blob: e4772a8340f831260e68906a48bdaa24aca5afba [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;
Huang Ying8cd79782016-10-07 17:00:24 -0700154 if (i_size > 0 && page_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;
Trond Myklebustcca588d2016-05-16 17:42:45 -0400248
NeilBrowne87b4c72014-09-18 16:09:27 +1000249 if (wbc->sync_mode == WB_SYNC_ALL)
250 ret = FLUSH_COND_STABLE;
NeilBrowne87b4c72014-09-18 16:09:27 +1000251 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252}
253
254/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800255 * NFS congestion control
256 */
257
258int nfs_congestion_kb;
259
260#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
261#define NFS_CONGESTION_OFF_THRESH \
262 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
263
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400264static void nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800265{
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400266 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400267 int ret = test_set_page_writeback(page);
268
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400269 WARN_ON_ONCE(ret != 0);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800270
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400271 if (atomic_long_inc_return(&nfss->writeback) >
272 NFS_CONGESTION_ON_THRESH) {
273 set_bdi_congested(&nfss->backing_dev_info,
274 BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800275 }
276}
277
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400278static void nfs_end_page_writeback(struct nfs_page *req)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800279{
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400280 struct inode *inode = page_file_mapping(req->wb_page)->host;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800281 struct nfs_server *nfss = NFS_SERVER(inode);
282
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400283 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
284 return;
285
286 end_page_writeback(req->wb_page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700287 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200288 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800289}
290
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400291
292/* nfs_page_group_clear_bits
293 * @req - an nfs request
294 * clears all page group related bits from @req
295 */
296static void
297nfs_page_group_clear_bits(struct nfs_page *req)
Trond Myklebuste261f512006-12-05 00:35:41 -0500298{
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400299 clear_bit(PG_TEARDOWN, &req->wb_flags);
300 clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
301 clear_bit(PG_UPTODATE, &req->wb_flags);
302 clear_bit(PG_WB_END, &req->wb_flags);
303 clear_bit(PG_REMOVE, &req->wb_flags);
304}
305
306
307/*
308 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
309 *
310 * this is a helper function for nfs_lock_and_join_requests
311 *
312 * @inode - inode associated with request page group, must be holding inode lock
313 * @head - head request of page group, must be holding head lock
314 * @req - request that couldn't lock and needs to wait on the req bit lock
315 * @nonblock - if true, don't actually wait
316 *
317 * NOTE: this must be called holding page_group bit lock and inode spin lock
318 * and BOTH will be released before returning.
319 *
320 * returns 0 on success, < 0 on error.
321 */
322static int
323nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
324 struct nfs_page *req, bool nonblock)
325 __releases(&inode->i_lock)
326{
327 struct nfs_page *tmp;
Trond Myklebuste261f512006-12-05 00:35:41 -0500328 int ret;
329
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400330 /* relinquish all the locks successfully grabbed this run */
331 for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
332 nfs_unlock_request(tmp);
333
334 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
335
336 /* grab a ref on the request that will be waited on */
337 kref_get(&req->wb_kref);
338
339 nfs_page_group_unlock(head);
Trond Myklebust587142f2007-07-02 09:57:54 -0400340 spin_unlock(&inode->i_lock);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400341
342 /* release ref from nfs_page_find_head_request_locked */
343 nfs_release_request(head);
344
345 if (!nonblock)
346 ret = nfs_wait_on_request(req);
347 else
348 ret = -EAGAIN;
349 nfs_release_request(req);
350
351 return ret;
352}
353
354/*
355 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
356 *
357 * @destroy_list - request list (using wb_this_page) terminated by @old_head
358 * @old_head - the old head of the list
359 *
360 * All subrequests must be locked and removed from all lists, so at this point
361 * they are only "active" in this function, and possibly in nfs_wait_on_request
362 * with a reference held by some other context.
363 */
364static void
365nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
366 struct nfs_page *old_head)
367{
368 while (destroy_list) {
369 struct nfs_page *subreq = destroy_list;
370
371 destroy_list = (subreq->wb_this_page == old_head) ?
372 NULL : subreq->wb_this_page;
373
374 WARN_ON_ONCE(old_head != subreq->wb_head);
375
376 /* make sure old group is not used */
377 subreq->wb_head = subreq;
378 subreq->wb_this_page = subreq;
379
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400380 /* subreq is now totally disconnected from page group or any
381 * write / commit lists. last chance to wake any waiters */
382 nfs_unlock_request(subreq);
383
384 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
385 /* release ref on old head request */
386 nfs_release_request(old_head);
387
388 nfs_page_group_clear_bits(subreq);
389
390 /* release the PG_INODE_REF reference */
391 if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
392 nfs_release_request(subreq);
393 else
394 WARN_ON_ONCE(1);
395 } else {
396 WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
397 /* zombie requests have already released the last
398 * reference and were waiting on the rest of the
399 * group to complete. Since it's no longer part of a
400 * group, simply free the request */
401 nfs_page_group_clear_bits(subreq);
402 nfs_free_request(subreq);
403 }
404 }
405}
406
407/*
408 * nfs_lock_and_join_requests - join all subreqs to the head req and return
409 * a locked reference, cancelling any pending
410 * operations for this page.
411 *
412 * @page - the page used to lookup the "page group" of nfs_page structures
413 * @nonblock - if true, don't block waiting for request locks
414 *
415 * This function joins all sub requests to the head request by first
416 * locking all requests in the group, cancelling any pending operations
417 * and finally updating the head request to cover the whole range covered by
418 * the (former) group. All subrequests are removed from any write or commit
419 * lists, unlinked from the group and destroyed.
420 *
421 * Returns a locked, referenced pointer to the head request - which after
422 * this call is guaranteed to be the only request associated with the page.
423 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
424 * error was encountered.
425 */
426static struct nfs_page *
427nfs_lock_and_join_requests(struct page *page, bool nonblock)
428{
429 struct inode *inode = page_file_mapping(page)->host;
430 struct nfs_page *head, *subreq;
431 struct nfs_page *destroy_list = NULL;
432 unsigned int total_bytes;
433 int ret;
434
435try_again:
436 total_bytes = 0;
437
438 WARN_ON_ONCE(destroy_list);
439
440 spin_lock(&inode->i_lock);
441
442 /*
443 * A reference is taken only on the head request which acts as a
444 * reference to the whole page group - the group will not be destroyed
445 * until the head reference is released.
446 */
447 head = nfs_page_find_head_request_locked(NFS_I(inode), page);
448
449 if (!head) {
450 spin_unlock(&inode->i_lock);
451 return NULL;
452 }
453
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400454 /* holding inode lock, so always make a non-blocking call to try the
455 * page group lock */
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400456 ret = nfs_page_group_lock(head, true);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400457 if (ret < 0) {
458 spin_unlock(&inode->i_lock);
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400459
460 if (!nonblock && ret == -EAGAIN) {
461 nfs_page_group_lock_wait(head);
462 nfs_release_request(head);
463 goto try_again;
464 }
465
Weston Andros Adamson94970012014-08-08 11:00:56 -0400466 nfs_release_request(head);
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400467 return ERR_PTR(ret);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400468 }
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400469
470 /* lock each request in the page group */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400471 subreq = head;
472 do {
473 /*
474 * Subrequests are always contiguous, non overlapping
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400475 * and in order - but may be repeated (mirrored writes).
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400476 */
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400477 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
478 /* keep track of how many bytes this group covers */
479 total_bytes += subreq->wb_bytes;
480 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
481 ((subreq->wb_offset + subreq->wb_bytes) >
482 (head->wb_offset + total_bytes)))) {
483 nfs_page_group_unlock(head);
484 spin_unlock(&inode->i_lock);
485 return ERR_PTR(-EIO);
486 }
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400487
488 if (!nfs_lock_request(subreq)) {
489 /* releases page group bit lock and
490 * inode spin lock and all references */
491 ret = nfs_unroll_locks_and_wait(inode, head,
492 subreq, nonblock);
493
494 if (ret == 0)
495 goto try_again;
496
497 return ERR_PTR(ret);
498 }
499
500 subreq = subreq->wb_this_page;
501 } while (subreq != head);
502
503 /* Now that all requests are locked, make sure they aren't on any list.
504 * Commit list removal accounting is done after locks are dropped */
505 subreq = head;
506 do {
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400507 nfs_clear_request_commit(subreq);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400508 subreq = subreq->wb_this_page;
509 } while (subreq != head);
510
511 /* unlink subrequests from head, destroy them later */
512 if (head->wb_this_page != head) {
513 /* destroy list will be terminated by head */
514 destroy_list = head->wb_this_page;
515 head->wb_this_page = head;
516
517 /* change head request to cover whole range that
518 * the former page group covered */
519 head->wb_bytes = total_bytes;
520 }
521
522 /*
523 * prepare head request to be added to new pgio descriptor
524 */
525 nfs_page_group_clear_bits(head);
526
527 /*
528 * some part of the group was still on the inode list - otherwise
529 * the group wouldn't be involved in async write.
530 * grab a reference for the head request, iff it needs one.
531 */
532 if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
533 kref_get(&head->wb_kref);
534
535 nfs_page_group_unlock(head);
536
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400537 /* drop lock to clean uprequests on destroy list */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400538 spin_unlock(&inode->i_lock);
539
540 nfs_destroy_unlinked_subrequests(destroy_list, head);
541
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400542 /* still holds ref on head from nfs_page_find_head_request_locked
543 * and still has lock on head from lock loop */
544 return head;
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400545}
546
Peng Tao0bcbf032015-12-05 15:57:31 +0800547static void nfs_write_error_remove_page(struct nfs_page *req)
548{
549 nfs_unlock_request(req);
550 nfs_end_page_writeback(req);
Peng Tao0bcbf032015-12-05 15:57:31 +0800551 generic_error_remove_page(page_file_mapping(req->wb_page),
552 req->wb_page);
Fred Isamana8c35e52017-04-14 14:24:28 -0400553 nfs_release_request(req);
Peng Tao0bcbf032015-12-05 15:57:31 +0800554}
555
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400556/*
557 * Find an associated nfs write request, and prepare to flush it out
558 * May return an error if the user signalled nfs_wait_on_request().
559 */
560static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Peng Taod6c843b2015-12-05 16:20:43 +0800561 struct page *page, bool nonblock,
562 bool launder)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400563{
564 struct nfs_page *req;
565 int ret = 0;
566
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400567 req = nfs_lock_and_join_requests(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400568 if (!req)
569 goto out;
570 ret = PTR_ERR(req);
571 if (IS_ERR(req))
572 goto out;
573
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400574 nfs_set_page_writeback(page);
575 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400576
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400577 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400578 if (!nfs_pageio_add_request(pgio, req)) {
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400579 ret = pgio->pg_error;
Peng Tao0bcbf032015-12-05 15:57:31 +0800580 /*
Peng Taod6c843b2015-12-05 16:20:43 +0800581 * Remove the problematic req upon fatal errors
582 * in launder case, while other dirty pages can
583 * still be around until they get flushed.
Peng Tao0bcbf032015-12-05 15:57:31 +0800584 */
585 if (nfs_error_is_fatal(ret)) {
586 nfs_context_set_write_error(req->wb_context, ret);
Peng Taod6c843b2015-12-05 16:20:43 +0800587 if (launder) {
588 nfs_write_error_remove_page(req);
589 goto out;
590 }
Peng Tao0bcbf032015-12-05 15:57:31 +0800591 }
Peng Taod6c843b2015-12-05 16:20:43 +0800592 nfs_redirty_request(req);
593 ret = -EAGAIN;
Trond Myklebust40f90272015-10-01 11:36:38 -0400594 } else
595 nfs_add_stats(page_file_mapping(page)->host,
596 NFSIOS_WRITEPAGES, 1);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400597out:
598 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500599}
600
Peng Taod6c843b2015-12-05 16:20:43 +0800601static int nfs_do_writepage(struct page *page, struct writeback_control *wbc,
602 struct nfs_pageio_descriptor *pgio, bool launder)
Trond Myklebustf758c8852007-07-22 19:27:32 -0400603{
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400604 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400605
Huang Ying8cd79782016-10-07 17:00:24 -0700606 nfs_pageio_cond_complete(pgio, page_index(page));
Peng Taod6c843b2015-12-05 16:20:43 +0800607 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE,
608 launder);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400609 if (ret == -EAGAIN) {
610 redirty_page_for_writepage(wbc, page);
611 ret = 0;
612 }
613 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400614}
615
Trond Myklebuste261f512006-12-05 00:35:41 -0500616/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 * Write an mmapped page to the server.
618 */
Peng Taod6c843b2015-12-05 16:20:43 +0800619static int nfs_writepage_locked(struct page *page,
620 struct writeback_control *wbc,
621 bool launder)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400623 struct nfs_pageio_descriptor pgio;
Trond Myklebust40f90272015-10-01 11:36:38 -0400624 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500625 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Trond Myklebust40f90272015-10-01 11:36:38 -0400627 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
Trond Myklebust811ed922016-06-01 18:25:56 -0400628 nfs_pageio_init_write(&pgio, inode, 0,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200629 false, &nfs_async_write_completion_ops);
Peng Taod6c843b2015-12-05 16:20:43 +0800630 err = nfs_do_writepage(page, wbc, &pgio, launder);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400631 nfs_pageio_complete(&pgio);
632 if (err < 0)
633 return err;
634 if (pgio.pg_error < 0)
635 return pgio.pg_error;
636 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500637}
638
639int nfs_writepage(struct page *page, struct writeback_control *wbc)
640{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400641 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500642
Peng Taod6c843b2015-12-05 16:20:43 +0800643 ret = nfs_writepage_locked(page, wbc, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400645 return ret;
646}
647
648static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
649{
650 int ret;
651
Peng Taod6c843b2015-12-05 16:20:43 +0800652 ret = nfs_do_writepage(page, wbc, data, false);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400653 unlock_page(page);
654 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655}
656
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
658{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 struct inode *inode = mapping->host;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400660 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 int err;
662
Chuck Lever91d5b472006-03-20 13:44:14 -0500663 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
664
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200665 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
666 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400667 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400668 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400669
Trond Myklebustf758c8852007-07-22 19:27:32 -0400670 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400671 goto out_err;
672 err = pgio.pg_error;
673 if (err < 0)
674 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400675 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400676out_err:
677 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
680/*
681 * Insert a write request into an inode
682 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500683static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684{
685 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400686
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400687 WARN_ON_ONCE(req->wb_this_page != req);
688
Trond Myklebuste7d39062008-06-13 12:12:32 -0400689 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400690 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400691
692 spin_lock(&inode->i_lock);
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500693 if (!nfsi->nrequests &&
694 NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700695 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700696 /*
697 * Swap-space should not get truncated. Hence no need to plug the race
698 * with invalidate/truncate.
699 */
700 if (likely(!PageSwapCache(req->wb_page))) {
701 set_bit(PG_MAPPED, &req->wb_flags);
702 SetPagePrivate(req->wb_page);
703 set_page_private(req->wb_page, (unsigned long)req);
704 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500705 nfsi->nrequests++;
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400706 /* this a head request for a page group - mark it as having an
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500707 * extra reference so sub groups can follow suit.
708 * This flag also informs pgio layer when to bump nrequests when
709 * adding subrequests. */
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400710 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
Trond Myklebustc03b4022007-06-17 13:26:38 -0400711 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400712 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713}
714
715/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800716 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 */
718static void nfs_inode_remove_request(struct nfs_page *req)
719{
David Howells2b0143b2015-03-17 22:25:59 +0000720 struct inode *inode = d_inode(req->wb_context->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400722 struct nfs_page *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400724 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
725 head = req->wb_head;
726
727 spin_lock(&inode->i_lock);
Anna Schumaker67911c82016-01-19 14:37:30 -0500728 if (likely(head->wb_page && !PageSwapCache(head->wb_page))) {
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400729 set_page_private(head->wb_page, 0);
730 ClearPagePrivate(head->wb_page);
NeilBrown95905442014-09-24 11:28:32 +1000731 smp_mb__after_atomic();
732 wake_up_page(head->wb_page, PG_private);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400733 clear_bit(PG_MAPPED, &head->wb_flags);
734 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500735 nfsi->nrequests--;
736 spin_unlock(&inode->i_lock);
737 } else {
738 spin_lock(&inode->i_lock);
739 nfsi->nrequests--;
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400740 spin_unlock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700741 }
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400742
743 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
744 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
746
Trond Myklebust61822ab2006-12-05 00:35:42 -0500747static void
Fred6d884e82008-03-19 11:24:38 -0400748nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500749{
Anna Schumaker67911c82016-01-19 14:37:30 -0500750 if (req->wb_page)
751 __set_page_dirty_nobuffers(req->wb_page);
Trond Myklebust61822ab2006-12-05 00:35:42 -0500752}
753
Trond Myklebust3a3908c2014-09-08 22:21:00 -0700754/*
755 * nfs_page_search_commits_for_head_request_locked
756 *
757 * Search through commit lists on @inode for the head request for @page.
758 * Must be called while holding the inode (which is cinfo) lock.
759 *
760 * Returns the head request if found, or NULL if not found.
761 */
762static struct nfs_page *
763nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
764 struct page *page)
765{
766 struct nfs_page *freq, *t;
767 struct nfs_commit_info cinfo;
768 struct inode *inode = &nfsi->vfs_inode;
769
770 nfs_init_cinfo_from_inode(&cinfo, inode);
771
772 /* search through pnfs commit lists */
773 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
774 if (freq)
775 return freq->wb_head;
776
777 /* Linearly search the commit list for the correct request */
778 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
779 if (freq->wb_page == page)
780 return freq->wb_head;
781 }
782
783 return NULL;
784}
785
Trond Myklebust8dd37752012-03-15 17:16:40 -0400786/**
Trond Myklebust86d80f92015-07-31 16:24:30 -0400787 * nfs_request_add_commit_list_locked - add request to a commit list
788 * @req: pointer to a struct nfs_page
789 * @dst: commit list head
790 * @cinfo: holds list lock and accounting info
791 *
792 * This sets the PG_CLEAN bit, updates the cinfo count of
793 * number of outstanding requests requiring a commit as well as
794 * the MM page stats.
795 *
Dave Wysochanskife238e62016-04-01 13:45:09 -0400796 * The caller must hold cinfo->inode->i_lock, and the nfs_page lock.
Trond Myklebust86d80f92015-07-31 16:24:30 -0400797 */
798void
799nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
800 struct nfs_commit_info *cinfo)
801{
802 set_bit(PG_CLEAN, &req->wb_flags);
803 nfs_list_add_request(req, dst);
804 cinfo->mds->ncommit++;
805}
806EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
807
808/**
Trond Myklebust8dd37752012-03-15 17:16:40 -0400809 * nfs_request_add_commit_list - add request to a commit list
810 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400811 * @dst: commit list head
812 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400813 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400814 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400815 * number of outstanding requests requiring a commit as well as
816 * the MM page stats.
817 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400818 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400819 * holding the nfs_page lock.
820 */
821void
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500822nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400823{
Dave Wysochanskife238e62016-04-01 13:45:09 -0400824 spin_lock(&cinfo->inode->i_lock);
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500825 nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo);
Dave Wysochanskife238e62016-04-01 13:45:09 -0400826 spin_unlock(&cinfo->inode->i_lock);
Anna Schumaker67911c82016-01-19 14:37:30 -0500827 if (req->wb_page)
828 nfs_mark_page_unstable(req->wb_page, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400829}
830EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
831
832/**
833 * nfs_request_remove_commit_list - Remove request from a commit list
834 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400835 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400836 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400837 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400838 * number of outstanding requests requiring a commit
839 * It does not update the MM page stats.
840 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400841 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400842 */
843void
Fred Isamanea2cf222012-04-20 14:47:53 -0400844nfs_request_remove_commit_list(struct nfs_page *req,
845 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400846{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400847 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
848 return;
849 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400850 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400851}
852EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
853
Fred Isamanea2cf222012-04-20 14:47:53 -0400854static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
855 struct inode *inode)
856{
Dave Wysochanskife238e62016-04-01 13:45:09 -0400857 cinfo->inode = inode;
Fred Isamanea2cf222012-04-20 14:47:53 -0400858 cinfo->mds = &NFS_I(inode)->commit_info;
859 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400860 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400861 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400862}
863
864void nfs_init_cinfo(struct nfs_commit_info *cinfo,
865 struct inode *inode,
866 struct nfs_direct_req *dreq)
867{
Fred Isaman1763da12012-04-20 14:47:57 -0400868 if (dreq)
869 nfs_init_cinfo_from_dreq(cinfo, dreq);
870 else
871 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400872}
873EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400874
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875/*
876 * Add a request to the inode's commit list.
877 */
Fred Isaman1763da12012-04-20 14:47:57 -0400878void
Fred Isamanea2cf222012-04-20 14:47:53 -0400879nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400880 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881{
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400882 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400883 return;
Anna Schumaker6272dcc2016-01-15 16:54:15 -0500884 nfs_request_add_commit_list(req, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400886
Fred Isamand6d6dc72012-03-08 17:29:35 -0500887static void
888nfs_clear_page_commit(struct page *page)
889{
Mel Gorman11fb9982016-07-28 15:46:20 -0700890 dec_node_page_state(page, NR_UNSTABLE_NFS);
Tejun Heo93f78d82015-05-22 17:13:27 -0400891 dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
892 WB_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500893}
894
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400895/* Called holding inode (/cinfo) lock */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400896static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400897nfs_clear_request_commit(struct nfs_page *req)
898{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400899 if (test_bit(PG_CLEAN, &req->wb_flags)) {
David Howells2b0143b2015-03-17 22:25:59 +0000900 struct inode *inode = d_inode(req->wb_context->dentry);
Fred Isamanea2cf222012-04-20 14:47:53 -0400901 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400902
Fred Isamanea2cf222012-04-20 14:47:53 -0400903 nfs_init_cinfo_from_inode(&cinfo, inode);
904 if (!pnfs_clear_request_commit(req, &cinfo)) {
Fred Isamanea2cf222012-04-20 14:47:53 -0400905 nfs_request_remove_commit_list(req, &cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400906 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500907 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400908 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400909}
910
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400911int nfs_write_need_commit(struct nfs_pgio_header *hdr)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400912{
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400913 if (hdr->verf.committed == NFS_DATA_SYNC)
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400914 return hdr->lseg == NULL;
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400915 return hdr->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400916}
917
Fred Isaman061ae2e2012-04-20 14:47:48 -0400918static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400919{
Fred Isamanea2cf222012-04-20 14:47:53 -0400920 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400921 unsigned long bytes = 0;
922
923 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
924 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400925 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400926 while (!list_empty(&hdr->pages)) {
927 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400928
929 bytes += req->wb_bytes;
930 nfs_list_remove_request(req);
931 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
932 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400933 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400934 nfs_context_set_write_error(req->wb_context, hdr->error);
935 goto remove_req;
936 }
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400937 if (nfs_write_need_commit(hdr)) {
Anna Schumakerf79d06f2014-05-06 09:12:28 -0400938 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400939 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -0400940 hdr->pgio_mirror_idx);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400941 goto next;
942 }
943remove_req:
944 nfs_inode_remove_request(req);
945next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400946 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400947 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400948 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400949 }
950out:
951 hdr->release(hdr);
952}
953
Anna Schumakerce595152014-05-06 09:12:34 -0400954unsigned long
Fred Isamanea2cf222012-04-20 14:47:53 -0400955nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400956{
Fred Isamanea2cf222012-04-20 14:47:53 -0400957 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400958}
959
Dave Wysochanskife238e62016-04-01 13:45:09 -0400960/* cinfo->inode->i_lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400961int
Fred Isamanea2cf222012-04-20 14:47:53 -0400962nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
963 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500964{
965 struct nfs_page *req, *tmp;
966 int ret = 0;
967
968 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400969 if (!nfs_lock_request(req))
970 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400971 kref_get(&req->wb_kref);
Dave Wysochanskife238e62016-04-01 13:45:09 -0400972 if (cond_resched_lock(&cinfo->inode->i_lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400973 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400974 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400975 nfs_list_add_request(req, dst);
976 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400977 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400978 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500979 }
980 return ret;
981}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983/*
984 * nfs_scan_commit - Scan an inode for commit requests
985 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400986 * @dst: mds destination list
987 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 *
989 * Moves requests from the inode's 'commit' request list.
990 * The requests are *not* checked to ensure that they form a contiguous set.
991 */
Fred Isaman1763da12012-04-20 14:47:57 -0400992int
Fred Isamanea2cf222012-04-20 14:47:53 -0400993nfs_scan_commit(struct inode *inode, struct list_head *dst,
994 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995{
Fred Isamand6d6dc72012-03-08 17:29:35 -0500996 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400997
Dave Wysochanskife238e62016-04-01 13:45:09 -0400998 spin_lock(&cinfo->inode->i_lock);
Fred Isamanea2cf222012-04-20 14:47:53 -0400999 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001000 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001001
Fred Isamanea2cf222012-04-20 14:47:53 -04001002 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1003 cinfo, max);
1004 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001005 }
Dave Wysochanskife238e62016-04-01 13:45:09 -04001006 spin_unlock(&cinfo->inode->i_lock);
Trond Myklebustff778d02010-02-19 16:53:39 -08001007 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008}
Fred Isamand6d6dc72012-03-08 17:29:35 -05001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010/*
Trond Myklebuste7d39062008-06-13 12:12:32 -04001011 * Search for an existing write request, and attempt to update
1012 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 *
Trond Myklebuste7d39062008-06-13 12:12:32 -04001014 * If the attempt fails, then the existing request is flushed out
1015 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 */
Trond Myklebuste7d39062008-06-13 12:12:32 -04001017static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1018 struct page *page,
1019 unsigned int offset,
1020 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021{
Trond Myklebuste7d39062008-06-13 12:12:32 -04001022 struct nfs_page *req;
1023 unsigned int rqend;
1024 unsigned int end;
1025 int error;
1026
1027 if (!PagePrivate(page))
1028 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001031 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 for (;;) {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001034 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001035 if (req == NULL)
1036 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -05001037
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001038 /* should be handled by nfs_flush_incompatible */
1039 WARN_ON_ONCE(req->wb_head != req);
1040 WARN_ON_ONCE(req->wb_this_page != req);
1041
Trond Myklebuste7d39062008-06-13 12:12:32 -04001042 rqend = req->wb_offset + req->wb_bytes;
1043 /*
1044 * Tell the caller to flush out the request if
1045 * the offsets are non-contiguous.
1046 * Note: nfs_flush_incompatible() will already
1047 * have flushed out requests having wrong owners.
1048 */
Trond Myklebuste468bae2008-06-13 13:25:22 -04001049 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -04001050 || end < req->wb_offset)
1051 goto out_flushme;
1052
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001053 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Trond Myklebuste7d39062008-06-13 12:12:32 -04001056 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -04001057 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001058 error = nfs_wait_on_request(req);
1059 nfs_release_request(req);
1060 if (error != 0)
1061 goto out_err;
1062 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 }
1064
1065 /* Okay, the request matches. Update the region */
1066 if (offset < req->wb_offset) {
1067 req->wb_offset = offset;
1068 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 if (end > rqend)
1071 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001072 else
1073 req->wb_bytes = rqend - req->wb_offset;
1074out_unlock:
Fred Isamanca138f32012-04-05 15:26:36 -04001075 if (req)
1076 nfs_clear_request_commit(req);
Weston Andros Adamson411a99a2014-07-17 20:42:19 -04001077 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001078 return req;
1079out_flushme:
1080 spin_unlock(&inode->i_lock);
1081 nfs_release_request(req);
1082 error = nfs_wb_page(inode, page);
1083out_err:
1084 return ERR_PTR(error);
1085}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Trond Myklebuste7d39062008-06-13 12:12:32 -04001087/*
1088 * Try to update an existing write request, or create one if there is none.
1089 *
1090 * Note: Should always be called with the Page Lock held to prevent races
1091 * if we have to add a new request. Also assumes that the caller has
1092 * already called nfs_flush_incompatible() if necessary.
1093 */
1094static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1095 struct page *page, unsigned int offset, unsigned int bytes)
1096{
Mel Gormand56b4dd2012-07-31 16:45:06 -07001097 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001098 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001099
1100 req = nfs_try_to_update_request(inode, page, offset, bytes);
1101 if (req != NULL)
1102 goto out;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001103 req = nfs_create_request(ctx, page, NULL, offset, bytes);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001104 if (IS_ERR(req))
1105 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001106 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -04001107out:
Trond Myklebust61e930a2007-10-18 17:08:05 -04001108 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109}
1110
Trond Myklebuste7d39062008-06-13 12:12:32 -04001111static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1112 unsigned int offset, unsigned int count)
1113{
1114 struct nfs_page *req;
1115
1116 req = nfs_setup_write_request(ctx, page, offset, count);
1117 if (IS_ERR(req))
1118 return PTR_ERR(req);
1119 /* Update file length */
1120 nfs_grow_file(page, offset, count);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -04001121 nfs_mark_uptodate(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001122 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001123 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001124 return 0;
1125}
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127int nfs_flush_incompatible(struct file *file, struct page *page)
1128{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001129 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -04001130 struct nfs_lock_context *l_ctx;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001131 struct file_lock_context *flctx = file_inode(file)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -05001133 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 /*
1135 * Look for a request corresponding to this page. If there
1136 * is one, and it belongs to another file, we flush it out
1137 * before we try to copy anything into the page. Do this
1138 * due to the lack of an ACCESS-type call in NFSv2.
1139 * Also do the same if we find a request from an existing
1140 * dropped page.
1141 */
Trond Myklebust1a545332006-12-05 00:35:40 -05001142 do {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001143 req = nfs_page_find_head_request(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001144 if (req == NULL)
1145 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -04001146 l_ctx = req->wb_lock_context;
Trond Myklebust138a2932015-10-01 17:17:06 -04001147 do_flush = req->wb_page != page ||
1148 !nfs_match_open_context(req->wb_context, ctx);
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001149 /* for now, flush if more than 1 request in page_group */
1150 do_flush |= req->wb_this_page != req;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001151 if (l_ctx && flctx &&
1152 !(list_empty_careful(&flctx->flc_posix) &&
1153 list_empty_careful(&flctx->flc_flock))) {
Jeff Layton5263e312015-01-16 15:05:55 -05001154 do_flush |= l_ctx->lockowner.l_owner != current->files
1155 || l_ctx->lockowner.l_pid != current->tgid;
1156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -05001158 if (!do_flush)
1159 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -07001160 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001161 } while (status == 0);
1162 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163}
1164
1165/*
Andy Adamsondc248262013-08-14 11:59:16 -04001166 * Avoid buffered writes when a open context credential's key would
1167 * expire soon.
1168 *
1169 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1170 *
1171 * Return 0 and set a credential flag which triggers the inode to flush
1172 * and performs NFS_FILE_SYNC writes if the key will expired within
1173 * RPC_KEY_EXPIRE_TIMEO.
1174 */
1175int
1176nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1177{
1178 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1179 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1180
1181 return rpcauth_key_timeout_notify(auth, ctx->cred);
1182}
1183
1184/*
1185 * Test if the open context credential key is marked to expire soon.
1186 */
Scott Mayhewce52914e2016-06-07 15:14:48 -04001187bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode)
Andy Adamsondc248262013-08-14 11:59:16 -04001188{
Scott Mayhewce52914e2016-06-07 15:14:48 -04001189 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1190
1191 return rpcauth_cred_key_to_expire(auth, ctx->cred);
Andy Adamsondc248262013-08-14 11:59:16 -04001192}
1193
1194/*
Trond Myklebust5d47a352008-02-07 17:24:07 -05001195 * If the page cache is marked as unsafe or invalid, then we can't rely on
1196 * the PageUptodate() flag. In this case, we will need to turn off
1197 * write optimisations that depend on the page contents being correct.
1198 */
Trond Myklebust8d197a52012-04-29 12:50:01 -04001199static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -05001200{
Jeff Laytond529ef82014-01-27 13:46:15 -05001201 struct nfs_inode *nfsi = NFS_I(inode);
1202
Trond Myklebust8d197a52012-04-29 12:50:01 -04001203 if (nfs_have_delegated_attributes(inode))
1204 goto out;
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001205 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
Jeff Laytond529ef82014-01-27 13:46:15 -05001206 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -05001207 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -05001208 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -04001209 return false;
1210out:
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001211 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1212 return false;
Trond Myklebust8d197a52012-04-29 12:50:01 -04001213 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -05001214}
1215
Jeff Layton5263e312015-01-16 15:05:55 -05001216static bool
1217is_whole_file_wrlock(struct file_lock *fl)
1218{
1219 return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1220 fl->fl_type == F_WRLCK;
1221}
1222
Scott Mayhewc7559662013-07-05 17:33:19 -04001223/* If we know the page is up to date, and we're not using byte range locks (or
1224 * if we have the whole file locked for writing), it may be more efficient to
1225 * extend the write to cover the entire page in order to avoid fragmentation
1226 * inefficiencies.
1227 *
Scott Mayhew263b4502014-01-17 15:12:05 -05001228 * If the file is opened for synchronous writes then we can just skip the rest
1229 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -04001230 */
1231static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1232{
Jeff Layton5263e312015-01-16 15:05:55 -05001233 int ret;
1234 struct file_lock_context *flctx = inode->i_flctx;
1235 struct file_lock *fl;
1236
Scott Mayhewc7559662013-07-05 17:33:19 -04001237 if (file->f_flags & O_DSYNC)
1238 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -05001239 if (!nfs_write_pageuptodate(page, inode))
1240 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -04001241 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1242 return 1;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001243 if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1244 list_empty_careful(&flctx->flc_posix)))
Trond Myklebust8fa45922015-10-01 18:38:27 -04001245 return 1;
Jeff Layton5263e312015-01-16 15:05:55 -05001246
1247 /* Check to see if there are whole file write locks */
Jeff Layton5263e312015-01-16 15:05:55 -05001248 ret = 0;
Jeff Layton6109c852015-01-16 15:05:57 -05001249 spin_lock(&flctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001250 if (!list_empty(&flctx->flc_posix)) {
1251 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1252 fl_list);
1253 if (is_whole_file_wrlock(fl))
1254 ret = 1;
1255 } else if (!list_empty(&flctx->flc_flock)) {
Jeff Layton5263e312015-01-16 15:05:55 -05001256 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1257 fl_list);
1258 if (fl->fl_type == F_WRLCK)
1259 ret = 1;
1260 }
Jeff Layton6109c852015-01-16 15:05:57 -05001261 spin_unlock(&flctx->flc_lock);
Jeff Layton5263e312015-01-16 15:05:55 -05001262 return ret;
Scott Mayhewc7559662013-07-05 17:33:19 -04001263}
1264
Trond Myklebust5d47a352008-02-07 17:24:07 -05001265/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 * Update and possibly write a cached page of an NFS file.
1267 *
1268 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1269 * things with a page scheduled for an RPC call (e.g. invalidate it).
1270 */
1271int nfs_updatepage(struct file *file, struct page *page,
1272 unsigned int offset, unsigned int count)
1273{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001274 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001275 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 int status = 0;
1277
Chuck Lever91d5b472006-03-20 13:44:14 -05001278 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1279
Al Viro6de14722013-09-16 10:53:17 -04001280 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1281 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
Benjamin Coddington149a4fd2016-07-18 10:41:57 -04001283 if (!count)
1284 goto out;
1285
Scott Mayhewc7559662013-07-05 17:33:19 -04001286 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -05001287 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
1290
Trond Myklebuste21195a2006-12-05 00:35:39 -05001291 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -04001292 if (status < 0)
1293 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -07001294 else
1295 __set_page_dirty_nobuffers(page);
Benjamin Coddington149a4fd2016-07-18 10:41:57 -04001296out:
Chuck Lever48186c72008-06-11 17:56:05 -04001297 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 return status;
1300}
1301
Trond Myklebust3ff75762007-07-14 15:40:00 -04001302static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303{
1304 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1305 case FLUSH_HIGHPRI:
1306 return RPC_PRIORITY_HIGH;
1307 case FLUSH_LOWPRI:
1308 return RPC_PRIORITY_LOW;
1309 }
1310 return RPC_PRIORITY_NORMAL;
1311}
1312
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001313static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1314 struct rpc_message *msg,
Tom Haynesabde71f2014-06-09 13:12:20 -07001315 const struct nfs_rpc_ops *rpc_ops,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001316 struct rpc_task_setup *task_setup_data, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317{
Trond Myklebust3ff75762007-07-14 15:40:00 -04001318 int priority = flush_task_priority(how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001320 task_setup_data->priority = priority;
Tom Haynesabde71f2014-06-09 13:12:20 -07001321 rpc_ops->write_setup(hdr, msg);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001322
Tom Haynesabde71f2014-06-09 13:12:20 -07001323 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001324 &task_setup_data->rpc_client, msg, hdr);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001325}
1326
Fred6d884e82008-03-19 11:24:38 -04001327/* If a nfs_flush_* function fails, it should remove reqs from @head and
1328 * call this on each, which will prepare them to be retried on next
1329 * writeback using standard nfs.
1330 */
1331static void nfs_redirty_request(struct nfs_page *req)
1332{
1333 nfs_mark_request_dirty(req);
Trond Myklebustc7070112015-06-17 19:56:22 -04001334 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001335 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -04001336 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001337 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001338}
1339
Fred Isaman061ae2e2012-04-20 14:47:48 -04001340static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001341{
1342 struct nfs_page *req;
1343
1344 while (!list_empty(head)) {
1345 req = nfs_list_entry(head->next);
1346 nfs_list_remove_request(req);
1347 nfs_redirty_request(req);
1348 }
1349}
1350
Trond Myklebustdc602dd2015-12-31 11:44:06 -05001351static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
1352{
1353 nfs_async_write_error(&hdr->pages);
1354}
1355
Fred Isaman061ae2e2012-04-20 14:47:48 -04001356static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1357 .error_cleanup = nfs_async_write_error,
1358 .completion = nfs_write_completion,
Trond Myklebustdc602dd2015-12-31 11:44:06 -05001359 .reschedule_io = nfs_async_write_reschedule_io,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001360};
1361
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001362void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001363 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001364 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001365{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001366 struct nfs_server *server = NFS_SERVER(inode);
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001367 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001368
1369#ifdef CONFIG_NFS_V4_1
1370 if (server->pnfs_curr_ld && !force_mds)
1371 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1372#endif
Anna Schumaker4a0de552014-05-06 09:12:30 -04001373 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1374 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001375}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001376EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001377
Trond Myklebustdce81292011-07-13 15:59:19 -04001378void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1379{
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001380 struct nfs_pgio_mirror *mirror;
1381
Kinglong Mee6f29b9b2015-09-20 23:03:28 +08001382 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1383 pgio->pg_ops->pg_cleanup(pgio);
1384
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001385 pgio->pg_ops = &nfs_pgio_rw_ops;
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001386
1387 nfs_pageio_stop_mirroring(pgio);
1388
1389 mirror = &pgio->pg_mirrors[0];
1390 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
Trond Myklebustdce81292011-07-13 15:59:19 -04001391}
Trond Myklebust1f945352011-07-13 15:59:57 -04001392EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
Fred Isaman0b7c0152012-04-20 14:47:39 -04001395void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1396{
1397 struct nfs_commit_data *data = calldata;
1398
1399 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1400}
1401
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001402/*
1403 * Special version of should_remove_suid() that ignores capabilities.
1404 */
1405static int nfs_should_remove_suid(const struct inode *inode)
1406{
1407 umode_t mode = inode->i_mode;
1408 int kill = 0;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001409
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001410 /* suid always must be killed */
1411 if (unlikely(mode & S_ISUID))
1412 kill = ATTR_KILL_SUID;
1413
1414 /*
1415 * sgid without any exec bits is just a mandatory locking mark; leave
1416 * it alone. If some exec bits are set, it's a real sgid; kill it.
1417 */
1418 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1419 kill |= ATTR_KILL_SGID;
1420
1421 if (unlikely(kill && S_ISREG(mode)))
1422 return kill;
1423
1424 return 0;
1425}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001426
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001427static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1428 struct nfs_fattr *fattr)
1429{
1430 struct nfs_pgio_args *argp = &hdr->args;
1431 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001432 u64 size = argp->offset + resp->count;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001433
1434 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001435 fattr->size = size;
1436 if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1437 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001438 return;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001439 }
1440 if (size != fattr->size)
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001441 return;
1442 /* Set attribute barrier */
1443 nfs_fattr_set_barrier(fattr);
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001444 /* ...and update size */
1445 fattr->valid |= NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001446}
1447
1448void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1449{
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001450 struct nfs_fattr *fattr = &hdr->fattr;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001451 struct inode *inode = hdr->inode;
1452
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001453 spin_lock(&inode->i_lock);
1454 nfs_writeback_check_extend(hdr, fattr);
1455 nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1456 spin_unlock(&inode->i_lock);
1457}
1458EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1459
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460/*
1461 * This function is called when the WRITE call is complete.
1462 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001463static int nfs_writeback_done(struct rpc_task *task,
1464 struct nfs_pgio_header *hdr,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001465 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001467 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Chuck Leverf551e442006-09-20 14:33:04 -04001469 /*
1470 * ->write_done will attempt to use post-op attributes to detect
1471 * conflicting writes by other clients. A strict interpretation
1472 * of close-to-open would allow us to continue caching even if
1473 * another writer had changed the file, but some applications
1474 * depend on tighter cache coherency when writing.
1475 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001476 status = NFS_PROTO(inode)->write_done(task, hdr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001477 if (status != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001478 return status;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001479 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001480
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001481 if (hdr->res.verf->committed < hdr->args.stable &&
1482 task->tk_status >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 /* We tried a write call, but the server did not
1484 * commit data to stable storage even though we
1485 * requested it.
1486 * Note: There is a known bug in Tru64 < 5.0 in which
1487 * the server reports NFS_DATA_SYNC, but performs
1488 * NFS_FILE_SYNC. We therefore implement this checking
1489 * as a dprintk() in order to avoid filling syslog.
1490 */
1491 static unsigned long complain;
1492
Fred Isamana69aef12011-03-03 15:13:47 +00001493 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001495 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001497 NFS_SERVER(inode)->nfs_client->cl_hostname,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001498 hdr->res.verf->committed, hdr->args.stable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 complain = jiffies + 300 * HZ;
1500 }
1501 }
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001502
1503 /* Deal with the suid/sgid bit corner case */
1504 if (nfs_should_remove_suid(inode))
1505 nfs_mark_for_revalidate(inode);
Anna Schumaker0eecb212014-05-06 09:12:32 -04001506 return 0;
1507}
1508
1509/*
1510 * This function is called when the WRITE call is complete.
1511 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001512static void nfs_writeback_result(struct rpc_task *task,
1513 struct nfs_pgio_header *hdr)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001514{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001515 struct nfs_pgio_args *argp = &hdr->args;
1516 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001517
1518 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 static unsigned long complain;
1520
Fred Isaman6c75dc02012-04-20 14:47:47 -04001521 /* This a short write! */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001522 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001523
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001525 if (resp->count == 0) {
1526 if (time_before(complain, jiffies)) {
1527 printk(KERN_WARNING
1528 "NFS: Server wrote zero bytes, expected %u.\n",
1529 argp->count);
1530 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001532 nfs_set_pgio_error(hdr, -EIO, argp->offset);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001533 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001534 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
Kinglong Meef8417b482015-10-16 17:23:29 +08001536
1537 /* For non rpc-based layout drivers, retry-through-MDS */
1538 if (!task->tk_ops) {
1539 hdr->pnfs_error = -EAGAIN;
1540 return;
1541 }
1542
Fred Isaman6c75dc02012-04-20 14:47:47 -04001543 /* Was this an NFSv2 write or an NFSv3 stable write? */
1544 if (resp->verf->committed != NFS_UNSTABLE) {
1545 /* Resend from where the server left off */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001546 hdr->mds_offset += resp->count;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001547 argp->offset += resp->count;
1548 argp->pgbase += resp->count;
1549 argp->count -= resp->count;
1550 } else {
1551 /* Resend as a stable write in order to avoid
1552 * headaches in the case of a server crash.
1553 */
1554 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001556 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558}
1559
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001560static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001561{
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001562 return wait_on_atomic_t(&cinfo->rpcs_out,
1563 nfs_wait_atomic_killable, TASK_KILLABLE);
1564}
1565
1566static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
1567{
1568 atomic_inc(&cinfo->rpcs_out);
1569}
1570
1571static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
1572{
1573 if (atomic_dec_and_test(&cinfo->rpcs_out))
1574 wake_up_atomic_t(&cinfo->rpcs_out);
Trond Myklebust71d0a612010-04-22 15:35:57 -04001575}
1576
Fred Isaman0b7c0152012-04-20 14:47:39 -04001577void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001579 put_nfs_open_context(data->context);
1580 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581}
Fred Isamane0c2b382011-03-23 13:27:53 +00001582EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Fred Isaman0b7c0152012-04-20 14:47:39 -04001584int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Peng Taoc36aae92014-06-09 07:10:14 +08001585 const struct nfs_rpc_ops *nfs_ops,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001586 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001587 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
Trond Myklebust07737692007-10-25 18:42:54 -04001589 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001590 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001591 struct rpc_message msg = {
1592 .rpc_argp = &data->args,
1593 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001594 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001595 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001596 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001597 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001598 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001599 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001600 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001601 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001602 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001603 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001604 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001605 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001606 /* Set up the initial task struct. */
Peng Taoc36aae92014-06-09 07:10:14 +08001607 nfs_ops->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
Kinglong Meeb4839eb2015-07-01 12:00:13 +08001609 dprintk("NFS: initiated commit call\n");
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001610
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001611 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1612 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1613
Trond Myklebust07737692007-10-25 18:42:54 -04001614 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001615 if (IS_ERR(task))
1616 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001617 if (how & FLUSH_SYNC)
1618 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001619 rpc_put_task(task);
1620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
Fred Isamane0c2b382011-03-23 13:27:53 +00001622EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Peng Tao378520b2014-08-07 10:15:02 +08001624static loff_t nfs_get_lwb(struct list_head *head)
1625{
1626 loff_t lwb = 0;
1627 struct nfs_page *req;
1628
1629 list_for_each_entry(req, head, wb_list)
1630 if (lwb < (req_offset(req) + req->wb_bytes))
1631 lwb = req_offset(req) + req->wb_bytes;
1632
1633 return lwb;
1634}
1635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001637 * Set up the argument/result storage required for the RPC call.
1638 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001639void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001640 struct list_head *head,
1641 struct pnfs_layout_segment *lseg,
1642 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001643{
1644 struct nfs_page *first = nfs_list_entry(head->next);
David Howells2b0143b2015-03-17 22:25:59 +00001645 struct inode *inode = d_inode(first->wb_context->dentry);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001646
1647 /* Set up the RPC argument and reply structs
1648 * NB: take care not to mess about with data->commit et al. */
1649
1650 list_splice_init(head, &data->pages);
1651
1652 data->inode = inode;
1653 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001654 data->lseg = lseg; /* reference transferred */
Peng Tao378520b2014-08-07 10:15:02 +08001655 /* only set lwb for pnfs commit */
1656 if (lseg)
1657 data->lwb = nfs_get_lwb(&data->pages);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001658 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001659 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001660 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001661
1662 data->args.fh = NFS_FH(data->inode);
1663 /* Note: we always request a commit of the entire inode */
1664 data->args.offset = 0;
1665 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001666 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001667 data->res.fattr = &data->fattr;
1668 data->res.verf = &data->verf;
1669 nfs_fattr_init(&data->fattr);
1670}
Fred Isamane0c2b382011-03-23 13:27:53 +00001671EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001672
Fred Isamane0c2b382011-03-23 13:27:53 +00001673void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001674 struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001675 struct nfs_commit_info *cinfo,
1676 u32 ds_commit_idx)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001677{
1678 struct nfs_page *req;
1679
1680 while (!list_empty(page_list)) {
1681 req = nfs_list_entry(page_list->next);
1682 nfs_list_remove_request(req);
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001683 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
Tom Haynes487b9b82015-02-13 13:19:54 -08001684 if (!cinfo->dreq)
1685 nfs_clear_page_commit(req->wb_page);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001686 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001687 }
1688}
Fred Isamane0c2b382011-03-23 13:27:53 +00001689EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001690
Trond Myklebustb20135d2015-12-31 09:28:06 -05001691static void
1692nfs_commit_resched_write(struct nfs_commit_info *cinfo,
1693 struct nfs_page *req)
1694{
1695 __set_page_dirty_nobuffers(req->wb_page);
1696}
1697
Fred Isaman9ace33c2011-03-23 13:27:45 +00001698/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 * Commit dirty pages
1700 */
1701static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001702nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1703 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001705 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706
Weston Andros Adamsonade8feb2016-05-25 10:07:23 -04001707 /* another commit raced with us */
1708 if (list_empty(head))
1709 return 0;
1710
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001711 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
1713 if (!data)
1714 goto out_bad;
1715
1716 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001717 nfs_init_commit(data, head, NULL, cinfo);
1718 atomic_inc(&cinfo->mds->rpcs_out);
Peng Taoc36aae92014-06-09 07:10:14 +08001719 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1720 data->mds_ops, how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 out_bad:
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001722 nfs_retry_commit(head, NULL, cinfo, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 return -ENOMEM;
1724}
1725
Anna Schumaker67911c82016-01-19 14:37:30 -05001726int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf)
1727{
1728 struct inode *inode = file_inode(file);
1729 struct nfs_open_context *open;
1730 struct nfs_commit_info cinfo;
1731 struct nfs_page *req;
1732 int ret;
1733
1734 open = get_nfs_open_context(nfs_file_open_context(file));
1735 req = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode));
Dan Carpenter2997bfd2016-05-23 13:21:01 +03001736 if (IS_ERR(req)) {
1737 ret = PTR_ERR(req);
Anna Schumaker67911c82016-01-19 14:37:30 -05001738 goto out_put;
1739 }
1740
1741 nfs_init_cinfo_from_inode(&cinfo, inode);
1742
1743 memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier));
1744 nfs_request_add_commit_list(req, &cinfo);
1745 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1746 if (ret > 0)
1747 ret = 0;
1748
1749 nfs_free_request(req);
1750out_put:
1751 put_nfs_open_context(open);
1752 return ret;
1753}
1754EXPORT_SYMBOL_GPL(nfs_commit_file);
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756/*
1757 * COMMIT call returned
1758 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001759static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001761 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
Chuck Levera3f565b2007-01-31 12:14:01 -05001763 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 task->tk_pid, task->tk_status);
1765
Trond Myklebust788e7a82006-03-20 13:44:27 -05001766 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001767 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001768}
1769
Fred Isamanf453a542012-04-20 14:47:54 -04001770static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001771{
Fred Isaman5917ce82011-03-23 13:27:48 +00001772 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001773 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001774 struct nfs_commit_info cinfo;
NeilBrown353db792014-09-24 11:28:32 +10001775 struct nfs_server *nfss;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 while (!list_empty(&data->pages)) {
1778 req = nfs_list_entry(data->pages.next);
1779 nfs_list_remove_request(req);
Anna Schumaker67911c82016-01-19 14:37:30 -05001780 if (req->wb_page)
1781 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782
Niels de Vos1e8968c2013-12-17 18:20:16 +01001783 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001784 req->wb_context->dentry->d_sb->s_id,
David Howells2b0143b2015-03-17 22:25:59 +00001785 (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 req->wb_bytes,
1787 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001788 if (status < 0) {
1789 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001791 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 goto next;
1793 }
1794
1795 /* Okay, COMMIT succeeded, apparently. Check the verifier
1796 * returned by the server against all stored verfs. */
Trond Myklebust8fc3c382016-06-01 21:32:24 -04001797 if (!nfs_write_verifier_cmp(&req->wb_verf, &data->verf.verifier)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 /* We have a match */
1799 nfs_inode_remove_request(req);
1800 dprintk(" OK\n");
1801 goto next;
1802 }
1803 /* We have a mismatch. Write the page again */
1804 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001805 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001806 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001808 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 }
NeilBrown353db792014-09-24 11:28:32 +10001810 nfss = NFS_SERVER(data->inode);
1811 if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1812 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1813
Fred Isamanf453a542012-04-20 14:47:54 -04001814 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001815 nfs_commit_end(cinfo.mds);
Fred Isaman5917ce82011-03-23 13:27:48 +00001816}
1817
1818static void nfs_commit_release(void *calldata)
1819{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001820 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001821
Fred Isamanf453a542012-04-20 14:47:54 -04001822 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001823 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001825
1826static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001827 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001828 .rpc_call_done = nfs_commit_done,
1829 .rpc_release = nfs_commit_release,
1830};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
Fred Isamanf453a542012-04-20 14:47:54 -04001832static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1833 .completion = nfs_commit_release_pages,
Trond Myklebustb20135d2015-12-31 09:28:06 -05001834 .resched_write = nfs_commit_resched_write,
Fred Isamanf453a542012-04-20 14:47:54 -04001835};
1836
Fred Isaman1763da12012-04-20 14:47:57 -04001837int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1838 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001839{
1840 int status;
1841
Fred Isamanea2cf222012-04-20 14:47:53 -04001842 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001843 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001844 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001845 return status;
1846}
1847
Trond Myklebustb608b282010-07-30 15:31:54 -04001848int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001851 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001852 int may_wait = how & FLUSH_SYNC;
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001853 int error = 0;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001854 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855
Fred Isamanea2cf222012-04-20 14:47:53 -04001856 nfs_init_cinfo_from_inode(&cinfo, inode);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001857 nfs_commit_begin(cinfo.mds);
Fred Isamanea2cf222012-04-20 14:47:53 -04001858 res = nfs_scan_commit(inode, &head, &cinfo);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001859 if (res)
Fred Isamanea2cf222012-04-20 14:47:53 -04001860 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001861 nfs_commit_end(cinfo.mds);
1862 if (error < 0)
1863 goto out_error;
1864 if (!may_wait)
1865 goto out_mark_dirty;
1866 error = wait_on_commit(cinfo.mds);
1867 if (error < 0)
1868 return error;
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001869 return res;
Trond Myklebustaf7cf052015-09-29 20:34:05 -04001870out_error:
1871 res = error;
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001872 /* Note: If we exit without ensuring that the commit is complete,
1873 * we must mark the inode as dirty. Otherwise, future calls to
1874 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1875 * that the data is on the disk.
1876 */
1877out_mark_dirty:
1878 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 return res;
1880}
Trond Myklebustb20135d2015-12-31 09:28:06 -05001881EXPORT_SYMBOL_GPL(nfs_commit_inode);
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001882
Anna Schumakerae09c312015-07-13 14:01:32 -04001883int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001884{
Trond Myklebust420e3642010-02-19 17:00:02 -08001885 struct nfs_inode *nfsi = NFS_I(inode);
1886 int flags = FLUSH_SYNC;
1887 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001888
Jeff Layton3236c3e2011-10-11 09:49:21 -04001889 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001890 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e2011-10-11 09:49:21 -04001891 return ret;
1892
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001893 if (wbc->sync_mode == WB_SYNC_NONE) {
1894 /* Don't commit yet if this is a non-blocking flush and there
1895 * are a lot of outstanding writes for this mapping.
1896 */
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -05001897 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001898 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001899
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001900 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001901 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001902 }
1903
Trond Myklebust420e3642010-02-19 17:00:02 -08001904 ret = nfs_commit_inode(inode, flags);
1905 if (ret >= 0) {
1906 if (wbc->sync_mode == WB_SYNC_NONE) {
1907 if (ret < wbc->nr_to_write)
1908 wbc->nr_to_write -= ret;
1909 else
1910 wbc->nr_to_write = 0;
1911 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001912 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001913 }
1914out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001915 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1916 return ret;
1917}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001918EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001919
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001920/*
Trond Myklebust837bb1d2016-06-25 18:12:03 -04001921 * Wrapper for filemap_write_and_wait_range()
1922 *
1923 * Needed for pNFS in order to ensure data becomes visible to the
1924 * client.
1925 */
1926int nfs_filemap_write_and_wait_range(struct address_space *mapping,
1927 loff_t lstart, loff_t lend)
1928{
1929 int ret;
1930
1931 ret = filemap_write_and_wait_range(mapping, lstart, lend);
1932 if (ret == 0)
1933 ret = pnfs_sync_inode(mapping->host, true);
1934 return ret;
1935}
1936EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range);
1937
1938/*
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001939 * flush the inode to disk.
1940 */
1941int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001942{
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001943 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001944
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001945 trace_nfs_writeback_inode_enter(inode);
1946
Trond Myklebust5bb89b42015-03-25 14:14:42 -04001947 ret = filemap_write_and_wait(inode->i_mapping);
Chuck Lever6b196872015-05-06 18:26:58 -04001948 if (ret)
1949 goto out;
1950 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1951 if (ret < 0)
1952 goto out;
1953 pnfs_sync_inode(inode, true);
1954 ret = 0;
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001955
Chuck Lever6b196872015-05-06 18:26:58 -04001956out:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001957 trace_nfs_writeback_inode_exit(inode, ret);
1958 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001959}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001960EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001961
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001962int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1963{
1964 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001965 int ret = 0;
1966
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001967 wait_on_page_writeback(page);
1968
1969 /* blocking call to cancel all requests and join to a single (head)
1970 * request */
1971 req = nfs_lock_and_join_requests(page, false);
1972
1973 if (IS_ERR(req)) {
1974 ret = PTR_ERR(req);
1975 } else if (req) {
1976 /* all requests from this page have been cancelled by
1977 * nfs_lock_and_join_requests, so just remove the head
1978 * request from the inode / page_private pointer and
1979 * release it */
1980 nfs_inode_remove_request(req);
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001981 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001982 }
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001983
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001984 return ret;
1985}
1986
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001987/*
1988 * Write back all requests on one page - we do this before reading it.
1989 */
Peng Taod6c843b2015-12-05 16:20:43 +08001990int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder)
Trond Myklebust1c759502006-10-09 16:18:38 -04001991{
Mel Gorman29418aa2012-07-31 16:45:10 -07001992 loff_t range_start = page_file_offset(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001993 loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001994 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001995 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001996 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001997 .range_start = range_start,
1998 .range_end = range_end,
1999 };
2000 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04002001
Trond Myklebustf4ce1292013-08-19 18:59:33 -04002002 trace_nfs_writeback_page_enter(inode);
2003
Trond Myklebust0522f6a2010-05-26 08:42:24 -04002004 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04002005 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04002006 if (clear_page_dirty_for_io(page)) {
Peng Taod6c843b2015-12-05 16:20:43 +08002007 ret = nfs_writepage_locked(page, &wbc, launder);
Trond Myklebust73e33022008-04-11 16:03:54 -04002008 if (ret < 0)
2009 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04002010 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08002011 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04002012 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04002013 if (!PagePrivate(page))
2014 break;
2015 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04002016 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04002017 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08002018 }
Trond Myklebust73e33022008-04-11 16:03:54 -04002019out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04002020 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05002021 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04002022}
2023
Trond Myklebust074cc1d2009-08-10 08:54:13 -04002024#ifdef CONFIG_MIGRATION
2025int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08002026 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04002027{
Jeff Layton2da95652011-10-12 10:57:42 -04002028 /*
2029 * If PagePrivate is set, then the page is currently associated with
2030 * an in-progress read or write request. Don't try to migrate it.
2031 *
2032 * FIXME: we could do this in principle, but we'll need a way to ensure
2033 * that we can safely release the inode reference while holding
2034 * the page lock.
2035 */
2036 if (PagePrivate(page))
2037 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04002038
David Howells8c209ce2012-12-05 13:34:49 +00002039 if (!nfs_fscache_release_page(page, GFP_KERNEL))
2040 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04002041
Mel Gormana6bc32b2012-01-12 17:19:43 -08002042 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04002043}
2044#endif
2045
David Howellsf7b422b2006-06-09 09:34:33 -04002046int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047{
2048 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04002049 sizeof(struct nfs_pgio_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09002051 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (nfs_wdata_cachep == NULL)
2053 return -ENOMEM;
2054
Matthew Dobson93d23412006-03-26 01:37:50 -08002055 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2056 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002058 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059
Fred Isaman0b7c0152012-04-20 14:47:39 -04002060 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2061 sizeof(struct nfs_commit_data),
2062 0, SLAB_HWCACHE_ALIGN,
2063 NULL);
2064 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002065 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04002066
Matthew Dobson93d23412006-03-26 01:37:50 -08002067 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08002068 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002070 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Peter Zijlstra89a09142007-03-16 13:38:26 -08002072 /*
2073 * NFS congestion size, scale with available memory.
2074 *
2075 * 64MB: 8192k
2076 * 128MB: 11585k
2077 * 256MB: 16384k
2078 * 512MB: 23170k
2079 * 1GB: 32768k
2080 * 2GB: 46340k
2081 * 4GB: 65536k
2082 * 8GB: 92681k
2083 * 16GB: 131072k
2084 *
2085 * This allows larger machines to have larger/more transfers.
2086 * Limit the default to 256M
2087 */
2088 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2089 if (nfs_congestion_kb > 256*1024)
2090 nfs_congestion_kb = 256*1024;
2091
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04002093
2094out_destroy_commit_cache:
2095 kmem_cache_destroy(nfs_cdata_cachep);
2096out_destroy_write_mempool:
2097 mempool_destroy(nfs_wdata_mempool);
2098out_destroy_write_cache:
2099 kmem_cache_destroy(nfs_wdata_cachep);
2100 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101}
2102
David Brownell266bee82006-06-27 12:59:15 -07002103void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
2105 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04002106 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002108 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109}
2110
Anna Schumaker4a0de552014-05-06 09:12:30 -04002111static const struct nfs_rw_ops nfs_rw_write_ops = {
Anna Schumakera4cdda52014-05-06 09:12:31 -04002112 .rw_mode = FMODE_WRITE,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002113 .rw_alloc_header = nfs_writehdr_alloc,
2114 .rw_free_header = nfs_writehdr_free,
Anna Schumaker0eecb212014-05-06 09:12:32 -04002115 .rw_done = nfs_writeback_done,
2116 .rw_result = nfs_writeback_result,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04002117 .rw_initiate = nfs_initiate_write,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002118};