 goto out_error;
1834 if (!may_wait)
1835 goto out_mark_dirty;
1836 error = wait_on_commit(cinfo.mds);
1837 if (error < 0)
1838 return error;
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001839 return res;
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001840out_error:
1841 res = error;
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001842 /* Note: If we exit without ensuring that the commit is complete,
1843 * we must mark the inode as dirty. Otherwise, future calls to
1844 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1845 * that the data is on the disk.
1846 */
1847out_mark_dirty:
1848 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 return res;
1850}
Trond Myklebustb20135d2015-12-31 09:28:06 -05001851EXPORT_SYMBOL_GPL(nfs_commit_inode);
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001852
Anna Schumakerae09c312015-07-13 14:01:32 -04001853int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001854{
Trond Myklebust420e3642010-02-19 17:00:02 -08001855 struct nfs_inode *nfsi = NFS_I(inode);
1856 int flags = FLUSH_SYNC;
1857 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001858
Jeff Layton3236c3e2011-10-11 09:49:21 -04001859 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001860 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e2011-10-11 09:49:21 -04001861 return ret;
1862
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001863 if (wbc->sync_mode == WB_SYNC_NONE) {
1864 /* Don't commit yet if this is a non-blocking flush and there
1865 * are a lot of outstanding writes for this mapping.
1866 */
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -05001867 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001868 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001869
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001870 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001871 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001872 }
1873
Trond Myklebust420e3642010-02-19 17:00:02 -08001874 ret = nfs_commit_inode(inode, flags);
1875 if (ret >= 0) {
1876 if (wbc->sync_mode == WB_SYNC_NONE) {
1877 if (ret < wbc->nr_to_write)
1878 wbc->nr_to_write -= ret;
1879 else
1880 wbc->nr_to_write = 0;
1881 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001882 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001883 }
1884out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001885 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1886 return ret;
1887}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001888EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001889
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001890/*
1891 * flush the inode to disk.
1892 */
1893int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001894{
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001895 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001896
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001897 trace_nfs_writeback_inode_enter(inode);
1898
Trond Myklebust5bb89b42015-03-25 14:14:42 -04001899 ret = filemap_write_and_wait(inode->i_mapping);
Chuck Lever6b196872015-05-06 18:26:58 -04001900 if (ret)
1901 goto out;
1902 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1903 if (ret < 0)
1904 goto out;
1905 pnfs_sync_inode(inode, true);
1906 ret = 0;
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001907
Chuck Lever6b196872015-05-06 18:26:58 -04001908out:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001909 trace_nfs_writeback_inode_exit(inode, ret);
1910 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001911}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001912EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001913
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001914int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1915{
1916 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001917 int ret = 0;
1918
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001919 wait_on_page_writeback(page);
1920
1921 /* blocking call to cancel all requests and join to a single (head)
1922 * request */
1923 req = nfs_lock_and_join_requests(page, false);
1924
1925 if (IS_ERR(req)) {
1926 ret = PTR_ERR(req);
1927 } else if (req) {
1928 /* all requests from this page have been cancelled by
1929 * nfs_lock_and_join_requests, so just remove the head
1930 * request from the inode / page_private pointer and
1931 * release it */
1932 nfs_inode_remove_request(req);
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001933 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001934 }
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001935
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001936 return ret;
1937}
1938
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001939/*
1940 * Write back all requests on one page - we do this before reading it.
1941 */
Peng Taod6c843b2015-12-05 16:20:43 +08001942int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder)
Trond Myklebust1c759502006-10-09 16:18:38 -04001943{
Mel Gorman29418aa2012-07-31 16:45:10 -07001944 loff_t range_start = page_file_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001945 loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001946 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001947 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001948 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001949 .range_start = range_start,
1950 .range_end = range_end,
1951 };
1952 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001953
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001954 trace_nfs_writeback_page_enter(inode);
1955
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001956 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001957 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001958 if (clear_page_dirty_for_io(page)) {
Peng Taod6c843b2015-12-05 16:20:43 +08001959 ret = nfs_writepage_locked(page, &wbc, launder);
Trond Myklebust73e33022008-04-11 16:03:54 -04001960 if (ret < 0)
1961 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001962 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001963 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001964 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001965 if (!PagePrivate(page))
1966 break;
1967 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001968 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001969 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001970 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001971out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001972 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001973 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001974}
1975
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001976#ifdef CONFIG_MIGRATION
1977int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001978 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001979{
Jeff Layton2da95652011-10-12 10:57:42 -04001980 /*
1981 * If PagePrivate is set, then the page is currently associated with
1982 * an in-progress read or write request. Don't try to migrate it.
1983 *
1984 * FIXME: we could do this in principle, but we'll need a way to ensure
1985 * that we can safely release the inode reference while holding
1986 * the page lock.
1987 */
1988 if (PagePrivate(page))
1989 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001990
David Howells8c209ce2012-12-05 13:34:49 +00001991 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1992 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001993
Mel Gormana6bc32b2012-01-12 17:19:43 -08001994 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001995}
1996#endif
1997
David Howellsf7b422b2006-06-09 09:34:33 -04001998int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999{
2000 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04002001 sizeof(struct nfs_pgio_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09002003 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 if (nfs_wdata_cachep == NULL)
2005 return -ENOMEM;
2006
Matthew Dobson93d23412006-03-26 01:37:50 -08002007 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2008 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002010 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Fred Isaman0b7c0152012-04-20 14:47:39 -04002012 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2013 sizeof(struct nfs_commit_data),
2014 0, SLAB_HWCACHE_ALIGN,
2015 NULL);
2016 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002017 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04002018
Matthew Dobson93d23412006-03-26 01:37:50 -08002019 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08002020 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002022 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Peter Zijlstra89a09142007-03-16 13:38:26 -08002024 /*
2025 * NFS congestion size, scale with available memory.
2026 *
2027 * 64MB: 8192k
2028 * 128MB: 11585k
2029 * 256MB: 16384k
2030 * 512MB: 23170k
2031 * 1GB: 32768k
2032 * 2GB: 46340k
2033 * 4GB: 65536k
2034 * 8GB: 92681k
2035 * 16GB: 131072k
2036 *
2037 * This allows larger machines to have larger/more transfers.
2038 * Limit the default to 256M
2039 */
2040 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2041 if (nfs_congestion_kb > 256*1024)
2042 nfs_congestion_kb = 256*1024;
2043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04002045
2046out_destroy_commit_cache:
2047 kmem_cache_destroy(nfs_cdata_cachep);
2048out_destroy_write_mempool:
2049 mempool_destroy(nfs_wdata_mempool);
2050out_destroy_write_cache:
2051 kmem_cache_destroy(nfs_wdata_cachep);
2052 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053}
2054
David Brownell266bee82006-06-27 12:59:15 -07002055void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056{
2057 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04002058 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002060 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061}
2062
Anna Schumaker4a0de552014-05-06 09:12:30 -04002063static const struct nfs_rw_ops nfs_rw_write_ops = {
Anna Schumakera4cdda52014-05-06 09:12:31 -04002064 .rw_mode = FMODE_WRITE,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002065 .rw_alloc_header = nfs_writehdr_alloc,
2066 .rw_free_header = nfs_writehdr_free,
Anna Schumaker0eecb212014-05-06 09:12:32 -04002067 .rw_done = nfs_writeback_done,
2068 .rw_result = nfs_writeback_result,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04002069 .rw_initiate = nfs_initiate_write,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002070};