blob: 7b93164069307346ea43f130e3883959ae6576b6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/write.c
3 *
Trond Myklebust7c85d902006-12-13 15:23:48 -05004 * Write file data over NFS.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
6 * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
7 */
8
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/writeback.h>
Peter Zijlstra89a09142007-03-16 13:38:26 -080015#include <linux/swap.h>
Trond Myklebust074cc1d2009-08-10 08:54:13 -040016#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070022#include <linux/backing-dev.h>
Paul Gortmakerafeacc82011-05-26 16:00:52 -040023#include <linux/export.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070024
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27#include "delegation.h"
Trond Myklebust49a70f22006-12-05 00:35:38 -050028#include "internal.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050029#include "iostat.h"
Andy Adamsondef6ed72009-04-01 09:22:26 -040030#include "nfs4_fs.h"
Trond Myklebust074cc1d2009-08-10 08:54:13 -040031#include "fscache.h"
Fred Isaman94ad1c82011-03-01 01:34:14 +000032#include "pnfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Trond Myklebustf4ce1292013-08-19 18:59:33 -040034#include "nfstrace.h"
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#define NFSDBG_FACILITY NFSDBG_PAGECACHE
37
38#define MIN_POOL_WRITE (32)
39#define MIN_POOL_COMMIT (4)
40
41/*
42 * Local function declarations
43 */
Fred Isamanf8512ad2008-03-19 11:24:39 -040044static void nfs_redirty_request(struct nfs_page *req);
Trond Myklebust788e7a82006-03-20 13:44:27 -050045static const struct rpc_call_ops nfs_commit_ops;
Fred Isaman061ae2e2012-04-20 14:47:48 -040046static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
Fred Isamanf453a542012-04-20 14:47:54 -040047static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
Anna Schumaker4a0de552014-05-06 09:12:30 -040048static const struct nfs_rw_ops nfs_rw_write_ops;
Weston Andros Adamsond4581382014-07-11 10:20:48 -040049static void nfs_clear_request_commit(struct nfs_page *req);
Weston Andros Adamson02d14262014-07-17 20:42:17 -040050static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
51 struct inode *inode);
Trond Myklebust3a3908c2014-09-08 22:21:00 -070052static struct nfs_page *
53nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
54 struct page *page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Christoph Lametere18b8902006-12-06 20:33:20 -080056static struct kmem_cache *nfs_wdata_cachep;
Trond Myklebust3feb2d42006-03-20 13:44:37 -050057static mempool_t *nfs_wdata_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -040058static struct kmem_cache *nfs_cdata_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059static mempool_t *nfs_commit_mempool;
60
Fred Isaman0b7c0152012-04-20 14:47:39 -040061struct nfs_commit_data *nfs_commitdata_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062{
Mel Gorman192e5012012-07-31 16:45:16 -070063 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
Chuck Lever40859d72005-11-30 18:09:02 -050064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 if (p) {
66 memset(p, 0, sizeof(*p));
67 INIT_LIST_HEAD(&p->pages);
68 }
69 return p;
70}
Fred Isamane0c2b382011-03-23 13:27:53 +000071EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Fred Isaman0b7c0152012-04-20 14:47:39 -040073void nfs_commit_free(struct nfs_commit_data *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 mempool_free(p, nfs_commit_mempool);
76}
Fred Isamane0c2b382011-03-23 13:27:53 +000077EXPORT_SYMBOL_GPL(nfs_commit_free);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040079static struct nfs_pgio_header *nfs_writehdr_alloc(void)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050080{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040081 struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050082
Anna Schumaker4a0de552014-05-06 09:12:30 -040083 if (p)
Trond Myklebust3feb2d42006-03-20 13:44:37 -050084 memset(p, 0, sizeof(*p));
Trond Myklebust3feb2d42006-03-20 13:44:37 -050085 return p;
86}
87
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040088static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -040089{
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -040090 mempool_free(hdr, nfs_wdata_mempool);
Trond Myklebust3feb2d42006-03-20 13:44:37 -050091}
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Trond Myklebust7b159fc2007-07-25 14:09:54 -040093static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
94{
95 ctx->error = error;
96 smp_wmb();
97 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
98}
99
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400100/*
101 * nfs_page_find_head_request_locked - find head request associated with @page
102 *
103 * must be called while holding the inode lock.
104 *
105 * returns matching head request with reference held, or NULL if not found.
106 */
Mel Gorman29418aa2012-07-31 16:45:10 -0700107static struct nfs_page *
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400108nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -0500109{
110 struct nfs_page *req = NULL;
111
Mel Gorman29418aa2012-07-31 16:45:10 -0700112 if (PagePrivate(page))
Trond Myklebust277459d2006-12-05 00:35:35 -0500113 req = (struct nfs_page *)page_private(page);
Weston Andros Adamson02d14262014-07-17 20:42:17 -0400114 else if (unlikely(PageSwapCache(page)))
115 req = nfs_page_search_commits_for_head_request_locked(nfsi,
116 page);
Mel Gorman29418aa2012-07-31 16:45:10 -0700117
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400118 if (req) {
119 WARN_ON_ONCE(req->wb_head != req);
Mel Gorman29418aa2012-07-31 16:45:10 -0700120 kref_get(&req->wb_kref);
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400121 }
Mel Gorman29418aa2012-07-31 16:45:10 -0700122
Trond Myklebust277459d2006-12-05 00:35:35 -0500123 return req;
124}
125
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400126/*
127 * nfs_page_find_head_request - find head request associated with @page
128 *
129 * returns matching head request with reference held, or NULL if not found.
130 */
131static struct nfs_page *nfs_page_find_head_request(struct page *page)
Trond Myklebust277459d2006-12-05 00:35:35 -0500132{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700133 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebust277459d2006-12-05 00:35:35 -0500134 struct nfs_page *req = NULL;
Trond Myklebust277459d2006-12-05 00:35:35 -0500135
Trond Myklebust587142f2007-07-02 09:57:54 -0400136 spin_lock(&inode->i_lock);
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -0400137 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebust587142f2007-07-02 09:57:54 -0400138 spin_unlock(&inode->i_lock);
Trond Myklebust277459d2006-12-05 00:35:35 -0500139 return req;
140}
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142/* Adjust the file length if we're writing beyond the end */
143static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
144{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700145 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebusta3d01452008-06-11 12:21:19 -0400146 loff_t end, i_size;
147 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Trond Myklebusta3d01452008-06-11 12:21:19 -0400149 spin_lock(&inode->i_lock);
150 i_size = i_size_read(inode);
151 end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700152 if (i_size > 0 && page_file_index(page) < end_index)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400153 goto out;
Mel Gormand56b4dd2012-07-31 16:45:06 -0700154 end = page_file_offset(page) + ((loff_t)offset+count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (i_size >= end)
Trond Myklebusta3d01452008-06-11 12:21:19 -0400156 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 i_size_write(inode, end);
Trond Myklebusta3d01452008-06-11 12:21:19 -0400158 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
159out:
160 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
Trond Myklebusta301b772007-02-06 11:07:15 -0800163/* A writeback failed: mark the page as bad, and invalidate the page cache */
164static void nfs_set_pageerror(struct page *page)
165{
Mel Gormand56b4dd2012-07-31 16:45:06 -0700166 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
Trond Myklebusta301b772007-02-06 11:07:15 -0800167}
168
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400169/*
170 * nfs_page_group_search_locked
171 * @head - head request of page group
172 * @page_offset - offset into page
173 *
174 * Search page group with head @head to find a request that contains the
175 * page offset @page_offset.
176 *
177 * Returns a pointer to the first matching nfs request, or NULL if no
178 * match is found.
179 *
180 * Must be called with the page group lock held
181 */
182static struct nfs_page *
183nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
184{
185 struct nfs_page *req;
186
187 WARN_ON_ONCE(head != head->wb_head);
188 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
189
190 req = head;
191 do {
192 if (page_offset >= req->wb_pgbase &&
193 page_offset < (req->wb_pgbase + req->wb_bytes))
194 return req;
195
196 req = req->wb_this_page;
197 } while (req != head);
198
199 return NULL;
200}
201
202/*
203 * nfs_page_group_covers_page
204 * @head - head request of page group
205 *
206 * Return true if the page group with head @head covers the whole page,
207 * returns false otherwise
208 */
209static bool nfs_page_group_covers_page(struct nfs_page *req)
210{
211 struct nfs_page *tmp;
212 unsigned int pos = 0;
213 unsigned int len = nfs_page_length(req->wb_page);
214
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400215 nfs_page_group_lock(req, false);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400216
217 do {
218 tmp = nfs_page_group_search_locked(req->wb_head, pos);
219 if (tmp) {
220 /* no way this should happen */
221 WARN_ON_ONCE(tmp->wb_pgbase != pos);
222 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
223 }
224 } while (tmp && pos < len);
225
226 nfs_page_group_unlock(req);
227 WARN_ON_ONCE(pos > len);
228 return pos == len;
229}
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231/* We can set the PG_uptodate flag if we see that a write request
232 * covers the full page.
233 */
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400234static void nfs_mark_uptodate(struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235{
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400236 if (PageUptodate(req->wb_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400238 if (!nfs_page_group_covers_page(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 return;
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -0400240 SetPageUptodate(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243static int wb_priority(struct writeback_control *wbc)
244{
NeilBrowne87b4c72014-09-18 16:09:27 +1000245 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 if (wbc->for_reclaim)
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400247 return FLUSH_HIGHPRI | FLUSH_STABLE;
NeilBrowne87b4c72014-09-18 16:09:27 +1000248 if (wbc->sync_mode == WB_SYNC_ALL)
249 ret = FLUSH_COND_STABLE;
Wu Fengguangb17621f2009-12-03 13:54:25 +0100250 if (wbc->for_kupdate || wbc->for_background)
NeilBrowne87b4c72014-09-18 16:09:27 +1000251 ret |= FLUSH_LOWPRI;
252 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
255/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800256 * NFS congestion control
257 */
258
259int nfs_congestion_kb;
260
261#define NFS_CONGESTION_ON_THRESH (nfs_congestion_kb >> (PAGE_SHIFT-10))
262#define NFS_CONGESTION_OFF_THRESH \
263 (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
264
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400265static void nfs_set_page_writeback(struct page *page)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800266{
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400267 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
Trond Myklebust5a6d41b2007-04-14 19:10:12 -0400268 int ret = test_set_page_writeback(page);
269
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400270 WARN_ON_ONCE(ret != 0);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800271
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400272 if (atomic_long_inc_return(&nfss->writeback) >
273 NFS_CONGESTION_ON_THRESH) {
274 set_bdi_congested(&nfss->backing_dev_info,
275 BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800276 }
277}
278
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400279static void nfs_end_page_writeback(struct nfs_page *req)
Peter Zijlstra89a09142007-03-16 13:38:26 -0800280{
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400281 struct inode *inode = page_file_mapping(req->wb_page)->host;
Peter Zijlstra89a09142007-03-16 13:38:26 -0800282 struct nfs_server *nfss = NFS_SERVER(inode);
283
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400284 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
285 return;
286
287 end_page_writeback(req->wb_page);
Peter Zijlstrac4dc4be2007-10-16 23:25:41 -0700288 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
Jens Axboe8aa7e842009-07-09 14:52:32 +0200289 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
Peter Zijlstra89a09142007-03-16 13:38:26 -0800290}
291
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400292
293/* nfs_page_group_clear_bits
294 * @req - an nfs request
295 * clears all page group related bits from @req
296 */
297static void
298nfs_page_group_clear_bits(struct nfs_page *req)
Trond Myklebuste261f512006-12-05 00:35:41 -0500299{
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400300 clear_bit(PG_TEARDOWN, &req->wb_flags);
301 clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
302 clear_bit(PG_UPTODATE, &req->wb_flags);
303 clear_bit(PG_WB_END, &req->wb_flags);
304 clear_bit(PG_REMOVE, &req->wb_flags);
305}
306
307
308/*
309 * nfs_unroll_locks_and_wait - unlock all newly locked reqs and wait on @req
310 *
311 * this is a helper function for nfs_lock_and_join_requests
312 *
313 * @inode - inode associated with request page group, must be holding inode lock
314 * @head - head request of page group, must be holding head lock
315 * @req - request that couldn't lock and needs to wait on the req bit lock
316 * @nonblock - if true, don't actually wait
317 *
318 * NOTE: this must be called holding page_group bit lock and inode spin lock
319 * and BOTH will be released before returning.
320 *
321 * returns 0 on success, < 0 on error.
322 */
323static int
324nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
325 struct nfs_page *req, bool nonblock)
326 __releases(&inode->i_lock)
327{
328 struct nfs_page *tmp;
Trond Myklebuste261f512006-12-05 00:35:41 -0500329 int ret;
330
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400331 /* relinquish all the locks successfully grabbed this run */
332 for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
333 nfs_unlock_request(tmp);
334
335 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
336
337 /* grab a ref on the request that will be waited on */
338 kref_get(&req->wb_kref);
339
340 nfs_page_group_unlock(head);
Trond Myklebust587142f2007-07-02 09:57:54 -0400341 spin_unlock(&inode->i_lock);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400342
343 /* release ref from nfs_page_find_head_request_locked */
344 nfs_release_request(head);
345
346 if (!nonblock)
347 ret = nfs_wait_on_request(req);
348 else
349 ret = -EAGAIN;
350 nfs_release_request(req);
351
352 return ret;
353}
354
355/*
356 * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
357 *
358 * @destroy_list - request list (using wb_this_page) terminated by @old_head
359 * @old_head - the old head of the list
360 *
361 * All subrequests must be locked and removed from all lists, so at this point
362 * they are only "active" in this function, and possibly in nfs_wait_on_request
363 * with a reference held by some other context.
364 */
365static void
366nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
367 struct nfs_page *old_head)
368{
369 while (destroy_list) {
370 struct nfs_page *subreq = destroy_list;
371
372 destroy_list = (subreq->wb_this_page == old_head) ?
373 NULL : subreq->wb_this_page;
374
375 WARN_ON_ONCE(old_head != subreq->wb_head);
376
377 /* make sure old group is not used */
378 subreq->wb_head = subreq;
379 subreq->wb_this_page = subreq;
380
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400381 /* subreq is now totally disconnected from page group or any
382 * write / commit lists. last chance to wake any waiters */
383 nfs_unlock_request(subreq);
384
385 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
386 /* release ref on old head request */
387 nfs_release_request(old_head);
388
389 nfs_page_group_clear_bits(subreq);
390
391 /* release the PG_INODE_REF reference */
392 if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
393 nfs_release_request(subreq);
394 else
395 WARN_ON_ONCE(1);
396 } else {
397 WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
398 /* zombie requests have already released the last
399 * reference and were waiting on the rest of the
400 * group to complete. Since it's no longer part of a
401 * group, simply free the request */
402 nfs_page_group_clear_bits(subreq);
403 nfs_free_request(subreq);
404 }
405 }
406}
407
408/*
409 * nfs_lock_and_join_requests - join all subreqs to the head req and return
410 * a locked reference, cancelling any pending
411 * operations for this page.
412 *
413 * @page - the page used to lookup the "page group" of nfs_page structures
414 * @nonblock - if true, don't block waiting for request locks
415 *
416 * This function joins all sub requests to the head request by first
417 * locking all requests in the group, cancelling any pending operations
418 * and finally updating the head request to cover the whole range covered by
419 * the (former) group. All subrequests are removed from any write or commit
420 * lists, unlinked from the group and destroyed.
421 *
422 * Returns a locked, referenced pointer to the head request - which after
423 * this call is guaranteed to be the only request associated with the page.
424 * Returns NULL if no requests are found for @page, or a ERR_PTR if an
425 * error was encountered.
426 */
427static struct nfs_page *
428nfs_lock_and_join_requests(struct page *page, bool nonblock)
429{
430 struct inode *inode = page_file_mapping(page)->host;
431 struct nfs_page *head, *subreq;
432 struct nfs_page *destroy_list = NULL;
433 unsigned int total_bytes;
434 int ret;
435
436try_again:
437 total_bytes = 0;
438
439 WARN_ON_ONCE(destroy_list);
440
441 spin_lock(&inode->i_lock);
442
443 /*
444 * A reference is taken only on the head request which acts as a
445 * reference to the whole page group - the group will not be destroyed
446 * until the head reference is released.
447 */
448 head = nfs_page_find_head_request_locked(NFS_I(inode), page);
449
450 if (!head) {
451 spin_unlock(&inode->i_lock);
452 return NULL;
453 }
454
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400455 /* holding inode lock, so always make a non-blocking call to try the
456 * page group lock */
Weston Andros Adamsonfd2f3a02014-08-08 11:00:53 -0400457 ret = nfs_page_group_lock(head, true);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400458 if (ret < 0) {
459 spin_unlock(&inode->i_lock);
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400460
461 if (!nonblock && ret == -EAGAIN) {
462 nfs_page_group_lock_wait(head);
463 nfs_release_request(head);
464 goto try_again;
465 }
466
Weston Andros Adamson94970012014-08-08 11:00:56 -0400467 nfs_release_request(head);
Weston Andros Adamsone7029202014-07-17 20:42:15 -0400468 return ERR_PTR(ret);
Weston Andros Adamson94970012014-08-08 11:00:56 -0400469 }
Weston Andros Adamson7c3af972014-08-08 11:00:57 -0400470
471 /* lock each request in the page group */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400472 subreq = head;
473 do {
474 /*
475 * Subrequests are always contiguous, non overlapping
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400476 * and in order - but may be repeated (mirrored writes).
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400477 */
Weston Andros Adamson309a1d62014-09-05 16:34:29 -0400478 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
479 /* keep track of how many bytes this group covers */
480 total_bytes += subreq->wb_bytes;
481 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
482 ((subreq->wb_offset + subreq->wb_bytes) >
483 (head->wb_offset + total_bytes)))) {
484 nfs_page_group_unlock(head);
485 spin_unlock(&inode->i_lock);
486 return ERR_PTR(-EIO);
487 }
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400488
489 if (!nfs_lock_request(subreq)) {
490 /* releases page group bit lock and
491 * inode spin lock and all references */
492 ret = nfs_unroll_locks_and_wait(inode, head,
493 subreq, nonblock);
494
495 if (ret == 0)
496 goto try_again;
497
498 return ERR_PTR(ret);
499 }
500
501 subreq = subreq->wb_this_page;
502 } while (subreq != head);
503
504 /* Now that all requests are locked, make sure they aren't on any list.
505 * Commit list removal accounting is done after locks are dropped */
506 subreq = head;
507 do {
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400508 nfs_clear_request_commit(subreq);
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400509 subreq = subreq->wb_this_page;
510 } while (subreq != head);
511
512 /* unlink subrequests from head, destroy them later */
513 if (head->wb_this_page != head) {
514 /* destroy list will be terminated by head */
515 destroy_list = head->wb_this_page;
516 head->wb_this_page = head;
517
518 /* change head request to cover whole range that
519 * the former page group covered */
520 head->wb_bytes = total_bytes;
521 }
522
523 /*
524 * prepare head request to be added to new pgio descriptor
525 */
526 nfs_page_group_clear_bits(head);
527
528 /*
529 * some part of the group was still on the inode list - otherwise
530 * the group wouldn't be involved in async write.
531 * grab a reference for the head request, iff it needs one.
532 */
533 if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
534 kref_get(&head->wb_kref);
535
536 nfs_page_group_unlock(head);
537
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400538 /* drop lock to clean uprequests on destroy list */
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400539 spin_unlock(&inode->i_lock);
540
541 nfs_destroy_unlinked_subrequests(destroy_list, head);
542
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400543 /* still holds ref on head from nfs_page_find_head_request_locked
544 * and still has lock on head from lock loop */
545 return head;
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400546}
547
548/*
549 * Find an associated nfs write request, and prepare to flush it out
550 * May return an error if the user signalled nfs_wait_on_request().
551 */
552static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400553 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400554{
555 struct nfs_page *req;
556 int ret = 0;
557
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400558 req = nfs_lock_and_join_requests(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400559 if (!req)
560 goto out;
561 ret = PTR_ERR(req);
562 if (IS_ERR(req))
563 goto out;
564
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400565 nfs_set_page_writeback(page);
566 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400567
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400568 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400569 if (!nfs_pageio_add_request(pgio, req)) {
570 nfs_redirty_request(req);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400571 ret = pgio->pg_error;
Trond Myklebust40f90272015-10-01 11:36:38 -0400572 } else
573 nfs_add_stats(page_file_mapping(page)->host,
574 NFSIOS_WRITEPAGES, 1);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400575out:
576 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500577}
578
Trond Myklebustf758c8852007-07-22 19:27:32 -0400579static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
580{
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400581 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400582
Mel Gormand56b4dd2012-07-31 16:45:06 -0700583 nfs_pageio_cond_complete(pgio, page_file_index(page));
Wu Fengguang1b430be2010-10-26 14:21:26 -0700584 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400585 if (ret == -EAGAIN) {
586 redirty_page_for_writepage(wbc, page);
587 ret = 0;
588 }
589 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400590}
591
Trond Myklebuste261f512006-12-05 00:35:41 -0500592/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 * Write an mmapped page to the server.
594 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500595static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400597 struct nfs_pageio_descriptor pgio;
Trond Myklebust40f90272015-10-01 11:36:38 -0400598 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500599 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
Trond Myklebust40f90272015-10-01 11:36:38 -0400601 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
602 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200603 false, &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400604 err = nfs_do_writepage(page, wbc, &pgio);
605 nfs_pageio_complete(&pgio);
606 if (err < 0)
607 return err;
608 if (pgio.pg_error < 0)
609 return pgio.pg_error;
610 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500611}
612
613int nfs_writepage(struct page *page, struct writeback_control *wbc)
614{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400615 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500616
Trond Myklebustf758c8852007-07-22 19:27:32 -0400617 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400619 return ret;
620}
621
622static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
623{
624 int ret;
625
626 ret = nfs_do_writepage(page, wbc, data);
627 unlock_page(page);
628 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
632{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400634 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400635 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 int err;
637
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400638 /* Stop dirtying of new pages while we sync */
NeilBrown74316202014-07-07 15:16:04 +1000639 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400640 nfs_wait_bit_killable, TASK_KILLABLE);
641 if (err)
642 goto out_err;
643
Chuck Lever91d5b472006-03-20 13:44:14 -0500644 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
645
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200646 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
647 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400648 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400649 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400650
651 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100652 smp_mb__after_atomic();
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400653 wake_up_bit(bitlock, NFS_INO_FLUSHING);
654
Trond Myklebustf758c8852007-07-22 19:27:32 -0400655 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400656 goto out_err;
657 err = pgio.pg_error;
658 if (err < 0)
659 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400660 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400661out_err:
662 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/*
666 * Insert a write request into an inode
667 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500668static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669{
670 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400671
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400672 WARN_ON_ONCE(req->wb_this_page != req);
673
Trond Myklebuste7d39062008-06-13 12:12:32 -0400674 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400675 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400676
677 spin_lock(&inode->i_lock);
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500678 if (!nfsi->nrequests &&
679 NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700680 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700681 /*
682 * Swap-space should not get truncated. Hence no need to plug the race
683 * with invalidate/truncate.
684 */
685 if (likely(!PageSwapCache(req->wb_page))) {
686 set_bit(PG_MAPPED, &req->wb_flags);
687 SetPagePrivate(req->wb_page);
688 set_page_private(req->wb_page, (unsigned long)req);
689 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500690 nfsi->nrequests++;
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400691 /* this a head request for a page group - mark it as having an
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500692 * extra reference so sub groups can follow suit.
693 * This flag also informs pgio layer when to bump nrequests when
694 * adding subrequests. */
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400695 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
Trond Myklebustc03b4022007-06-17 13:26:38 -0400696 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400697 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698}
699
700/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800701 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 */
703static void nfs_inode_remove_request(struct nfs_page *req)
704{
David Howells2b0143b2015-03-17 22:25:59 +0000705 struct inode *inode = d_inode(req->wb_context->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400707 struct nfs_page *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400709 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
710 head = req->wb_head;
711
712 spin_lock(&inode->i_lock);
713 if (likely(!PageSwapCache(head->wb_page))) {
714 set_page_private(head->wb_page, 0);
715 ClearPagePrivate(head->wb_page);
NeilBrown95905442014-09-24 11:28:32 +1000716 smp_mb__after_atomic();
717 wake_up_page(head->wb_page, PG_private);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400718 clear_bit(PG_MAPPED, &head->wb_flags);
719 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500720 nfsi->nrequests--;
721 spin_unlock(&inode->i_lock);
722 } else {
723 spin_lock(&inode->i_lock);
724 nfsi->nrequests--;
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400725 spin_unlock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700726 }
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400727
728 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
729 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
Trond Myklebust61822ab2006-12-05 00:35:42 -0500732static void
Fred6d884e82008-03-19 11:24:38 -0400733nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500734{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500735 __set_page_dirty_nobuffers(req->wb_page);
736}
737
Trond Myklebust3a3908c2014-09-08 22:21:00 -0700738/*
739 * nfs_page_search_commits_for_head_request_locked
740 *
741 * Search through commit lists on @inode for the head request for @page.
742 * Must be called while holding the inode (which is cinfo) lock.
743 *
744 * Returns the head request if found, or NULL if not found.
745 */
746static struct nfs_page *
747nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
748 struct page *page)
749{
750 struct nfs_page *freq, *t;
751 struct nfs_commit_info cinfo;
752 struct inode *inode = &nfsi->vfs_inode;
753
754 nfs_init_cinfo_from_inode(&cinfo, inode);
755
756 /* search through pnfs commit lists */
757 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
758 if (freq)
759 return freq->wb_head;
760
761 /* Linearly search the commit list for the correct request */
762 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
763 if (freq->wb_page == page)
764 return freq->wb_head;
765 }
766
767 return NULL;
768}
769
Trond Myklebust8dd37752012-03-15 17:16:40 -0400770/**
Trond Myklebust86d80f92015-07-31 16:24:30 -0400771 * nfs_request_add_commit_list_locked - add request to a commit list
772 * @req: pointer to a struct nfs_page
773 * @dst: commit list head
774 * @cinfo: holds list lock and accounting info
775 *
776 * This sets the PG_CLEAN bit, updates the cinfo count of
777 * number of outstanding requests requiring a commit as well as
778 * the MM page stats.
779 *
780 * The caller must hold the cinfo->lock, and the nfs_page lock.
781 */
782void
783nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
784 struct nfs_commit_info *cinfo)
785{
786 set_bit(PG_CLEAN, &req->wb_flags);
787 nfs_list_add_request(req, dst);
788 cinfo->mds->ncommit++;
789}
790EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
791
792/**
Trond Myklebust8dd37752012-03-15 17:16:40 -0400793 * nfs_request_add_commit_list - add request to a commit list
794 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400795 * @dst: commit list head
796 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400797 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400798 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400799 * number of outstanding requests requiring a commit as well as
800 * the MM page stats.
801 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400802 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400803 * holding the nfs_page lock.
804 */
805void
Fred Isamanea2cf222012-04-20 14:47:53 -0400806nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
807 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400808{
Fred Isamanea2cf222012-04-20 14:47:53 -0400809 spin_lock(cinfo->lock);
Trond Myklebust86d80f92015-07-31 16:24:30 -0400810 nfs_request_add_commit_list_locked(req, dst, cinfo);
Fred Isamanea2cf222012-04-20 14:47:53 -0400811 spin_unlock(cinfo->lock);
Trond Myklebust86d80f92015-07-31 16:24:30 -0400812 nfs_mark_page_unstable(req->wb_page, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400813}
814EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
815
816/**
817 * nfs_request_remove_commit_list - Remove request from a commit list
818 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400819 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400820 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400821 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400822 * number of outstanding requests requiring a commit
823 * It does not update the MM page stats.
824 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400825 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400826 */
827void
Fred Isamanea2cf222012-04-20 14:47:53 -0400828nfs_request_remove_commit_list(struct nfs_page *req,
829 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400830{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400831 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
832 return;
833 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400834 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400835}
836EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
837
Fred Isamanea2cf222012-04-20 14:47:53 -0400838static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
839 struct inode *inode)
840{
841 cinfo->lock = &inode->i_lock;
842 cinfo->mds = &NFS_I(inode)->commit_info;
843 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400844 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400845 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400846}
847
848void nfs_init_cinfo(struct nfs_commit_info *cinfo,
849 struct inode *inode,
850 struct nfs_direct_req *dreq)
851{
Fred Isaman1763da12012-04-20 14:47:57 -0400852 if (dreq)
853 nfs_init_cinfo_from_dreq(cinfo, dreq);
854 else
855 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400856}
857EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/*
860 * Add a request to the inode's commit list.
861 */
Fred Isaman1763da12012-04-20 14:47:57 -0400862void
Fred Isamanea2cf222012-04-20 14:47:53 -0400863nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400864 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400866 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400867 return;
Fred Isamanea2cf222012-04-20 14:47:53 -0400868 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400870
Fred Isamand6d6dc72012-03-08 17:29:35 -0500871static void
872nfs_clear_page_commit(struct page *page)
873{
874 dec_zone_page_state(page, NR_UNSTABLE_NFS);
Tejun Heo93f78d82015-05-22 17:13:27 -0400875 dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
876 WB_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500877}
878
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400879/* Called holding inode (/cinfo) lock */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400880static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400881nfs_clear_request_commit(struct nfs_page *req)
882{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400883 if (test_bit(PG_CLEAN, &req->wb_flags)) {
David Howells2b0143b2015-03-17 22:25:59 +0000884 struct inode *inode = d_inode(req->wb_context->dentry);
Fred Isamanea2cf222012-04-20 14:47:53 -0400885 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400886
Fred Isamanea2cf222012-04-20 14:47:53 -0400887 nfs_init_cinfo_from_inode(&cinfo, inode);
888 if (!pnfs_clear_request_commit(req, &cinfo)) {
Fred Isamanea2cf222012-04-20 14:47:53 -0400889 nfs_request_remove_commit_list(req, &cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400890 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500891 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400892 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400893}
894
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400895int nfs_write_need_commit(struct nfs_pgio_header *hdr)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400896{
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400897 if (hdr->verf.committed == NFS_DATA_SYNC)
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400898 return hdr->lseg == NULL;
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400899 return hdr->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400900}
901
Fred Isaman061ae2e2012-04-20 14:47:48 -0400902static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400903{
Fred Isamanea2cf222012-04-20 14:47:53 -0400904 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400905 unsigned long bytes = 0;
906
907 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
908 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400909 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400910 while (!list_empty(&hdr->pages)) {
911 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400912
913 bytes += req->wb_bytes;
914 nfs_list_remove_request(req);
915 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
916 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400917 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400918 nfs_context_set_write_error(req->wb_context, hdr->error);
919 goto remove_req;
920 }
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400921 if (nfs_write_need_commit(hdr)) {
Anna Schumakerf79d06f2014-05-06 09:12:28 -0400922 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400923 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -0400924 hdr->pgio_mirror_idx);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400925 goto next;
926 }
927remove_req:
928 nfs_inode_remove_request(req);
929next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400930 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400931 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400932 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400933 }
934out:
935 hdr->release(hdr);
936}
937
Anna Schumakerce595152014-05-06 09:12:34 -0400938unsigned long
Fred Isamanea2cf222012-04-20 14:47:53 -0400939nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400940{
Fred Isamanea2cf222012-04-20 14:47:53 -0400941 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400942}
943
Fred Isamanea2cf222012-04-20 14:47:53 -0400944/* cinfo->lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400945int
Fred Isamanea2cf222012-04-20 14:47:53 -0400946nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
947 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500948{
949 struct nfs_page *req, *tmp;
950 int ret = 0;
951
952 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400953 if (!nfs_lock_request(req))
954 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400955 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -0400956 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400957 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400958 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400959 nfs_list_add_request(req, dst);
960 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400961 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400962 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500963 }
964 return ret;
965}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500966
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967/*
968 * nfs_scan_commit - Scan an inode for commit requests
969 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400970 * @dst: mds destination list
971 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 *
973 * Moves requests from the inode's 'commit' request list.
974 * The requests are *not* checked to ensure that they form a contiguous set.
975 */
Fred Isaman1763da12012-04-20 14:47:57 -0400976int
Fred Isamanea2cf222012-04-20 14:47:53 -0400977nfs_scan_commit(struct inode *inode, struct list_head *dst,
978 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
Fred Isamand6d6dc72012-03-08 17:29:35 -0500980 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400981
Fred Isamanea2cf222012-04-20 14:47:53 -0400982 spin_lock(cinfo->lock);
983 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400984 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500985
Fred Isamanea2cf222012-04-20 14:47:53 -0400986 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
987 cinfo, max);
988 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500989 }
Fred Isamanea2cf222012-04-20 14:47:53 -0400990 spin_unlock(cinfo->lock);
Trond Myklebustff778d02010-02-19 16:53:39 -0800991 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994/*
Trond Myklebuste7d39062008-06-13 12:12:32 -0400995 * Search for an existing write request, and attempt to update
996 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 *
Trond Myklebuste7d39062008-06-13 12:12:32 -0400998 * If the attempt fails, then the existing request is flushed out
999 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 */
Trond Myklebuste7d39062008-06-13 12:12:32 -04001001static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1002 struct page *page,
1003 unsigned int offset,
1004 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
Trond Myklebuste7d39062008-06-13 12:12:32 -04001006 struct nfs_page *req;
1007 unsigned int rqend;
1008 unsigned int end;
1009 int error;
1010
1011 if (!PagePrivate(page))
1012 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
1014 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001015 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 for (;;) {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001018 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001019 if (req == NULL)
1020 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -05001021
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001022 /* should be handled by nfs_flush_incompatible */
1023 WARN_ON_ONCE(req->wb_head != req);
1024 WARN_ON_ONCE(req->wb_this_page != req);
1025
Trond Myklebuste7d39062008-06-13 12:12:32 -04001026 rqend = req->wb_offset + req->wb_bytes;
1027 /*
1028 * Tell the caller to flush out the request if
1029 * the offsets are non-contiguous.
1030 * Note: nfs_flush_incompatible() will already
1031 * have flushed out requests having wrong owners.
1032 */
Trond Myklebuste468bae2008-06-13 13:25:22 -04001033 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -04001034 || end < req->wb_offset)
1035 goto out_flushme;
1036
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001037 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Trond Myklebuste7d39062008-06-13 12:12:32 -04001040 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -04001041 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001042 error = nfs_wait_on_request(req);
1043 nfs_release_request(req);
1044 if (error != 0)
1045 goto out_err;
1046 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048
1049 /* Okay, the request matches. Update the region */
1050 if (offset < req->wb_offset) {
1051 req->wb_offset = offset;
1052 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 if (end > rqend)
1055 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001056 else
1057 req->wb_bytes = rqend - req->wb_offset;
1058out_unlock:
Fred Isamanca138f32012-04-05 15:26:36 -04001059 if (req)
1060 nfs_clear_request_commit(req);
Weston Andros Adamson411a99a2014-07-17 20:42:19 -04001061 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001062 return req;
1063out_flushme:
1064 spin_unlock(&inode->i_lock);
1065 nfs_release_request(req);
1066 error = nfs_wb_page(inode, page);
1067out_err:
1068 return ERR_PTR(error);
1069}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Trond Myklebuste7d39062008-06-13 12:12:32 -04001071/*
1072 * Try to update an existing write request, or create one if there is none.
1073 *
1074 * Note: Should always be called with the Page Lock held to prevent races
1075 * if we have to add a new request. Also assumes that the caller has
1076 * already called nfs_flush_incompatible() if necessary.
1077 */
1078static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1079 struct page *page, unsigned int offset, unsigned int bytes)
1080{
Mel Gormand56b4dd2012-07-31 16:45:06 -07001081 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001082 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001083
1084 req = nfs_try_to_update_request(inode, page, offset, bytes);
1085 if (req != NULL)
1086 goto out;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001087 req = nfs_create_request(ctx, page, NULL, offset, bytes);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001088 if (IS_ERR(req))
1089 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001090 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -04001091out:
Trond Myklebust61e930a2007-10-18 17:08:05 -04001092 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093}
1094
Trond Myklebuste7d39062008-06-13 12:12:32 -04001095static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1096 unsigned int offset, unsigned int count)
1097{
1098 struct nfs_page *req;
1099
1100 req = nfs_setup_write_request(ctx, page, offset, count);
1101 if (IS_ERR(req))
1102 return PTR_ERR(req);
1103 /* Update file length */
1104 nfs_grow_file(page, offset, count);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -04001105 nfs_mark_uptodate(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001106 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001107 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001108 return 0;
1109}
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111int nfs_flush_incompatible(struct file *file, struct page *page)
1112{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001113 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -04001114 struct nfs_lock_context *l_ctx;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001115 struct file_lock_context *flctx = file_inode(file)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -05001117 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 /*
1119 * Look for a request corresponding to this page. If there
1120 * is one, and it belongs to another file, we flush it out
1121 * before we try to copy anything into the page. Do this
1122 * due to the lack of an ACCESS-type call in NFSv2.
1123 * Also do the same if we find a request from an existing
1124 * dropped page.
1125 */
Trond Myklebust1a545332006-12-05 00:35:40 -05001126 do {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001127 req = nfs_page_find_head_request(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001128 if (req == NULL)
1129 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -04001130 l_ctx = req->wb_lock_context;
1131 do_flush = req->wb_page != page || req->wb_context != ctx;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001132 /* for now, flush if more than 1 request in page_group */
1133 do_flush |= req->wb_this_page != req;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001134 if (l_ctx && flctx &&
1135 !(list_empty_careful(&flctx->flc_posix) &&
1136 list_empty_careful(&flctx->flc_flock))) {
Jeff Layton5263e312015-01-16 15:05:55 -05001137 do_flush |= l_ctx->lockowner.l_owner != current->files
1138 || l_ctx->lockowner.l_pid != current->tgid;
1139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -05001141 if (!do_flush)
1142 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -07001143 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001144 } while (status == 0);
1145 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
1147
1148/*
Andy Adamsondc248262013-08-14 11:59:16 -04001149 * Avoid buffered writes when a open context credential's key would
1150 * expire soon.
1151 *
1152 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1153 *
1154 * Return 0 and set a credential flag which triggers the inode to flush
1155 * and performs NFS_FILE_SYNC writes if the key will expired within
1156 * RPC_KEY_EXPIRE_TIMEO.
1157 */
1158int
1159nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1160{
1161 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1162 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1163
1164 return rpcauth_key_timeout_notify(auth, ctx->cred);
1165}
1166
1167/*
1168 * Test if the open context credential key is marked to expire soon.
1169 */
1170bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1171{
1172 return rpcauth_cred_key_to_expire(ctx->cred);
1173}
1174
1175/*
Trond Myklebust5d47a352008-02-07 17:24:07 -05001176 * If the page cache is marked as unsafe or invalid, then we can't rely on
1177 * the PageUptodate() flag. In this case, we will need to turn off
1178 * write optimisations that depend on the page contents being correct.
1179 */
Trond Myklebust8d197a52012-04-29 12:50:01 -04001180static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -05001181{
Jeff Laytond529ef82014-01-27 13:46:15 -05001182 struct nfs_inode *nfsi = NFS_I(inode);
1183
Trond Myklebust8d197a52012-04-29 12:50:01 -04001184 if (nfs_have_delegated_attributes(inode))
1185 goto out;
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001186 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
Jeff Laytond529ef82014-01-27 13:46:15 -05001187 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -05001188 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -05001189 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -04001190 return false;
1191out:
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001192 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1193 return false;
Trond Myklebust8d197a52012-04-29 12:50:01 -04001194 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -05001195}
1196
Jeff Layton5263e312015-01-16 15:05:55 -05001197static bool
1198is_whole_file_wrlock(struct file_lock *fl)
1199{
1200 return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1201 fl->fl_type == F_WRLCK;
1202}
1203
Scott Mayhewc7559662013-07-05 17:33:19 -04001204/* If we know the page is up to date, and we're not using byte range locks (or
1205 * if we have the whole file locked for writing), it may be more efficient to
1206 * extend the write to cover the entire page in order to avoid fragmentation
1207 * inefficiencies.
1208 *
Scott Mayhew263b4502014-01-17 15:12:05 -05001209 * If the file is opened for synchronous writes then we can just skip the rest
1210 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -04001211 */
1212static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1213{
Jeff Layton5263e312015-01-16 15:05:55 -05001214 int ret;
1215 struct file_lock_context *flctx = inode->i_flctx;
1216 struct file_lock *fl;
1217
Scott Mayhewc7559662013-07-05 17:33:19 -04001218 if (file->f_flags & O_DSYNC)
1219 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -05001220 if (!nfs_write_pageuptodate(page, inode))
1221 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -04001222 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1223 return 1;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001224 if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1225 list_empty_careful(&flctx->flc_posix)))
Trond Myklebust8fa45922015-10-01 18:38:27 -04001226 return 1;
Jeff Layton5263e312015-01-16 15:05:55 -05001227
1228 /* Check to see if there are whole file write locks */
Jeff Layton5263e312015-01-16 15:05:55 -05001229 ret = 0;
Jeff Layton6109c852015-01-16 15:05:57 -05001230 spin_lock(&flctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001231 if (!list_empty(&flctx->flc_posix)) {
1232 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1233 fl_list);
1234 if (is_whole_file_wrlock(fl))
1235 ret = 1;
1236 } else if (!list_empty(&flctx->flc_flock)) {
Jeff Layton5263e312015-01-16 15:05:55 -05001237 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1238 fl_list);
1239 if (fl->fl_type == F_WRLCK)
1240 ret = 1;
1241 }
Jeff Layton6109c852015-01-16 15:05:57 -05001242 spin_unlock(&flctx->flc_lock);
Jeff Layton5263e312015-01-16 15:05:55 -05001243 return ret;
Scott Mayhewc7559662013-07-05 17:33:19 -04001244}
1245
Trond Myklebust5d47a352008-02-07 17:24:07 -05001246/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 * Update and possibly write a cached page of an NFS file.
1248 *
1249 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1250 * things with a page scheduled for an RPC call (e.g. invalidate it).
1251 */
1252int nfs_updatepage(struct file *file, struct page *page,
1253 unsigned int offset, unsigned int count)
1254{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001255 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001256 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 int status = 0;
1258
Chuck Lever91d5b472006-03-20 13:44:14 -05001259 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1260
Al Viro6de14722013-09-16 10:53:17 -04001261 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1262 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Scott Mayhewc7559662013-07-05 17:33:19 -04001264 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -05001265 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 }
1268
Trond Myklebuste21195a2006-12-05 00:35:39 -05001269 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -04001270 if (status < 0)
1271 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -07001272 else
1273 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Chuck Lever48186c72008-06-11 17:56:05 -04001275 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 return status;
1278}
1279
Trond Myklebust3ff75762007-07-14 15:40:00 -04001280static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
1282 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1283 case FLUSH_HIGHPRI:
1284 return RPC_PRIORITY_HIGH;
1285 case FLUSH_LOWPRI:
1286 return RPC_PRIORITY_LOW;
1287 }
1288 return RPC_PRIORITY_NORMAL;
1289}
1290
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001291static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1292 struct rpc_message *msg,
Tom Haynesabde71f2014-06-09 13:12:20 -07001293 const struct nfs_rpc_ops *rpc_ops,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001294 struct rpc_task_setup *task_setup_data, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295{
Trond Myklebust3ff75762007-07-14 15:40:00 -04001296 int priority = flush_task_priority(how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001298 task_setup_data->priority = priority;
Tom Haynesabde71f2014-06-09 13:12:20 -07001299 rpc_ops->write_setup(hdr, msg);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001300
Tom Haynesabde71f2014-06-09 13:12:20 -07001301 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001302 &task_setup_data->rpc_client, msg, hdr);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001303}
1304
Fred6d884e82008-03-19 11:24:38 -04001305/* If a nfs_flush_* function fails, it should remove reqs from @head and
1306 * call this on each, which will prepare them to be retried on next
1307 * writeback using standard nfs.
1308 */
1309static void nfs_redirty_request(struct nfs_page *req)
1310{
1311 nfs_mark_request_dirty(req);
Trond Myklebustc7070112015-06-17 19:56:22 -04001312 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001313 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -04001314 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001315 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001316}
1317
Fred Isaman061ae2e2012-04-20 14:47:48 -04001318static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001319{
1320 struct nfs_page *req;
1321
1322 while (!list_empty(head)) {
1323 req = nfs_list_entry(head->next);
1324 nfs_list_remove_request(req);
1325 nfs_redirty_request(req);
1326 }
1327}
1328
Fred Isaman061ae2e2012-04-20 14:47:48 -04001329static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1330 .error_cleanup = nfs_async_write_error,
1331 .completion = nfs_write_completion,
1332};
1333
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001334void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001335 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001336 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001337{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001338 struct nfs_server *server = NFS_SERVER(inode);
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001339 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001340
1341#ifdef CONFIG_NFS_V4_1
1342 if (server->pnfs_curr_ld && !force_mds)
1343 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1344#endif
Anna Schumaker4a0de552014-05-06 09:12:30 -04001345 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1346 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001347}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001348EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001349
Trond Myklebustdce81292011-07-13 15:59:19 -04001350void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1351{
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001352 struct nfs_pgio_mirror *mirror;
1353
Kinglong Mee6f29b9b2015-09-20 23:03:28 +08001354 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1355 pgio->pg_ops->pg_cleanup(pgio);
1356
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001357 pgio->pg_ops = &nfs_pgio_rw_ops;
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001358
1359 nfs_pageio_stop_mirroring(pgio);
1360
1361 mirror = &pgio->pg_mirrors[0];
1362 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
Trond Myklebustdce81292011-07-13 15:59:19 -04001363}
Trond Myklebust1f945352011-07-13 15:59:57 -04001364EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
Fred Isaman0b7c0152012-04-20 14:47:39 -04001367void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1368{
1369 struct nfs_commit_data *data = calldata;
1370
1371 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1372}
1373
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001374/*
1375 * Special version of should_remove_suid() that ignores capabilities.
1376 */
1377static int nfs_should_remove_suid(const struct inode *inode)
1378{
1379 umode_t mode = inode->i_mode;
1380 int kill = 0;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001381
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001382 /* suid always must be killed */
1383 if (unlikely(mode & S_ISUID))
1384 kill = ATTR_KILL_SUID;
1385
1386 /*
1387 * sgid without any exec bits is just a mandatory locking mark; leave
1388 * it alone. If some exec bits are set, it's a real sgid; kill it.
1389 */
1390 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1391 kill |= ATTR_KILL_SGID;
1392
1393 if (unlikely(kill && S_ISREG(mode)))
1394 return kill;
1395
1396 return 0;
1397}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001398
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001399static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1400 struct nfs_fattr *fattr)
1401{
1402 struct nfs_pgio_args *argp = &hdr->args;
1403 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001404 u64 size = argp->offset + resp->count;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001405
1406 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001407 fattr->size = size;
1408 if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1409 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001410 return;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001411 }
1412 if (size != fattr->size)
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001413 return;
1414 /* Set attribute barrier */
1415 nfs_fattr_set_barrier(fattr);
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001416 /* ...and update size */
1417 fattr->valid |= NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001418}
1419
1420void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1421{
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001422 struct nfs_fattr *fattr = &hdr->fattr;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001423 struct inode *inode = hdr->inode;
1424
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001425 spin_lock(&inode->i_lock);
1426 nfs_writeback_check_extend(hdr, fattr);
1427 nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1428 spin_unlock(&inode->i_lock);
1429}
1430EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432/*
1433 * This function is called when the WRITE call is complete.
1434 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001435static int nfs_writeback_done(struct rpc_task *task,
1436 struct nfs_pgio_header *hdr,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001437 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001439 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Chuck Leverf551e442006-09-20 14:33:04 -04001441 /*
1442 * ->write_done will attempt to use post-op attributes to detect
1443 * conflicting writes by other clients. A strict interpretation
1444 * of close-to-open would allow us to continue caching even if
1445 * another writer had changed the file, but some applications
1446 * depend on tighter cache coherency when writing.
1447 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001448 status = NFS_PROTO(inode)->write_done(task, hdr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001449 if (status != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001450 return status;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001451 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001452
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001453 if (hdr->res.verf->committed < hdr->args.stable &&
1454 task->tk_status >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 /* We tried a write call, but the server did not
1456 * commit data to stable storage even though we
1457 * requested it.
1458 * Note: There is a known bug in Tru64 < 5.0 in which
1459 * the server reports NFS_DATA_SYNC, but performs
1460 * NFS_FILE_SYNC. We therefore implement this checking
1461 * as a dprintk() in order to avoid filling syslog.
1462 */
1463 static unsigned long complain;
1464
Fred Isamana69aef12011-03-03 15:13:47 +00001465 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001467 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001469 NFS_SERVER(inode)->nfs_client->cl_hostname,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001470 hdr->res.verf->committed, hdr->args.stable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 complain = jiffies + 300 * HZ;
1472 }
1473 }
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001474
1475 /* Deal with the suid/sgid bit corner case */
1476 if (nfs_should_remove_suid(inode))
1477 nfs_mark_for_revalidate(inode);
Anna Schumaker0eecb212014-05-06 09:12:32 -04001478 return 0;
1479}
1480
1481/*
1482 * This function is called when the WRITE call is complete.
1483 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001484static void nfs_writeback_result(struct rpc_task *task,
1485 struct nfs_pgio_header *hdr)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001486{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001487 struct nfs_pgio_args *argp = &hdr->args;
1488 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001489
1490 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 static unsigned long complain;
1492
Fred Isaman6c75dc02012-04-20 14:47:47 -04001493 /* This a short write! */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001494 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001497 if (resp->count == 0) {
1498 if (time_before(complain, jiffies)) {
1499 printk(KERN_WARNING
1500 "NFS: Server wrote zero bytes, expected %u.\n",
1501 argp->count);
1502 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 }
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001504 nfs_set_pgio_error(hdr, -EIO, argp->offset);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001505 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001506 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 }
Kinglong Meef8417b482015-10-16 17:23:29 +08001508
1509 /* For non rpc-based layout drivers, retry-through-MDS */
1510 if (!task->tk_ops) {
1511 hdr->pnfs_error = -EAGAIN;
1512 return;
1513 }
1514
Fred Isaman6c75dc02012-04-20 14:47:47 -04001515 /* Was this an NFSv2 write or an NFSv3 stable write? */
1516 if (resp->verf->committed != NFS_UNSTABLE) {
1517 /* Resend from where the server left off */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001518 hdr->mds_offset += resp->count;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001519 argp->offset += resp->count;
1520 argp->pgbase += resp->count;
1521 argp->count -= resp->count;
1522 } else {
1523 /* Resend as a stable write in order to avoid
1524 * headaches in the case of a server crash.
1525 */
1526 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001528 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
1532
Trond Myklebust71d0a612010-04-22 15:35:57 -04001533static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1534{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001535 int ret;
1536
Trond Myklebust71d0a612010-04-22 15:35:57 -04001537 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1538 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001539 if (!may_wait)
1540 return 0;
1541 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1542 NFS_INO_COMMIT,
1543 nfs_wait_bit_killable,
1544 TASK_KILLABLE);
1545 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001546}
1547
Fred Isamanf453a542012-04-20 14:47:54 -04001548static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001549{
1550 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001551 smp_mb__after_atomic();
Trond Myklebust71d0a612010-04-22 15:35:57 -04001552 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1553}
1554
Fred Isaman0b7c0152012-04-20 14:47:39 -04001555void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001557 put_nfs_open_context(data->context);
1558 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
Fred Isamane0c2b382011-03-23 13:27:53 +00001560EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
Fred Isaman0b7c0152012-04-20 14:47:39 -04001562int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Peng Taoc36aae92014-06-09 07:10:14 +08001563 const struct nfs_rpc_ops *nfs_ops,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001564 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001565 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566{
Trond Myklebust07737692007-10-25 18:42:54 -04001567 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001568 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001569 struct rpc_message msg = {
1570 .rpc_argp = &data->args,
1571 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001572 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001573 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001574 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001575 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001576 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001577 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001578 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001579 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001580 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001581 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001582 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001583 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001584 /* Set up the initial task struct. */
Peng Taoc36aae92014-06-09 07:10:14 +08001585 nfs_ops->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Kinglong Meeb4839eb2015-07-01 12:00:13 +08001587 dprintk("NFS: initiated commit call\n");
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001588
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001589 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1590 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1591
Trond Myklebust07737692007-10-25 18:42:54 -04001592 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001593 if (IS_ERR(task))
1594 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001595 if (how & FLUSH_SYNC)
1596 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001597 rpc_put_task(task);
1598 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599}
Fred Isamane0c2b382011-03-23 13:27:53 +00001600EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
Peng Tao378520b2014-08-07 10:15:02 +08001602static loff_t nfs_get_lwb(struct list_head *head)
1603{
1604 loff_t lwb = 0;
1605 struct nfs_page *req;
1606
1607 list_for_each_entry(req, head, wb_list)
1608 if (lwb < (req_offset(req) + req->wb_bytes))
1609 lwb = req_offset(req) + req->wb_bytes;
1610
1611 return lwb;
1612}
1613
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001615 * Set up the argument/result storage required for the RPC call.
1616 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001617void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001618 struct list_head *head,
1619 struct pnfs_layout_segment *lseg,
1620 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001621{
1622 struct nfs_page *first = nfs_list_entry(head->next);
David Howells2b0143b2015-03-17 22:25:59 +00001623 struct inode *inode = d_inode(first->wb_context->dentry);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001624
1625 /* Set up the RPC argument and reply structs
1626 * NB: take care not to mess about with data->commit et al. */
1627
1628 list_splice_init(head, &data->pages);
1629
1630 data->inode = inode;
1631 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001632 data->lseg = lseg; /* reference transferred */
Peng Tao378520b2014-08-07 10:15:02 +08001633 /* only set lwb for pnfs commit */
1634 if (lseg)
1635 data->lwb = nfs_get_lwb(&data->pages);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001636 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001637 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001638 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001639
1640 data->args.fh = NFS_FH(data->inode);
1641 /* Note: we always request a commit of the entire inode */
1642 data->args.offset = 0;
1643 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001644 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001645 data->res.fattr = &data->fattr;
1646 data->res.verf = &data->verf;
1647 nfs_fattr_init(&data->fattr);
1648}
Fred Isamane0c2b382011-03-23 13:27:53 +00001649EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001650
Fred Isamane0c2b382011-03-23 13:27:53 +00001651void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001652 struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001653 struct nfs_commit_info *cinfo,
1654 u32 ds_commit_idx)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001655{
1656 struct nfs_page *req;
1657
1658 while (!list_empty(page_list)) {
1659 req = nfs_list_entry(page_list->next);
1660 nfs_list_remove_request(req);
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001661 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
Tom Haynes487b9b82015-02-13 13:19:54 -08001662 if (!cinfo->dreq)
1663 nfs_clear_page_commit(req->wb_page);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001664 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001665 }
1666}
Fred Isamane0c2b382011-03-23 13:27:53 +00001667EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001668
Fred Isaman9ace33c2011-03-23 13:27:45 +00001669/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 * Commit dirty pages
1671 */
1672static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001673nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1674 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001676 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001678 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
1680 if (!data)
1681 goto out_bad;
1682
1683 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001684 nfs_init_commit(data, head, NULL, cinfo);
1685 atomic_inc(&cinfo->mds->rpcs_out);
Peng Taoc36aae92014-06-09 07:10:14 +08001686 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1687 data->mds_ops, how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 out_bad:
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001689 nfs_retry_commit(head, NULL, cinfo, 0);
Fred Isamanf453a542012-04-20 14:47:54 -04001690 cinfo->completion_ops->error_cleanup(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 return -ENOMEM;
1692}
1693
1694/*
1695 * COMMIT call returned
1696 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001697static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001699 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Chuck Levera3f565b2007-01-31 12:14:01 -05001701 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 task->tk_pid, task->tk_status);
1703
Trond Myklebust788e7a82006-03-20 13:44:27 -05001704 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001705 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001706}
1707
Fred Isamanf453a542012-04-20 14:47:54 -04001708static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001709{
Fred Isaman5917ce82011-03-23 13:27:48 +00001710 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001711 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001712 struct nfs_commit_info cinfo;
NeilBrown353db792014-09-24 11:28:32 +10001713 struct nfs_server *nfss;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 while (!list_empty(&data->pages)) {
1716 req = nfs_list_entry(data->pages.next);
1717 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001718 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Niels de Vos1e8968c2013-12-17 18:20:16 +01001720 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001721 req->wb_context->dentry->d_sb->s_id,
David Howells2b0143b2015-03-17 22:25:59 +00001722 (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 req->wb_bytes,
1724 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001725 if (status < 0) {
1726 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001728 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 goto next;
1730 }
1731
1732 /* Okay, COMMIT succeeded, apparently. Check the verifier
1733 * returned by the server against all stored verfs. */
Trond Myklebust2f2c63b2012-06-08 11:56:09 -04001734 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 /* We have a match */
1736 nfs_inode_remove_request(req);
1737 dprintk(" OK\n");
1738 goto next;
1739 }
1740 /* We have a mismatch. Write the page again */
1741 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001742 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001743 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001745 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
NeilBrown353db792014-09-24 11:28:32 +10001747 nfss = NFS_SERVER(data->inode);
1748 if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1749 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1750
Fred Isamanf453a542012-04-20 14:47:54 -04001751 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1752 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1753 nfs_commit_clear_lock(NFS_I(data->inode));
Fred Isaman5917ce82011-03-23 13:27:48 +00001754}
1755
1756static void nfs_commit_release(void *calldata)
1757{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001758 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001759
Fred Isamanf453a542012-04-20 14:47:54 -04001760 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001761 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001763
1764static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001765 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001766 .rpc_call_done = nfs_commit_done,
1767 .rpc_release = nfs_commit_release,
1768};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769
Fred Isamanf453a542012-04-20 14:47:54 -04001770static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1771 .completion = nfs_commit_release_pages,
1772 .error_cleanup = nfs_commit_clear_lock,
1773};
1774
Fred Isaman1763da12012-04-20 14:47:57 -04001775int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1776 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001777{
1778 int status;
1779
Fred Isamanea2cf222012-04-20 14:47:53 -04001780 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001781 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001782 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001783 return status;
1784}
1785
Trond Myklebustb608b282010-07-30 15:31:54 -04001786int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001789 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001790 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001791 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Trond Myklebustb8413f92011-03-21 15:37:01 -04001793 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1794 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001795 goto out_mark_dirty;
Fred Isamanea2cf222012-04-20 14:47:53 -04001796 nfs_init_cinfo_from_inode(&cinfo, inode);
1797 res = nfs_scan_commit(inode, &head, &cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001799 int error;
1800
Fred Isamanea2cf222012-04-20 14:47:53 -04001801 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001802 if (error < 0)
1803 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001804 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001805 goto out_mark_dirty;
NeilBrown74316202014-07-07 15:16:04 +10001806 error = wait_on_bit_action(&NFS_I(inode)->flags,
Trond Myklebustb8413f92011-03-21 15:37:01 -04001807 NFS_INO_COMMIT,
1808 nfs_wait_bit_killable,
1809 TASK_KILLABLE);
1810 if (error < 0)
1811 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001812 } else
1813 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001814 return res;
1815 /* Note: If we exit without ensuring that the commit is complete,
1816 * we must mark the inode as dirty. Otherwise, future calls to
1817 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1818 * that the data is on the disk.
1819 */
1820out_mark_dirty:
1821 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 return res;
1823}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001824
Anna Schumakerae09c312015-07-13 14:01:32 -04001825int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001826{
Trond Myklebust420e3642010-02-19 17:00:02 -08001827 struct nfs_inode *nfsi = NFS_I(inode);
1828 int flags = FLUSH_SYNC;
1829 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001830
Jeff Layton3236c3e2011-10-11 09:49:21 -04001831 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001832 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e2011-10-11 09:49:21 -04001833 return ret;
1834
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001835 if (wbc->sync_mode == WB_SYNC_NONE) {
1836 /* Don't commit yet if this is a non-blocking flush and there
1837 * are a lot of outstanding writes for this mapping.
1838 */
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -05001839 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001840 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001841
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001842 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001843 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001844 }
1845
Trond Myklebust420e3642010-02-19 17:00:02 -08001846 ret = nfs_commit_inode(inode, flags);
1847 if (ret >= 0) {
1848 if (wbc->sync_mode == WB_SYNC_NONE) {
1849 if (ret < wbc->nr_to_write)
1850 wbc->nr_to_write -= ret;
1851 else
1852 wbc->nr_to_write = 0;
1853 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001854 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001855 }
1856out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001857 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1858 return ret;
1859}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001860EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001861
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001862/*
1863 * flush the inode to disk.
1864 */
1865int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001866{
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001867 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001868
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001869 trace_nfs_writeback_inode_enter(inode);
1870
Trond Myklebust5bb89b42015-03-25 14:14:42 -04001871 ret = filemap_write_and_wait(inode->i_mapping);
Chuck Lever6b196872015-05-06 18:26:58 -04001872 if (ret)
1873 goto out;
1874 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1875 if (ret < 0)
1876 goto out;
1877 pnfs_sync_inode(inode, true);
1878 ret = 0;
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001879
Chuck Lever6b196872015-05-06 18:26:58 -04001880out:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001881 trace_nfs_writeback_inode_exit(inode, ret);
1882 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001883}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001884EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001885
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001886int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1887{
1888 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001889 int ret = 0;
1890
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001891 wait_on_page_writeback(page);
1892
1893 /* blocking call to cancel all requests and join to a single (head)
1894 * request */
1895 req = nfs_lock_and_join_requests(page, false);
1896
1897 if (IS_ERR(req)) {
1898 ret = PTR_ERR(req);
1899 } else if (req) {
1900 /* all requests from this page have been cancelled by
1901 * nfs_lock_and_join_requests, so just remove the head
1902 * request from the inode / page_private pointer and
1903 * release it */
1904 nfs_inode_remove_request(req);
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001905 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001906 }
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001907
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001908 return ret;
1909}
1910
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001911/*
1912 * Write back all requests on one page - we do this before reading it.
1913 */
1914int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001915{
Mel Gorman29418aa2012-07-31 16:45:10 -07001916 loff_t range_start = page_file_offset(page);
Trond Myklebust1c759502006-10-09 16:18:38 -04001917 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001918 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001919 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001920 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001921 .range_start = range_start,
1922 .range_end = range_end,
1923 };
1924 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001925
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001926 trace_nfs_writeback_page_enter(inode);
1927
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001928 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001929 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001930 if (clear_page_dirty_for_io(page)) {
1931 ret = nfs_writepage_locked(page, &wbc);
1932 if (ret < 0)
1933 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001934 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001935 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001936 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001937 if (!PagePrivate(page))
1938 break;
1939 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001940 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001941 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001942 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001943out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001944 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001945 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001946}
1947
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001948#ifdef CONFIG_MIGRATION
1949int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001950 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001951{
Jeff Layton2da95652011-10-12 10:57:42 -04001952 /*
1953 * If PagePrivate is set, then the page is currently associated with
1954 * an in-progress read or write request. Don't try to migrate it.
1955 *
1956 * FIXME: we could do this in principle, but we'll need a way to ensure
1957 * that we can safely release the inode reference while holding
1958 * the page lock.
1959 */
1960 if (PagePrivate(page))
1961 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001962
David Howells8c209ce2012-12-05 13:34:49 +00001963 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1964 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001965
Mel Gormana6bc32b2012-01-12 17:19:43 -08001966 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001967}
1968#endif
1969
David Howellsf7b422b2006-06-09 09:34:33 -04001970int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971{
1972 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04001973 sizeof(struct nfs_pgio_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001975 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 if (nfs_wdata_cachep == NULL)
1977 return -ENOMEM;
1978
Matthew Dobson93d23412006-03-26 01:37:50 -08001979 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1980 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001982 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Fred Isaman0b7c0152012-04-20 14:47:39 -04001984 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1985 sizeof(struct nfs_commit_data),
1986 0, SLAB_HWCACHE_ALIGN,
1987 NULL);
1988 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001989 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001990
Matthew Dobson93d23412006-03-26 01:37:50 -08001991 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08001992 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04001994 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995
Peter Zijlstra89a09142007-03-16 13:38:26 -08001996 /*
1997 * NFS congestion size, scale with available memory.
1998 *
1999 * 64MB: 8192k
2000 * 128MB: 11585k
2001 * 256MB: 16384k
2002 * 512MB: 23170k
2003 * 1GB: 32768k
2004 * 2GB: 46340k
2005 * 4GB: 65536k
2006 * 8GB: 92681k
2007 * 16GB: 131072k
2008 *
2009 * This allows larger machines to have larger/more transfers.
2010 * Limit the default to 256M
2011 */
2012 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2013 if (nfs_congestion_kb > 256*1024)
2014 nfs_congestion_kb = 256*1024;
2015
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04002017
2018out_destroy_commit_cache:
2019 kmem_cache_destroy(nfs_cdata_cachep);
2020out_destroy_write_mempool:
2021 mempool_destroy(nfs_wdata_mempool);
2022out_destroy_write_cache:
2023 kmem_cache_destroy(nfs_wdata_cachep);
2024 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025}
2026
David Brownell266bee82006-06-27 12:59:15 -07002027void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028{
2029 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04002030 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002032 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033}
2034
Anna Schumaker4a0de552014-05-06 09:12:30 -04002035static const struct nfs_rw_ops nfs_rw_write_ops = {
Anna Schumakera4cdda52014-05-06 09:12:31 -04002036 .rw_mode = FMODE_WRITE,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002037 .rw_alloc_header = nfs_writehdr_alloc,
2038 .rw_free_header = nfs_writehdr_free,
Anna Schumaker0eecb212014-05-06 09:12:32 -04002039 .rw_done = nfs_writeback_done,
2040 .rw_result = nfs_writeback_result,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04002041 .rw_initiate = nfs_initiate_write,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002042};