blob: 9dafb08ddae533981f149ecebfd83d5f01625cb2 [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
Peng Tao0bcbf032015-12-05 15:57:31 +0800548static void nfs_write_error_remove_page(struct nfs_page *req)
549{
550 nfs_unlock_request(req);
551 nfs_end_page_writeback(req);
552 nfs_release_request(req);
553 generic_error_remove_page(page_file_mapping(req->wb_page),
554 req->wb_page);
555}
556
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400557/*
558 * Find an associated nfs write request, and prepare to flush it out
559 * May return an error if the user signalled nfs_wait_on_request().
560 */
561static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400562 struct page *page, bool nonblock)
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400563{
564 struct nfs_page *req;
565 int ret = 0;
566
Weston Andros Adamsond4581382014-07-11 10:20:48 -0400567 req = nfs_lock_and_join_requests(page, nonblock);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400568 if (!req)
569 goto out;
570 ret = PTR_ERR(req);
571 if (IS_ERR(req))
572 goto out;
573
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400574 nfs_set_page_writeback(page);
575 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400576
Trond Myklebustdeed85e2012-10-15 15:02:01 -0400577 ret = 0;
Fred Isamanf8512ad2008-03-19 11:24:39 -0400578 if (!nfs_pageio_add_request(pgio, req)) {
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400579 ret = pgio->pg_error;
Peng Tao0bcbf032015-12-05 15:57:31 +0800580 /*
581 * Remove the problematic req upon fatal errors,
582 * while other dirty pages can still be around
583 * until they get flushed.
584 */
585 if (nfs_error_is_fatal(ret)) {
586 nfs_context_set_write_error(req->wb_context, ret);
587 nfs_write_error_remove_page(req);
588 } else {
589 nfs_redirty_request(req);
590 ret = -EAGAIN;
591 }
Trond Myklebust40f90272015-10-01 11:36:38 -0400592 } else
593 nfs_add_stats(page_file_mapping(page)->host,
594 NFSIOS_WRITEPAGES, 1);
Trond Myklebust074cc1d2009-08-10 08:54:13 -0400595out:
596 return ret;
Trond Myklebuste261f512006-12-05 00:35:41 -0500597}
598
Trond Myklebustf758c8852007-07-22 19:27:32 -0400599static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
600{
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400601 int ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400602
Mel Gormand56b4dd2012-07-31 16:45:06 -0700603 nfs_pageio_cond_complete(pgio, page_file_index(page));
Wu Fengguang1b430be2010-10-26 14:21:26 -0700604 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
Trond Myklebustcfb506e2010-07-30 15:31:57 -0400605 if (ret == -EAGAIN) {
606 redirty_page_for_writepage(wbc, page);
607 ret = 0;
608 }
609 return ret;
Trond Myklebustf758c8852007-07-22 19:27:32 -0400610}
611
Trond Myklebuste261f512006-12-05 00:35:41 -0500612/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 * Write an mmapped page to the server.
614 */
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500615static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400617 struct nfs_pageio_descriptor pgio;
Trond Myklebust40f90272015-10-01 11:36:38 -0400618 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste261f512006-12-05 00:35:41 -0500619 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Trond Myklebust40f90272015-10-01 11:36:38 -0400621 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
622 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200623 false, &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400624 err = nfs_do_writepage(page, wbc, &pgio);
625 nfs_pageio_complete(&pgio);
626 if (err < 0)
627 return err;
628 if (pgio.pg_error < 0)
629 return pgio.pg_error;
630 return 0;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500631}
632
633int nfs_writepage(struct page *page, struct writeback_control *wbc)
634{
Trond Myklebustf758c8852007-07-22 19:27:32 -0400635 int ret;
Trond Myklebust4d770cc2006-12-05 00:35:41 -0500636
Trond Myklebustf758c8852007-07-22 19:27:32 -0400637 ret = nfs_writepage_locked(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 unlock_page(page);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400639 return ret;
640}
641
642static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
643{
644 int ret;
645
646 ret = nfs_do_writepage(page, wbc, data);
647 unlock_page(page);
648 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
652{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 struct inode *inode = mapping->host;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400654 unsigned long *bitlock = &NFS_I(inode)->flags;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400655 struct nfs_pageio_descriptor pgio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 int err;
657
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400658 /* Stop dirtying of new pages while we sync */
NeilBrown74316202014-07-07 15:16:04 +1000659 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400660 nfs_wait_bit_killable, TASK_KILLABLE);
661 if (err)
662 goto out_err;
663
Chuck Lever91d5b472006-03-20 13:44:14 -0500664 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
665
Christoph Hellwiga20c93e2014-04-16 15:07:21 +0200666 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
667 &nfs_async_write_completion_ops);
Trond Myklebustf758c8852007-07-22 19:27:32 -0400668 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400669 nfs_pageio_complete(&pgio);
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400670
671 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100672 smp_mb__after_atomic();
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400673 wake_up_bit(bitlock, NFS_INO_FLUSHING);
674
Trond Myklebustf758c8852007-07-22 19:27:32 -0400675 if (err < 0)
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400676 goto out_err;
677 err = pgio.pg_error;
678 if (err < 0)
679 goto out_err;
Trond Myklebustc63c7b02007-04-02 19:29:52 -0400680 return 0;
Trond Myklebust72cb77f2009-03-11 14:10:30 -0400681out_err:
682 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
685/*
686 * Insert a write request into an inode
687 */
Fred Isamand6d6dc72012-03-08 17:29:35 -0500688static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
690 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400691
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -0400692 WARN_ON_ONCE(req->wb_this_page != req);
693
Trond Myklebuste7d39062008-06-13 12:12:32 -0400694 /* Lock the request! */
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400695 nfs_lock_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400696
697 spin_lock(&inode->i_lock);
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500698 if (!nfsi->nrequests &&
699 NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
Trond Myklebusta9a4a872011-10-17 16:08:46 -0700700 inode->i_version++;
Mel Gorman29418aa2012-07-31 16:45:10 -0700701 /*
702 * Swap-space should not get truncated. Hence no need to plug the race
703 * with invalidate/truncate.
704 */
705 if (likely(!PageSwapCache(req->wb_page))) {
706 set_bit(PG_MAPPED, &req->wb_flags);
707 SetPagePrivate(req->wb_page);
708 set_page_private(req->wb_page, (unsigned long)req);
709 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500710 nfsi->nrequests++;
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400711 /* this a head request for a page group - mark it as having an
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500712 * extra reference so sub groups can follow suit.
713 * This flag also informs pgio layer when to bump nrequests when
714 * adding subrequests. */
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400715 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
Trond Myklebustc03b4022007-06-17 13:26:38 -0400716 kref_get(&req->wb_kref);
Trond Myklebuste7d39062008-06-13 12:12:32 -0400717 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718}
719
720/*
Peter Zijlstra89a09142007-03-16 13:38:26 -0800721 * Remove a write request from an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 */
723static void nfs_inode_remove_request(struct nfs_page *req)
724{
David Howells2b0143b2015-03-17 22:25:59 +0000725 struct inode *inode = d_inode(req->wb_context->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 struct nfs_inode *nfsi = NFS_I(inode);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400727 struct nfs_page *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400729 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
730 head = req->wb_head;
731
732 spin_lock(&inode->i_lock);
733 if (likely(!PageSwapCache(head->wb_page))) {
734 set_page_private(head->wb_page, 0);
735 ClearPagePrivate(head->wb_page);
NeilBrown95905442014-09-24 11:28:32 +1000736 smp_mb__after_atomic();
737 wake_up_page(head->wb_page, PG_private);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400738 clear_bit(PG_MAPPED, &head->wb_flags);
739 }
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -0500740 nfsi->nrequests--;
741 spin_unlock(&inode->i_lock);
742 } else {
743 spin_lock(&inode->i_lock);
744 nfsi->nrequests--;
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400745 spin_unlock(&inode->i_lock);
Mel Gorman29418aa2012-07-31 16:45:10 -0700746 }
Weston Andros Adamson17089a22014-07-11 10:20:45 -0400747
748 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
749 nfs_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750}
751
Trond Myklebust61822ab2006-12-05 00:35:42 -0500752static void
Fred6d884e82008-03-19 11:24:38 -0400753nfs_mark_request_dirty(struct nfs_page *req)
Trond Myklebust61822ab2006-12-05 00:35:42 -0500754{
Trond Myklebust61822ab2006-12-05 00:35:42 -0500755 __set_page_dirty_nobuffers(req->wb_page);
756}
757
Trond Myklebust3a3908c2014-09-08 22:21:00 -0700758/*
759 * nfs_page_search_commits_for_head_request_locked
760 *
761 * Search through commit lists on @inode for the head request for @page.
762 * Must be called while holding the inode (which is cinfo) lock.
763 *
764 * Returns the head request if found, or NULL if not found.
765 */
766static struct nfs_page *
767nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
768 struct page *page)
769{
770 struct nfs_page *freq, *t;
771 struct nfs_commit_info cinfo;
772 struct inode *inode = &nfsi->vfs_inode;
773
774 nfs_init_cinfo_from_inode(&cinfo, inode);
775
776 /* search through pnfs commit lists */
777 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
778 if (freq)
779 return freq->wb_head;
780
781 /* Linearly search the commit list for the correct request */
782 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
783 if (freq->wb_page == page)
784 return freq->wb_head;
785 }
786
787 return NULL;
788}
789
Trond Myklebust8dd37752012-03-15 17:16:40 -0400790/**
Trond Myklebust86d80f92015-07-31 16:24:30 -0400791 * nfs_request_add_commit_list_locked - add request to a commit list
792 * @req: pointer to a struct nfs_page
793 * @dst: commit list head
794 * @cinfo: holds list lock and accounting info
795 *
796 * This sets the PG_CLEAN bit, updates the cinfo count of
797 * number of outstanding requests requiring a commit as well as
798 * the MM page stats.
799 *
800 * The caller must hold the cinfo->lock, and the nfs_page lock.
801 */
802void
803nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
804 struct nfs_commit_info *cinfo)
805{
806 set_bit(PG_CLEAN, &req->wb_flags);
807 nfs_list_add_request(req, dst);
808 cinfo->mds->ncommit++;
809}
810EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
811
812/**
Trond Myklebust8dd37752012-03-15 17:16:40 -0400813 * nfs_request_add_commit_list - add request to a commit list
814 * @req: pointer to a struct nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400815 * @dst: commit list head
816 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400817 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400818 * This sets the PG_CLEAN bit, updates the cinfo count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400819 * number of outstanding requests requiring a commit as well as
820 * the MM page stats.
821 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400822 * The caller must _not_ hold the cinfo->lock, but must be
Trond Myklebust8dd37752012-03-15 17:16:40 -0400823 * holding the nfs_page lock.
824 */
825void
Fred Isamanea2cf222012-04-20 14:47:53 -0400826nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
827 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400828{
Fred Isamanea2cf222012-04-20 14:47:53 -0400829 spin_lock(cinfo->lock);
Trond Myklebust86d80f92015-07-31 16:24:30 -0400830 nfs_request_add_commit_list_locked(req, dst, cinfo);
Fred Isamanea2cf222012-04-20 14:47:53 -0400831 spin_unlock(cinfo->lock);
Trond Myklebust86d80f92015-07-31 16:24:30 -0400832 nfs_mark_page_unstable(req->wb_page, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400833}
834EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
835
836/**
837 * nfs_request_remove_commit_list - Remove request from a commit list
838 * @req: pointer to a nfs_page
Fred Isamanea2cf222012-04-20 14:47:53 -0400839 * @cinfo: holds list lock and accounting info
Trond Myklebust8dd37752012-03-15 17:16:40 -0400840 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400841 * This clears the PG_CLEAN bit, and updates the cinfo's count of
Trond Myklebust8dd37752012-03-15 17:16:40 -0400842 * number of outstanding requests requiring a commit
843 * It does not update the MM page stats.
844 *
Fred Isamanea2cf222012-04-20 14:47:53 -0400845 * The caller _must_ hold the cinfo->lock and the nfs_page lock.
Trond Myklebust8dd37752012-03-15 17:16:40 -0400846 */
847void
Fred Isamanea2cf222012-04-20 14:47:53 -0400848nfs_request_remove_commit_list(struct nfs_page *req,
849 struct nfs_commit_info *cinfo)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400850{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400851 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
852 return;
853 nfs_list_remove_request(req);
Fred Isamanea2cf222012-04-20 14:47:53 -0400854 cinfo->mds->ncommit--;
Trond Myklebust8dd37752012-03-15 17:16:40 -0400855}
856EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
857
Fred Isamanea2cf222012-04-20 14:47:53 -0400858static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
859 struct inode *inode)
860{
861 cinfo->lock = &inode->i_lock;
862 cinfo->mds = &NFS_I(inode)->commit_info;
863 cinfo->ds = pnfs_get_ds_info(inode);
Fred Isamanb359f9d2012-04-20 14:47:55 -0400864 cinfo->dreq = NULL;
Fred Isamanf453a542012-04-20 14:47:54 -0400865 cinfo->completion_ops = &nfs_commit_completion_ops;
Fred Isamanea2cf222012-04-20 14:47:53 -0400866}
867
868void nfs_init_cinfo(struct nfs_commit_info *cinfo,
869 struct inode *inode,
870 struct nfs_direct_req *dreq)
871{
Fred Isaman1763da12012-04-20 14:47:57 -0400872 if (dreq)
873 nfs_init_cinfo_from_dreq(cinfo, dreq);
874 else
875 nfs_init_cinfo_from_inode(cinfo, inode);
Fred Isamanea2cf222012-04-20 14:47:53 -0400876}
877EXPORT_SYMBOL_GPL(nfs_init_cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879/*
880 * Add a request to the inode's commit list.
881 */
Fred Isaman1763da12012-04-20 14:47:57 -0400882void
Fred Isamanea2cf222012-04-20 14:47:53 -0400883nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400884 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400886 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
Trond Myklebust8dd37752012-03-15 17:16:40 -0400887 return;
Fred Isamanea2cf222012-04-20 14:47:53 -0400888 nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889}
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400890
Fred Isamand6d6dc72012-03-08 17:29:35 -0500891static void
892nfs_clear_page_commit(struct page *page)
893{
894 dec_zone_page_state(page, NR_UNSTABLE_NFS);
Tejun Heo93f78d82015-05-22 17:13:27 -0400895 dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
896 WB_RECLAIMABLE);
Fred Isamand6d6dc72012-03-08 17:29:35 -0500897}
898
Weston Andros Adamson411a99a2014-07-17 20:42:19 -0400899/* Called holding inode (/cinfo) lock */
Trond Myklebust8dd37752012-03-15 17:16:40 -0400900static void
Trond Myklebuste468bae2008-06-13 13:25:22 -0400901nfs_clear_request_commit(struct nfs_page *req)
902{
Trond Myklebust8dd37752012-03-15 17:16:40 -0400903 if (test_bit(PG_CLEAN, &req->wb_flags)) {
David Howells2b0143b2015-03-17 22:25:59 +0000904 struct inode *inode = d_inode(req->wb_context->dentry);
Fred Isamanea2cf222012-04-20 14:47:53 -0400905 struct nfs_commit_info cinfo;
Trond Myklebuste468bae2008-06-13 13:25:22 -0400906
Fred Isamanea2cf222012-04-20 14:47:53 -0400907 nfs_init_cinfo_from_inode(&cinfo, inode);
908 if (!pnfs_clear_request_commit(req, &cinfo)) {
Fred Isamanea2cf222012-04-20 14:47:53 -0400909 nfs_request_remove_commit_list(req, &cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400910 }
Fred Isamand6d6dc72012-03-08 17:29:35 -0500911 nfs_clear_page_commit(req->wb_page);
Trond Myklebuste468bae2008-06-13 13:25:22 -0400912 }
Trond Myklebuste468bae2008-06-13 13:25:22 -0400913}
914
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400915int nfs_write_need_commit(struct nfs_pgio_header *hdr)
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400916{
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400917 if (hdr->verf.committed == NFS_DATA_SYNC)
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -0400918 return hdr->lseg == NULL;
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400919 return hdr->verf.committed != NFS_FILE_SYNC;
Trond Myklebust8e821ca2007-04-20 16:12:34 -0400920}
921
Fred Isaman061ae2e2012-04-20 14:47:48 -0400922static void nfs_write_completion(struct nfs_pgio_header *hdr)
Fred Isaman6c75dc02012-04-20 14:47:47 -0400923{
Fred Isamanea2cf222012-04-20 14:47:53 -0400924 struct nfs_commit_info cinfo;
Fred Isaman6c75dc02012-04-20 14:47:47 -0400925 unsigned long bytes = 0;
926
927 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
928 goto out;
Fred Isamanea2cf222012-04-20 14:47:53 -0400929 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400930 while (!list_empty(&hdr->pages)) {
931 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400932
933 bytes += req->wb_bytes;
934 nfs_list_remove_request(req);
935 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
936 (hdr->good_bytes < bytes)) {
Trond Myklebustd1182b32012-05-09 13:37:43 -0400937 nfs_set_pageerror(req->wb_page);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400938 nfs_context_set_write_error(req->wb_context, hdr->error);
939 goto remove_req;
940 }
Weston Andros Adamsonc65e6252014-06-09 11:48:36 -0400941 if (nfs_write_need_commit(hdr)) {
Anna Schumakerf79d06f2014-05-06 09:12:28 -0400942 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -0400943 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -0400944 hdr->pgio_mirror_idx);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400945 goto next;
946 }
947remove_req:
948 nfs_inode_remove_request(req);
949next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -0400950 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -0400951 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -0400952 nfs_release_request(req);
Fred Isaman6c75dc02012-04-20 14:47:47 -0400953 }
954out:
955 hdr->release(hdr);
956}
957
Anna Schumakerce595152014-05-06 09:12:34 -0400958unsigned long
Fred Isamanea2cf222012-04-20 14:47:53 -0400959nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400960{
Fred Isamanea2cf222012-04-20 14:47:53 -0400961 return cinfo->mds->ncommit;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -0400962}
963
Fred Isamanea2cf222012-04-20 14:47:53 -0400964/* cinfo->lock held by caller */
Fred Isaman1763da12012-04-20 14:47:57 -0400965int
Fred Isamanea2cf222012-04-20 14:47:53 -0400966nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
967 struct nfs_commit_info *cinfo, int max)
Fred Isamand6d6dc72012-03-08 17:29:35 -0500968{
969 struct nfs_page *req, *tmp;
970 int ret = 0;
971
972 list_for_each_entry_safe(req, tmp, src, wb_list) {
Trond Myklebust8dd37752012-03-15 17:16:40 -0400973 if (!nfs_lock_request(req))
974 continue;
Trond Myklebust7ad84aa2012-05-09 13:19:15 -0400975 kref_get(&req->wb_kref);
Fred Isamanea2cf222012-04-20 14:47:53 -0400976 if (cond_resched_lock(cinfo->lock))
Trond Myklebust3b3be882012-03-17 11:59:30 -0400977 list_safe_reset_next(req, tmp, wb_list);
Fred Isamanea2cf222012-04-20 14:47:53 -0400978 nfs_request_remove_commit_list(req, cinfo);
Trond Myklebust8dd37752012-03-15 17:16:40 -0400979 nfs_list_add_request(req, dst);
980 ret++;
Fred Isaman1763da12012-04-20 14:47:57 -0400981 if ((ret == max) && !cinfo->dreq)
Trond Myklebust8dd37752012-03-15 17:16:40 -0400982 break;
Fred Isamand6d6dc72012-03-08 17:29:35 -0500983 }
984 return ret;
985}
Fred Isamand6d6dc72012-03-08 17:29:35 -0500986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987/*
988 * nfs_scan_commit - Scan an inode for commit requests
989 * @inode: NFS inode to scan
Fred Isamanea2cf222012-04-20 14:47:53 -0400990 * @dst: mds destination list
991 * @cinfo: mds and ds lists of reqs ready to commit
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 *
993 * Moves requests from the inode's 'commit' request list.
994 * The requests are *not* checked to ensure that they form a contiguous set.
995 */
Fred Isaman1763da12012-04-20 14:47:57 -0400996int
Fred Isamanea2cf222012-04-20 14:47:53 -0400997nfs_scan_commit(struct inode *inode, struct list_head *dst,
998 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
Fred Isamand6d6dc72012-03-08 17:29:35 -05001000 int ret = 0;
Trond Myklebustfb8a1f12009-03-11 14:10:29 -04001001
Fred Isamanea2cf222012-04-20 14:47:53 -04001002 spin_lock(cinfo->lock);
1003 if (cinfo->mds->ncommit > 0) {
Trond Myklebust8dd37752012-03-15 17:16:40 -04001004 const int max = INT_MAX;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001005
Fred Isamanea2cf222012-04-20 14:47:53 -04001006 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1007 cinfo, max);
1008 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001009 }
Fred Isamanea2cf222012-04-20 14:47:53 -04001010 spin_unlock(cinfo->lock);
Trond Myklebustff778d02010-02-19 16:53:39 -08001011 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
Fred Isamand6d6dc72012-03-08 17:29:35 -05001013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014/*
Trond Myklebuste7d39062008-06-13 12:12:32 -04001015 * Search for an existing write request, and attempt to update
1016 * it to reflect a new dirty region on a given page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 *
Trond Myklebuste7d39062008-06-13 12:12:32 -04001018 * If the attempt fails, then the existing request is flushed out
1019 * to disk.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 */
Trond Myklebuste7d39062008-06-13 12:12:32 -04001021static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1022 struct page *page,
1023 unsigned int offset,
1024 unsigned int bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
Trond Myklebuste7d39062008-06-13 12:12:32 -04001026 struct nfs_page *req;
1027 unsigned int rqend;
1028 unsigned int end;
1029 int error;
1030
1031 if (!PagePrivate(page))
1032 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 end = offset + bytes;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001035 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 for (;;) {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001038 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001039 if (req == NULL)
1040 goto out_unlock;
Trond Myklebust277459d2006-12-05 00:35:35 -05001041
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001042 /* should be handled by nfs_flush_incompatible */
1043 WARN_ON_ONCE(req->wb_head != req);
1044 WARN_ON_ONCE(req->wb_this_page != req);
1045
Trond Myklebuste7d39062008-06-13 12:12:32 -04001046 rqend = req->wb_offset + req->wb_bytes;
1047 /*
1048 * Tell the caller to flush out the request if
1049 * the offsets are non-contiguous.
1050 * Note: nfs_flush_incompatible() will already
1051 * have flushed out requests having wrong owners.
1052 */
Trond Myklebuste468bae2008-06-13 13:25:22 -04001053 if (offset > rqend
Trond Myklebuste7d39062008-06-13 12:12:32 -04001054 || end < req->wb_offset)
1055 goto out_flushme;
1056
Trond Myklebust7ad84aa2012-05-09 13:19:15 -04001057 if (nfs_lock_request(req))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
Trond Myklebuste7d39062008-06-13 12:12:32 -04001060 /* The request is locked, so wait and then retry */
Trond Myklebust587142f2007-07-02 09:57:54 -04001061 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001062 error = nfs_wait_on_request(req);
1063 nfs_release_request(req);
1064 if (error != 0)
1065 goto out_err;
1066 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 }
1068
1069 /* Okay, the request matches. Update the region */
1070 if (offset < req->wb_offset) {
1071 req->wb_offset = offset;
1072 req->wb_pgbase = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (end > rqend)
1075 req->wb_bytes = end - req->wb_offset;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001076 else
1077 req->wb_bytes = rqend - req->wb_offset;
1078out_unlock:
Fred Isamanca138f32012-04-05 15:26:36 -04001079 if (req)
1080 nfs_clear_request_commit(req);
Weston Andros Adamson411a99a2014-07-17 20:42:19 -04001081 spin_unlock(&inode->i_lock);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001082 return req;
1083out_flushme:
1084 spin_unlock(&inode->i_lock);
1085 nfs_release_request(req);
1086 error = nfs_wb_page(inode, page);
1087out_err:
1088 return ERR_PTR(error);
1089}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Trond Myklebuste7d39062008-06-13 12:12:32 -04001091/*
1092 * Try to update an existing write request, or create one if there is none.
1093 *
1094 * Note: Should always be called with the Page Lock held to prevent races
1095 * if we have to add a new request. Also assumes that the caller has
1096 * already called nfs_flush_incompatible() if necessary.
1097 */
1098static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1099 struct page *page, unsigned int offset, unsigned int bytes)
1100{
Mel Gormand56b4dd2012-07-31 16:45:06 -07001101 struct inode *inode = page_file_mapping(page)->host;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001102 struct nfs_page *req;
Trond Myklebuste7d39062008-06-13 12:12:32 -04001103
1104 req = nfs_try_to_update_request(inode, page, offset, bytes);
1105 if (req != NULL)
1106 goto out;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001107 req = nfs_create_request(ctx, page, NULL, offset, bytes);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001108 if (IS_ERR(req))
1109 goto out;
Fred Isamand6d6dc72012-03-08 17:29:35 -05001110 nfs_inode_add_request(inode, req);
Trond Myklebustefc91ed2008-06-10 18:31:00 -04001111out:
Trond Myklebust61e930a2007-10-18 17:08:05 -04001112 return req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113}
1114
Trond Myklebuste7d39062008-06-13 12:12:32 -04001115static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1116 unsigned int offset, unsigned int count)
1117{
1118 struct nfs_page *req;
1119
1120 req = nfs_setup_write_request(ctx, page, offset, count);
1121 if (IS_ERR(req))
1122 return PTR_ERR(req);
1123 /* Update file length */
1124 nfs_grow_file(page, offset, count);
Weston Andros Adamsond72ddcb2014-05-15 11:56:48 -04001125 nfs_mark_uptodate(req);
Trond Myklebusta6305dd2010-04-09 19:07:08 -04001126 nfs_mark_request_dirty(req);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001127 nfs_unlock_and_release_request(req);
Trond Myklebuste7d39062008-06-13 12:12:32 -04001128 return 0;
1129}
1130
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131int nfs_flush_incompatible(struct file *file, struct page *page)
1132{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001133 struct nfs_open_context *ctx = nfs_file_open_context(file);
Trond Myklebust2a369152012-08-13 18:54:45 -04001134 struct nfs_lock_context *l_ctx;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001135 struct file_lock_context *flctx = file_inode(file)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 struct nfs_page *req;
Trond Myklebust1a545332006-12-05 00:35:40 -05001137 int do_flush, status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 /*
1139 * Look for a request corresponding to this page. If there
1140 * is one, and it belongs to another file, we flush it out
1141 * before we try to copy anything into the page. Do this
1142 * due to the lack of an ACCESS-type call in NFSv2.
1143 * Also do the same if we find a request from an existing
1144 * dropped page.
1145 */
Trond Myklebust1a545332006-12-05 00:35:40 -05001146 do {
Weston Andros Adamson84d3a9a2014-07-11 10:20:47 -04001147 req = nfs_page_find_head_request(page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001148 if (req == NULL)
1149 return 0;
Trond Myklebust2a369152012-08-13 18:54:45 -04001150 l_ctx = req->wb_lock_context;
1151 do_flush = req->wb_page != page || req->wb_context != ctx;
Weston Andros Adamson2bfc6e52014-05-15 11:56:45 -04001152 /* for now, flush if more than 1 request in page_group */
1153 do_flush |= req->wb_this_page != req;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001154 if (l_ctx && flctx &&
1155 !(list_empty_careful(&flctx->flc_posix) &&
1156 list_empty_careful(&flctx->flc_flock))) {
Jeff Layton5263e312015-01-16 15:05:55 -05001157 do_flush |= l_ctx->lockowner.l_owner != current->files
1158 || l_ctx->lockowner.l_pid != current->tgid;
1159 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 nfs_release_request(req);
Trond Myklebust1a545332006-12-05 00:35:40 -05001161 if (!do_flush)
1162 return 0;
Mel Gormand56b4dd2012-07-31 16:45:06 -07001163 status = nfs_wb_page(page_file_mapping(page)->host, page);
Trond Myklebust1a545332006-12-05 00:35:40 -05001164 } while (status == 0);
1165 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166}
1167
1168/*
Andy Adamsondc248262013-08-14 11:59:16 -04001169 * Avoid buffered writes when a open context credential's key would
1170 * expire soon.
1171 *
1172 * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1173 *
1174 * Return 0 and set a credential flag which triggers the inode to flush
1175 * and performs NFS_FILE_SYNC writes if the key will expired within
1176 * RPC_KEY_EXPIRE_TIMEO.
1177 */
1178int
1179nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1180{
1181 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1182 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1183
1184 return rpcauth_key_timeout_notify(auth, ctx->cred);
1185}
1186
1187/*
1188 * Test if the open context credential key is marked to expire soon.
1189 */
1190bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1191{
1192 return rpcauth_cred_key_to_expire(ctx->cred);
1193}
1194
1195/*
Trond Myklebust5d47a352008-02-07 17:24:07 -05001196 * If the page cache is marked as unsafe or invalid, then we can't rely on
1197 * the PageUptodate() flag. In this case, we will need to turn off
1198 * write optimisations that depend on the page contents being correct.
1199 */
Trond Myklebust8d197a52012-04-29 12:50:01 -04001200static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
Trond Myklebust5d47a352008-02-07 17:24:07 -05001201{
Jeff Laytond529ef82014-01-27 13:46:15 -05001202 struct nfs_inode *nfsi = NFS_I(inode);
1203
Trond Myklebust8d197a52012-04-29 12:50:01 -04001204 if (nfs_have_delegated_attributes(inode))
1205 goto out;
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001206 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
Jeff Laytond529ef82014-01-27 13:46:15 -05001207 return false;
Jeff Layton4db72b42014-01-28 13:47:46 -05001208 smp_rmb();
Jeff Laytond529ef82014-01-27 13:46:15 -05001209 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
Trond Myklebust8d197a52012-04-29 12:50:01 -04001210 return false;
1211out:
Scott Mayhew18dd78c2014-06-20 08:44:42 -04001212 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1213 return false;
Trond Myklebust8d197a52012-04-29 12:50:01 -04001214 return PageUptodate(page) != 0;
Trond Myklebust5d47a352008-02-07 17:24:07 -05001215}
1216
Jeff Layton5263e312015-01-16 15:05:55 -05001217static bool
1218is_whole_file_wrlock(struct file_lock *fl)
1219{
1220 return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1221 fl->fl_type == F_WRLCK;
1222}
1223
Scott Mayhewc7559662013-07-05 17:33:19 -04001224/* If we know the page is up to date, and we're not using byte range locks (or
1225 * if we have the whole file locked for writing), it may be more efficient to
1226 * extend the write to cover the entire page in order to avoid fragmentation
1227 * inefficiencies.
1228 *
Scott Mayhew263b4502014-01-17 15:12:05 -05001229 * If the file is opened for synchronous writes then we can just skip the rest
1230 * of the checks.
Scott Mayhewc7559662013-07-05 17:33:19 -04001231 */
1232static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1233{
Jeff Layton5263e312015-01-16 15:05:55 -05001234 int ret;
1235 struct file_lock_context *flctx = inode->i_flctx;
1236 struct file_lock *fl;
1237
Scott Mayhewc7559662013-07-05 17:33:19 -04001238 if (file->f_flags & O_DSYNC)
1239 return 0;
Scott Mayhew263b4502014-01-17 15:12:05 -05001240 if (!nfs_write_pageuptodate(page, inode))
1241 return 0;
Scott Mayhewc7559662013-07-05 17:33:19 -04001242 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1243 return 1;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001244 if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1245 list_empty_careful(&flctx->flc_posix)))
Trond Myklebust8fa45922015-10-01 18:38:27 -04001246 return 1;
Jeff Layton5263e312015-01-16 15:05:55 -05001247
1248 /* Check to see if there are whole file write locks */
Jeff Layton5263e312015-01-16 15:05:55 -05001249 ret = 0;
Jeff Layton6109c852015-01-16 15:05:57 -05001250 spin_lock(&flctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001251 if (!list_empty(&flctx->flc_posix)) {
1252 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1253 fl_list);
1254 if (is_whole_file_wrlock(fl))
1255 ret = 1;
1256 } else if (!list_empty(&flctx->flc_flock)) {
Jeff Layton5263e312015-01-16 15:05:55 -05001257 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1258 fl_list);
1259 if (fl->fl_type == F_WRLCK)
1260 ret = 1;
1261 }
Jeff Layton6109c852015-01-16 15:05:57 -05001262 spin_unlock(&flctx->flc_lock);
Jeff Layton5263e312015-01-16 15:05:55 -05001263 return ret;
Scott Mayhewc7559662013-07-05 17:33:19 -04001264}
1265
Trond Myklebust5d47a352008-02-07 17:24:07 -05001266/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 * Update and possibly write a cached page of an NFS file.
1268 *
1269 * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1270 * things with a page scheduled for an RPC call (e.g. invalidate it).
1271 */
1272int nfs_updatepage(struct file *file, struct page *page,
1273 unsigned int offset, unsigned int count)
1274{
Trond Myklebustcd3758e2007-08-10 17:44:32 -04001275 struct nfs_open_context *ctx = nfs_file_open_context(file);
Mel Gormand56b4dd2012-07-31 16:45:06 -07001276 struct inode *inode = page_file_mapping(page)->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 int status = 0;
1278
Chuck Lever91d5b472006-03-20 13:44:14 -05001279 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1280
Al Viro6de14722013-09-16 10:53:17 -04001281 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1282 file, count, (long long)(page_file_offset(page) + offset));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Scott Mayhewc7559662013-07-05 17:33:19 -04001284 if (nfs_can_extend_write(file, page, inode)) {
Trond Myklebust49a70f22006-12-05 00:35:38 -05001285 count = max(count + offset, nfs_page_length(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288
Trond Myklebuste21195a2006-12-05 00:35:39 -05001289 status = nfs_writepage_setup(ctx, page, offset, count);
Trond Myklebust03fa9e82008-06-05 16:02:35 -04001290 if (status < 0)
1291 nfs_set_pageerror(page);
Trond Myklebust59b7c052011-10-17 18:22:55 -07001292 else
1293 __set_page_dirty_nobuffers(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
Chuck Lever48186c72008-06-11 17:56:05 -04001295 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 status, (long long)i_size_read(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 return status;
1298}
1299
Trond Myklebust3ff75762007-07-14 15:40:00 -04001300static int flush_task_priority(int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301{
1302 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1303 case FLUSH_HIGHPRI:
1304 return RPC_PRIORITY_HIGH;
1305 case FLUSH_LOWPRI:
1306 return RPC_PRIORITY_LOW;
1307 }
1308 return RPC_PRIORITY_NORMAL;
1309}
1310
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001311static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1312 struct rpc_message *msg,
Tom Haynesabde71f2014-06-09 13:12:20 -07001313 const struct nfs_rpc_ops *rpc_ops,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001314 struct rpc_task_setup *task_setup_data, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Trond Myklebust3ff75762007-07-14 15:40:00 -04001316 int priority = flush_task_priority(how);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
Anna Schumaker1ed26f32014-05-06 09:12:37 -04001318 task_setup_data->priority = priority;
Tom Haynesabde71f2014-06-09 13:12:20 -07001319 rpc_ops->write_setup(hdr, msg);
Andy Adamsond138d5d2011-03-03 15:13:41 +00001320
Tom Haynesabde71f2014-06-09 13:12:20 -07001321 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001322 &task_setup_data->rpc_client, msg, hdr);
Trond Myklebust275acaa2011-07-12 13:42:02 -04001323}
1324
Fred6d884e82008-03-19 11:24:38 -04001325/* If a nfs_flush_* function fails, it should remove reqs from @head and
1326 * call this on each, which will prepare them to be retried on next
1327 * writeback using standard nfs.
1328 */
1329static void nfs_redirty_request(struct nfs_page *req)
1330{
1331 nfs_mark_request_dirty(req);
Trond Myklebustc7070112015-06-17 19:56:22 -04001332 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001333 nfs_unlock_request(req);
Weston Andros Adamson20633f02014-05-15 11:56:47 -04001334 nfs_end_page_writeback(req);
Trond Myklebust3aff4eb2012-05-09 14:30:35 -04001335 nfs_release_request(req);
Fred6d884e82008-03-19 11:24:38 -04001336}
1337
Fred Isaman061ae2e2012-04-20 14:47:48 -04001338static void nfs_async_write_error(struct list_head *head)
Fred Isaman6c75dc02012-04-20 14:47:47 -04001339{
1340 struct nfs_page *req;
1341
1342 while (!list_empty(head)) {
1343 req = nfs_list_entry(head->next);
1344 nfs_list_remove_request(req);
1345 nfs_redirty_request(req);
1346 }
1347}
1348
Fred Isaman061ae2e2012-04-20 14:47:48 -04001349static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1350 .error_cleanup = nfs_async_write_error,
1351 .completion = nfs_write_completion,
1352};
1353
Bryan Schumaker57208fa2012-06-20 15:53:48 -04001354void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001355 struct inode *inode, int ioflags, bool force_mds,
Fred Isaman061ae2e2012-04-20 14:47:48 -04001356 const struct nfs_pgio_completion_ops *compl_ops)
Trond Myklebust1751c362011-06-10 13:30:23 -04001357{
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001358 struct nfs_server *server = NFS_SERVER(inode);
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001359 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
Christoph Hellwiga20c93e2014-04-16 15:07:21 +02001360
1361#ifdef CONFIG_NFS_V4_1
1362 if (server->pnfs_curr_ld && !force_mds)
1363 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1364#endif
Anna Schumaker4a0de552014-05-06 09:12:30 -04001365 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1366 server->wsize, ioflags);
Trond Myklebust1751c362011-06-10 13:30:23 -04001367}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001368EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
Trond Myklebust1751c362011-06-10 13:30:23 -04001369
Trond Myklebustdce81292011-07-13 15:59:19 -04001370void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1371{
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001372 struct nfs_pgio_mirror *mirror;
1373
Kinglong Mee6f29b9b2015-09-20 23:03:28 +08001374 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1375 pgio->pg_ops->pg_cleanup(pgio);
1376
Anna Schumaker41d8d5b2014-05-06 09:12:40 -04001377 pgio->pg_ops = &nfs_pgio_rw_ops;
Weston Andros Adamsona7d42dd2014-09-19 10:55:07 -04001378
1379 nfs_pageio_stop_mirroring(pgio);
1380
1381 mirror = &pgio->pg_mirrors[0];
1382 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
Trond Myklebustdce81292011-07-13 15:59:19 -04001383}
Trond Myklebust1f945352011-07-13 15:59:57 -04001384EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
Trond Myklebustdce81292011-07-13 15:59:19 -04001385
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Fred Isaman0b7c0152012-04-20 14:47:39 -04001387void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1388{
1389 struct nfs_commit_data *data = calldata;
1390
1391 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1392}
1393
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001394/*
1395 * Special version of should_remove_suid() that ignores capabilities.
1396 */
1397static int nfs_should_remove_suid(const struct inode *inode)
1398{
1399 umode_t mode = inode->i_mode;
1400 int kill = 0;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001401
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001402 /* suid always must be killed */
1403 if (unlikely(mode & S_ISUID))
1404 kill = ATTR_KILL_SUID;
1405
1406 /*
1407 * sgid without any exec bits is just a mandatory locking mark; leave
1408 * it alone. If some exec bits are set, it's a real sgid; kill it.
1409 */
1410 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1411 kill |= ATTR_KILL_SGID;
1412
1413 if (unlikely(kill && S_ISREG(mode)))
1414 return kill;
1415
1416 return 0;
1417}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001418
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001419static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1420 struct nfs_fattr *fattr)
1421{
1422 struct nfs_pgio_args *argp = &hdr->args;
1423 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001424 u64 size = argp->offset + resp->count;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001425
1426 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001427 fattr->size = size;
1428 if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1429 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001430 return;
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001431 }
1432 if (size != fattr->size)
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001433 return;
1434 /* Set attribute barrier */
1435 nfs_fattr_set_barrier(fattr);
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001436 /* ...and update size */
1437 fattr->valid |= NFS_ATTR_FATTR_SIZE;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001438}
1439
1440void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1441{
Trond Myklebust2b83d3d2015-07-05 20:06:38 -04001442 struct nfs_fattr *fattr = &hdr->fattr;
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001443 struct inode *inode = hdr->inode;
1444
Trond Myklebusta08a8cd2015-02-26 17:36:09 -05001445 spin_lock(&inode->i_lock);
1446 nfs_writeback_check_extend(hdr, fattr);
1447 nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1448 spin_unlock(&inode->i_lock);
1449}
1450EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452/*
1453 * This function is called when the WRITE call is complete.
1454 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001455static int nfs_writeback_done(struct rpc_task *task,
1456 struct nfs_pgio_header *hdr,
Anna Schumaker0eecb212014-05-06 09:12:32 -04001457 struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Trond Myklebust788e7a82006-03-20 13:44:27 -05001459 int status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Chuck Leverf551e442006-09-20 14:33:04 -04001461 /*
1462 * ->write_done will attempt to use post-op attributes to detect
1463 * conflicting writes by other clients. A strict interpretation
1464 * of close-to-open would allow us to continue caching even if
1465 * another writer had changed the file, but some applications
1466 * depend on tighter cache coherency when writing.
1467 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001468 status = NFS_PROTO(inode)->write_done(task, hdr);
Trond Myklebust788e7a82006-03-20 13:44:27 -05001469 if (status != 0)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001470 return status;
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001471 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
Chuck Lever91d5b472006-03-20 13:44:14 -05001472
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001473 if (hdr->res.verf->committed < hdr->args.stable &&
1474 task->tk_status >= 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 /* We tried a write call, but the server did not
1476 * commit data to stable storage even though we
1477 * requested it.
1478 * Note: There is a known bug in Tru64 < 5.0 in which
1479 * the server reports NFS_DATA_SYNC, but performs
1480 * NFS_FILE_SYNC. We therefore implement this checking
1481 * as a dprintk() in order to avoid filling syslog.
1482 */
1483 static unsigned long complain;
1484
Fred Isamana69aef12011-03-03 15:13:47 +00001485 /* Note this will print the MDS for a DS write */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (time_before(complain, jiffies)) {
Chuck Lever48186c72008-06-11 17:56:05 -04001487 dprintk("NFS: faulty NFS server %s:"
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 " (committed = %d) != (stable = %d)\n",
Fred Isamancd841602012-04-20 14:47:44 -04001489 NFS_SERVER(inode)->nfs_client->cl_hostname,
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001490 hdr->res.verf->committed, hdr->args.stable);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 complain = jiffies + 300 * HZ;
1492 }
1493 }
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001494
1495 /* Deal with the suid/sgid bit corner case */
1496 if (nfs_should_remove_suid(inode))
1497 nfs_mark_for_revalidate(inode);
Anna Schumaker0eecb212014-05-06 09:12:32 -04001498 return 0;
1499}
1500
1501/*
1502 * This function is called when the WRITE call is complete.
1503 */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001504static void nfs_writeback_result(struct rpc_task *task,
1505 struct nfs_pgio_header *hdr)
Anna Schumaker0eecb212014-05-06 09:12:32 -04001506{
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001507 struct nfs_pgio_args *argp = &hdr->args;
1508 struct nfs_pgio_res *resp = &hdr->res;
Trond Myklebust1f2edbe2014-04-13 11:11:31 -04001509
1510 if (resp->count < argp->count) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 static unsigned long complain;
1512
Fred Isaman6c75dc02012-04-20 14:47:47 -04001513 /* This a short write! */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001514 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
Chuck Lever91d5b472006-03-20 13:44:14 -05001515
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /* Has the server at least made some progress? */
Fred Isaman6c75dc02012-04-20 14:47:47 -04001517 if (resp->count == 0) {
1518 if (time_before(complain, jiffies)) {
1519 printk(KERN_WARNING
1520 "NFS: Server wrote zero bytes, expected %u.\n",
1521 argp->count);
1522 complain = jiffies + 300 * HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 }
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001524 nfs_set_pgio_error(hdr, -EIO, argp->offset);
Fred Isaman6c75dc02012-04-20 14:47:47 -04001525 task->tk_status = -EIO;
Fred Isaman13602892011-02-11 15:42:38 +00001526 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 }
Kinglong Meef8417b482015-10-16 17:23:29 +08001528
1529 /* For non rpc-based layout drivers, retry-through-MDS */
1530 if (!task->tk_ops) {
1531 hdr->pnfs_error = -EAGAIN;
1532 return;
1533 }
1534
Fred Isaman6c75dc02012-04-20 14:47:47 -04001535 /* Was this an NFSv2 write or an NFSv3 stable write? */
1536 if (resp->verf->committed != NFS_UNSTABLE) {
1537 /* Resend from where the server left off */
Weston Andros Adamsond45f60c2014-06-09 11:48:35 -04001538 hdr->mds_offset += resp->count;
Fred Isaman6c75dc02012-04-20 14:47:47 -04001539 argp->offset += resp->count;
1540 argp->pgbase += resp->count;
1541 argp->count -= resp->count;
1542 } else {
1543 /* Resend as a stable write in order to avoid
1544 * headaches in the case of a server crash.
1545 */
1546 argp->stable = NFS_FILE_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
Fred Isaman6c75dc02012-04-20 14:47:47 -04001548 rpc_restart_call_prepare(task);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550}
1551
1552
Trond Myklebust71d0a612010-04-22 15:35:57 -04001553static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1554{
Trond Myklebustb8413f92011-03-21 15:37:01 -04001555 int ret;
1556
Trond Myklebust71d0a612010-04-22 15:35:57 -04001557 if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1558 return 1;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001559 if (!may_wait)
1560 return 0;
1561 ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1562 NFS_INO_COMMIT,
1563 nfs_wait_bit_killable,
1564 TASK_KILLABLE);
1565 return (ret < 0) ? ret : 1;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001566}
1567
Fred Isamanf453a542012-04-20 14:47:54 -04001568static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
Trond Myklebust71d0a612010-04-22 15:35:57 -04001569{
1570 clear_bit(NFS_INO_COMMIT, &nfsi->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001571 smp_mb__after_atomic();
Trond Myklebust71d0a612010-04-22 15:35:57 -04001572 wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1573}
1574
Fred Isaman0b7c0152012-04-20 14:47:39 -04001575void nfs_commitdata_release(struct nfs_commit_data *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001577 put_nfs_open_context(data->context);
1578 nfs_commit_free(data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
Fred Isamane0c2b382011-03-23 13:27:53 +00001580EXPORT_SYMBOL_GPL(nfs_commitdata_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Fred Isaman0b7c0152012-04-20 14:47:39 -04001582int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
Peng Taoc36aae92014-06-09 07:10:14 +08001583 const struct nfs_rpc_ops *nfs_ops,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001584 const struct rpc_call_ops *call_ops,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001585 int how, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
Trond Myklebust07737692007-10-25 18:42:54 -04001587 struct rpc_task *task;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001588 int priority = flush_task_priority(how);
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001589 struct rpc_message msg = {
1590 .rpc_argp = &data->args,
1591 .rpc_resp = &data->res,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001592 .rpc_cred = data->cred,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001593 };
Trond Myklebust84115e12007-07-14 15:39:59 -04001594 struct rpc_task_setup task_setup_data = {
Trond Myklebust07737692007-10-25 18:42:54 -04001595 .task = &data->task,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001596 .rpc_client = clnt,
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001597 .rpc_message = &msg,
Fred Isaman9ace33c2011-03-23 13:27:45 +00001598 .callback_ops = call_ops,
Trond Myklebust84115e12007-07-14 15:39:59 -04001599 .callback_data = data,
Trond Myklebust101070c2008-02-19 20:04:23 -05001600 .workqueue = nfsiod_workqueue,
Andy Adamson9f0ec1762012-04-27 17:53:44 -04001601 .flags = RPC_TASK_ASYNC | flags,
Trond Myklebust3ff75762007-07-14 15:40:00 -04001602 .priority = priority,
Trond Myklebust84115e12007-07-14 15:39:59 -04001603 };
Trond Myklebust788e7a82006-03-20 13:44:27 -05001604 /* Set up the initial task struct. */
Peng Taoc36aae92014-06-09 07:10:14 +08001605 nfs_ops->commit_setup(data, &msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
Kinglong Meeb4839eb2015-07-01 12:00:13 +08001607 dprintk("NFS: initiated commit call\n");
Trond Myklebustbdc7f022007-07-14 15:40:00 -04001608
Weston Andros Adamson8c21c622013-08-13 16:37:37 -04001609 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1610 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1611
Trond Myklebust07737692007-10-25 18:42:54 -04001612 task = rpc_run_task(&task_setup_data);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001613 if (IS_ERR(task))
1614 return PTR_ERR(task);
Jeff Laytond2224e72011-03-06 17:14:13 +00001615 if (how & FLUSH_SYNC)
1616 rpc_wait_for_completion_task(task);
Trond Myklebustdbae4c72008-04-14 14:54:53 -04001617 rpc_put_task(task);
1618 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619}
Fred Isamane0c2b382011-03-23 13:27:53 +00001620EXPORT_SYMBOL_GPL(nfs_initiate_commit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
Peng Tao378520b2014-08-07 10:15:02 +08001622static loff_t nfs_get_lwb(struct list_head *head)
1623{
1624 loff_t lwb = 0;
1625 struct nfs_page *req;
1626
1627 list_for_each_entry(req, head, wb_list)
1628 if (lwb < (req_offset(req) + req->wb_bytes))
1629 lwb = req_offset(req) + req->wb_bytes;
1630
1631 return lwb;
1632}
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634/*
Fred Isaman9ace33c2011-03-23 13:27:45 +00001635 * Set up the argument/result storage required for the RPC call.
1636 */
Fred Isaman0b7c0152012-04-20 14:47:39 -04001637void nfs_init_commit(struct nfs_commit_data *data,
Fred Isamanf453a542012-04-20 14:47:54 -04001638 struct list_head *head,
1639 struct pnfs_layout_segment *lseg,
1640 struct nfs_commit_info *cinfo)
Fred Isaman9ace33c2011-03-23 13:27:45 +00001641{
1642 struct nfs_page *first = nfs_list_entry(head->next);
David Howells2b0143b2015-03-17 22:25:59 +00001643 struct inode *inode = d_inode(first->wb_context->dentry);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001644
1645 /* Set up the RPC argument and reply structs
1646 * NB: take care not to mess about with data->commit et al. */
1647
1648 list_splice_init(head, &data->pages);
1649
1650 data->inode = inode;
1651 data->cred = first->wb_context->cred;
Fred Isaman988b6dc2011-03-23 13:27:52 +00001652 data->lseg = lseg; /* reference transferred */
Peng Tao378520b2014-08-07 10:15:02 +08001653 /* only set lwb for pnfs commit */
1654 if (lseg)
1655 data->lwb = nfs_get_lwb(&data->pages);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001656 data->mds_ops = &nfs_commit_ops;
Fred Isamanf453a542012-04-20 14:47:54 -04001657 data->completion_ops = cinfo->completion_ops;
Fred Isamanb359f9d2012-04-20 14:47:55 -04001658 data->dreq = cinfo->dreq;
Fred Isaman9ace33c2011-03-23 13:27:45 +00001659
1660 data->args.fh = NFS_FH(data->inode);
1661 /* Note: we always request a commit of the entire inode */
1662 data->args.offset = 0;
1663 data->args.count = 0;
Fred Isaman0b7c0152012-04-20 14:47:39 -04001664 data->context = get_nfs_open_context(first->wb_context);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001665 data->res.fattr = &data->fattr;
1666 data->res.verf = &data->verf;
1667 nfs_fattr_init(&data->fattr);
1668}
Fred Isamane0c2b382011-03-23 13:27:53 +00001669EXPORT_SYMBOL_GPL(nfs_init_commit);
Fred Isaman9ace33c2011-03-23 13:27:45 +00001670
Fred Isamane0c2b382011-03-23 13:27:53 +00001671void nfs_retry_commit(struct list_head *page_list,
Fred Isamanea2cf222012-04-20 14:47:53 -04001672 struct pnfs_layout_segment *lseg,
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001673 struct nfs_commit_info *cinfo,
1674 u32 ds_commit_idx)
Fred Isaman64bfeb42011-03-23 13:27:47 +00001675{
1676 struct nfs_page *req;
1677
1678 while (!list_empty(page_list)) {
1679 req = nfs_list_entry(page_list->next);
1680 nfs_list_remove_request(req);
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001681 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
Tom Haynes487b9b82015-02-13 13:19:54 -08001682 if (!cinfo->dreq)
1683 nfs_clear_page_commit(req->wb_page);
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001684 nfs_unlock_and_release_request(req);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001685 }
1686}
Fred Isamane0c2b382011-03-23 13:27:53 +00001687EXPORT_SYMBOL_GPL(nfs_retry_commit);
Fred Isaman64bfeb42011-03-23 13:27:47 +00001688
Fred Isaman9ace33c2011-03-23 13:27:45 +00001689/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 * Commit dirty pages
1691 */
1692static int
Fred Isamanea2cf222012-04-20 14:47:53 -04001693nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1694 struct nfs_commit_info *cinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001696 struct nfs_commit_data *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001698 data = nfs_commitdata_alloc();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699
1700 if (!data)
1701 goto out_bad;
1702
1703 /* Set up the argument struct */
Fred Isamanf453a542012-04-20 14:47:54 -04001704 nfs_init_commit(data, head, NULL, cinfo);
1705 atomic_inc(&cinfo->mds->rpcs_out);
Peng Taoc36aae92014-06-09 07:10:14 +08001706 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1707 data->mds_ops, how, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 out_bad:
Weston Andros Adamsonb57ff132014-09-05 18:20:21 -04001709 nfs_retry_commit(head, NULL, cinfo, 0);
Fred Isamanf453a542012-04-20 14:47:54 -04001710 cinfo->completion_ops->error_cleanup(NFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return -ENOMEM;
1712}
1713
1714/*
1715 * COMMIT call returned
1716 */
Trond Myklebust788e7a82006-03-20 13:44:27 -05001717static void nfs_commit_done(struct rpc_task *task, void *calldata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001719 struct nfs_commit_data *data = calldata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Chuck Levera3f565b2007-01-31 12:14:01 -05001721 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 task->tk_pid, task->tk_status);
1723
Trond Myklebust788e7a82006-03-20 13:44:27 -05001724 /* Call the NFS version-specific code */
Trond Myklebustc0d0e962011-04-12 12:29:15 -04001725 NFS_PROTO(data->inode)->commit_done(task, data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001726}
1727
Fred Isamanf453a542012-04-20 14:47:54 -04001728static void nfs_commit_release_pages(struct nfs_commit_data *data)
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001729{
Fred Isaman5917ce82011-03-23 13:27:48 +00001730 struct nfs_page *req;
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001731 int status = data->task.tk_status;
Fred Isamanf453a542012-04-20 14:47:54 -04001732 struct nfs_commit_info cinfo;
NeilBrown353db792014-09-24 11:28:32 +10001733 struct nfs_server *nfss;
Trond Myklebust788e7a82006-03-20 13:44:27 -05001734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 while (!list_empty(&data->pages)) {
1736 req = nfs_list_entry(data->pages.next);
1737 nfs_list_remove_request(req);
Fred Isamand6d6dc72012-03-08 17:29:35 -05001738 nfs_clear_page_commit(req->wb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
Niels de Vos1e8968c2013-12-17 18:20:16 +01001740 dprintk("NFS: commit (%s/%llu %d@%lld)",
Al Viro3d4ff432011-06-22 18:40:12 -04001741 req->wb_context->dentry->d_sb->s_id,
David Howells2b0143b2015-03-17 22:25:59 +00001742 (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 req->wb_bytes,
1744 (long long)req_offset(req));
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001745 if (status < 0) {
1746 nfs_context_set_write_error(req->wb_context, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 nfs_inode_remove_request(req);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001748 dprintk(", error = %d\n", status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 goto next;
1750 }
1751
1752 /* Okay, COMMIT succeeded, apparently. Check the verifier
1753 * returned by the server against all stored verfs. */
Trond Myklebust2f2c63b2012-06-08 11:56:09 -04001754 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 /* We have a match */
1756 nfs_inode_remove_request(req);
1757 dprintk(" OK\n");
1758 goto next;
1759 }
1760 /* We have a mismatch. Write the page again */
1761 dprintk(" mismatch\n");
Fred6d884e82008-03-19 11:24:38 -04001762 nfs_mark_request_dirty(req);
Trond Myklebust05990d12012-09-11 16:01:22 -04001763 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 next:
Trond Myklebust1d1afcb2012-05-09 14:04:55 -04001765 nfs_unlock_and_release_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 }
NeilBrown353db792014-09-24 11:28:32 +10001767 nfss = NFS_SERVER(data->inode);
1768 if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1769 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1770
Fred Isamanf453a542012-04-20 14:47:54 -04001771 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1772 if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1773 nfs_commit_clear_lock(NFS_I(data->inode));
Fred Isaman5917ce82011-03-23 13:27:48 +00001774}
1775
1776static void nfs_commit_release(void *calldata)
1777{
Fred Isaman0b7c0152012-04-20 14:47:39 -04001778 struct nfs_commit_data *data = calldata;
Fred Isaman5917ce82011-03-23 13:27:48 +00001779
Fred Isamanf453a542012-04-20 14:47:54 -04001780 data->completion_ops->completion(data);
Trond Myklebustc9d8f892008-04-15 16:56:39 -04001781 nfs_commitdata_release(calldata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782}
Trond Myklebust788e7a82006-03-20 13:44:27 -05001783
1784static const struct rpc_call_ops nfs_commit_ops = {
Fred Isaman0b7c0152012-04-20 14:47:39 -04001785 .rpc_call_prepare = nfs_commit_prepare,
Trond Myklebust788e7a82006-03-20 13:44:27 -05001786 .rpc_call_done = nfs_commit_done,
1787 .rpc_release = nfs_commit_release,
1788};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Fred Isamanf453a542012-04-20 14:47:54 -04001790static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1791 .completion = nfs_commit_release_pages,
1792 .error_cleanup = nfs_commit_clear_lock,
1793};
1794
Fred Isaman1763da12012-04-20 14:47:57 -04001795int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1796 int how, struct nfs_commit_info *cinfo)
Fred Isaman84c53ab2012-04-20 14:47:52 -04001797{
1798 int status;
1799
Fred Isamanea2cf222012-04-20 14:47:53 -04001800 status = pnfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001801 if (status == PNFS_NOT_ATTEMPTED)
Fred Isamanea2cf222012-04-20 14:47:53 -04001802 status = nfs_commit_list(inode, head, how, cinfo);
Fred Isaman84c53ab2012-04-20 14:47:52 -04001803 return status;
1804}
1805
Trond Myklebustb608b282010-07-30 15:31:54 -04001806int nfs_commit_inode(struct inode *inode, int how)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 LIST_HEAD(head);
Fred Isamanea2cf222012-04-20 14:47:53 -04001809 struct nfs_commit_info cinfo;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001810 int may_wait = how & FLUSH_SYNC;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001811 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Trond Myklebustb8413f92011-03-21 15:37:01 -04001813 res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1814 if (res <= 0)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001815 goto out_mark_dirty;
Fred Isamanea2cf222012-04-20 14:47:53 -04001816 nfs_init_cinfo_from_inode(&cinfo, inode);
1817 res = nfs_scan_commit(inode, &head, &cinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 if (res) {
Fred Isamana861a1e2011-03-23 13:27:51 +00001819 int error;
1820
Fred Isamanea2cf222012-04-20 14:47:53 -04001821 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
Trond Myklebust3da28eb2005-06-22 17:16:31 +00001822 if (error < 0)
1823 return error;
Trond Myklebustb8413f92011-03-21 15:37:01 -04001824 if (!may_wait)
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001825 goto out_mark_dirty;
NeilBrown74316202014-07-07 15:16:04 +10001826 error = wait_on_bit_action(&NFS_I(inode)->flags,
Trond Myklebustb8413f92011-03-21 15:37:01 -04001827 NFS_INO_COMMIT,
1828 nfs_wait_bit_killable,
1829 TASK_KILLABLE);
1830 if (error < 0)
1831 return error;
Trond Myklebust71d0a612010-04-22 15:35:57 -04001832 } else
1833 nfs_commit_clear_lock(NFS_I(inode));
Trond Myklebustc5efa5f2010-05-26 08:42:11 -04001834 return res;
1835 /* Note: If we exit without ensuring that the commit is complete,
1836 * we must mark the inode as dirty. Otherwise, future calls to
1837 * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1838 * that the data is on the disk.
1839 */
1840out_mark_dirty:
1841 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 return res;
1843}
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001844
Anna Schumakerae09c312015-07-13 14:01:32 -04001845int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001846{
Trond Myklebust420e3642010-02-19 17:00:02 -08001847 struct nfs_inode *nfsi = NFS_I(inode);
1848 int flags = FLUSH_SYNC;
1849 int ret = 0;
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001850
Jeff Layton3236c3e12011-10-11 09:49:21 -04001851 /* no commits means nothing needs to be done */
Fred Isamanea2cf222012-04-20 14:47:53 -04001852 if (!nfsi->commit_info.ncommit)
Jeff Layton3236c3e12011-10-11 09:49:21 -04001853 return ret;
1854
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001855 if (wbc->sync_mode == WB_SYNC_NONE) {
1856 /* Don't commit yet if this is a non-blocking flush and there
1857 * are a lot of outstanding writes for this mapping.
1858 */
Weston Andros Adamsoncb1410c2014-11-12 12:08:00 -05001859 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001860 goto out_mark_dirty;
Trond Myklebust420e3642010-02-19 17:00:02 -08001861
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001862 /* don't wait for the COMMIT response */
Trond Myklebust420e3642010-02-19 17:00:02 -08001863 flags = 0;
Jeff Laytona00dd6c2010-09-28 09:14:01 -04001864 }
1865
Trond Myklebust420e3642010-02-19 17:00:02 -08001866 ret = nfs_commit_inode(inode, flags);
1867 if (ret >= 0) {
1868 if (wbc->sync_mode == WB_SYNC_NONE) {
1869 if (ret < wbc->nr_to_write)
1870 wbc->nr_to_write -= ret;
1871 else
1872 wbc->nr_to_write = 0;
1873 }
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001874 return 0;
Trond Myklebust420e3642010-02-19 17:00:02 -08001875 }
1876out_mark_dirty:
Trond Myklebust8fc795f2010-02-19 16:46:56 -08001877 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1878 return ret;
1879}
Bryan Schumaker89d77c82012-07-30 16:05:25 -04001880EXPORT_SYMBOL_GPL(nfs_write_inode);
Andy Adamson863a3c62011-03-23 13:27:54 +00001881
Trond Myklebustacdc53b2010-02-19 17:03:26 -08001882/*
1883 * flush the inode to disk.
1884 */
1885int nfs_wb_all(struct inode *inode)
Trond Myklebust34901f72007-07-25 14:09:54 -04001886{
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001887 int ret;
Trond Myklebust34901f72007-07-25 14:09:54 -04001888
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001889 trace_nfs_writeback_inode_enter(inode);
1890
Trond Myklebust5bb89b42015-03-25 14:14:42 -04001891 ret = filemap_write_and_wait(inode->i_mapping);
Chuck Lever6b196872015-05-06 18:26:58 -04001892 if (ret)
1893 goto out;
1894 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1895 if (ret < 0)
1896 goto out;
1897 pnfs_sync_inode(inode, true);
1898 ret = 0;
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001899
Chuck Lever6b196872015-05-06 18:26:58 -04001900out:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001901 trace_nfs_writeback_inode_exit(inode, ret);
1902 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001903}
Bryan Schumakerddda8e02012-07-30 16:05:23 -04001904EXPORT_SYMBOL_GPL(nfs_wb_all);
Trond Myklebust1c759502006-10-09 16:18:38 -04001905
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001906int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1907{
1908 struct nfs_page *req;
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001909 int ret = 0;
1910
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001911 wait_on_page_writeback(page);
1912
1913 /* blocking call to cancel all requests and join to a single (head)
1914 * request */
1915 req = nfs_lock_and_join_requests(page, false);
1916
1917 if (IS_ERR(req)) {
1918 ret = PTR_ERR(req);
1919 } else if (req) {
1920 /* all requests from this page have been cancelled by
1921 * nfs_lock_and_join_requests, so just remove the head
1922 * request from the inode / page_private pointer and
1923 * release it */
1924 nfs_inode_remove_request(req);
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001925 nfs_unlock_and_release_request(req);
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001926 }
Weston Andros Adamson3e217042014-07-11 10:20:49 -04001927
Trond Myklebust1b3b4a12007-08-28 10:29:36 -04001928 return ret;
1929}
1930
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001931/*
1932 * Write back all requests on one page - we do this before reading it.
1933 */
1934int nfs_wb_page(struct inode *inode, struct page *page)
Trond Myklebust1c759502006-10-09 16:18:38 -04001935{
Mel Gorman29418aa2012-07-31 16:45:10 -07001936 loff_t range_start = page_file_offset(page);
Trond Myklebust1c759502006-10-09 16:18:38 -04001937 loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001938 struct writeback_control wbc = {
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001939 .sync_mode = WB_SYNC_ALL,
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001940 .nr_to_write = 0,
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001941 .range_start = range_start,
1942 .range_end = range_end,
1943 };
1944 int ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001945
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001946 trace_nfs_writeback_page_enter(inode);
1947
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001948 for (;;) {
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001949 wait_on_page_writeback(page);
Trond Myklebust73e33022008-04-11 16:03:54 -04001950 if (clear_page_dirty_for_io(page)) {
1951 ret = nfs_writepage_locked(page, &wbc);
1952 if (ret < 0)
1953 goto out_error;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001954 continue;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001955 }
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001956 ret = 0;
Trond Myklebust0522f6a2010-05-26 08:42:24 -04001957 if (!PagePrivate(page))
1958 break;
1959 ret = nfs_commit_inode(inode, FLUSH_SYNC);
Trond Myklebustba8b06e2010-04-27 18:33:54 -04001960 if (ret < 0)
Trond Myklebust73e33022008-04-11 16:03:54 -04001961 goto out_error;
Trond Myklebust7f2f12d2010-02-19 17:03:28 -08001962 }
Trond Myklebust73e33022008-04-11 16:03:54 -04001963out_error:
Trond Myklebustf4ce1292013-08-19 18:59:33 -04001964 trace_nfs_writeback_page_exit(inode, ret);
Trond Myklebust4d770cc2006-12-05 00:35:41 -05001965 return ret;
Trond Myklebust1c759502006-10-09 16:18:38 -04001966}
1967
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001968#ifdef CONFIG_MIGRATION
1969int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
Mel Gormana6bc32b2012-01-12 17:19:43 -08001970 struct page *page, enum migrate_mode mode)
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001971{
Jeff Layton2da95652011-10-12 10:57:42 -04001972 /*
1973 * If PagePrivate is set, then the page is currently associated with
1974 * an in-progress read or write request. Don't try to migrate it.
1975 *
1976 * FIXME: we could do this in principle, but we'll need a way to ensure
1977 * that we can safely release the inode reference while holding
1978 * the page lock.
1979 */
1980 if (PagePrivate(page))
1981 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001982
David Howells8c209ce2012-12-05 13:34:49 +00001983 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1984 return -EBUSY;
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001985
Mel Gormana6bc32b2012-01-12 17:19:43 -08001986 return migrate_page(mapping, newpage, page, mode);
Trond Myklebust074cc1d2009-08-10 08:54:13 -04001987}
1988#endif
1989
David Howellsf7b422b2006-06-09 09:34:33 -04001990int __init nfs_init_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991{
1992 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
Weston Andros Adamson1e7f3a42014-06-09 11:48:33 -04001993 sizeof(struct nfs_pgio_header),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 0, SLAB_HWCACHE_ALIGN,
Paul Mundt20c2df82007-07-20 10:11:58 +09001995 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 if (nfs_wdata_cachep == NULL)
1997 return -ENOMEM;
1998
Matthew Dobson93d23412006-03-26 01:37:50 -08001999 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2000 nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (nfs_wdata_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002002 goto out_destroy_write_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
Fred Isaman0b7c0152012-04-20 14:47:39 -04002004 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2005 sizeof(struct nfs_commit_data),
2006 0, SLAB_HWCACHE_ALIGN,
2007 NULL);
2008 if (nfs_cdata_cachep == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002009 goto out_destroy_write_mempool;
Fred Isaman0b7c0152012-04-20 14:47:39 -04002010
Matthew Dobson93d23412006-03-26 01:37:50 -08002011 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
Yanchuan Nian4c100212012-11-12 09:27:37 +08002012 nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 if (nfs_commit_mempool == NULL)
Jeff Layton3dd47652012-08-02 14:30:56 -04002014 goto out_destroy_commit_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Peter Zijlstra89a09142007-03-16 13:38:26 -08002016 /*
2017 * NFS congestion size, scale with available memory.
2018 *
2019 * 64MB: 8192k
2020 * 128MB: 11585k
2021 * 256MB: 16384k
2022 * 512MB: 23170k
2023 * 1GB: 32768k
2024 * 2GB: 46340k
2025 * 4GB: 65536k
2026 * 8GB: 92681k
2027 * 16GB: 131072k
2028 *
2029 * This allows larger machines to have larger/more transfers.
2030 * Limit the default to 256M
2031 */
2032 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2033 if (nfs_congestion_kb > 256*1024)
2034 nfs_congestion_kb = 256*1024;
2035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return 0;
Jeff Layton3dd47652012-08-02 14:30:56 -04002037
2038out_destroy_commit_cache:
2039 kmem_cache_destroy(nfs_cdata_cachep);
2040out_destroy_write_mempool:
2041 mempool_destroy(nfs_wdata_mempool);
2042out_destroy_write_cache:
2043 kmem_cache_destroy(nfs_wdata_cachep);
2044 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045}
2046
David Brownell266bee82006-06-27 12:59:15 -07002047void nfs_destroy_writepagecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048{
2049 mempool_destroy(nfs_commit_mempool);
Jeff Layton3dd47652012-08-02 14:30:56 -04002050 kmem_cache_destroy(nfs_cdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 mempool_destroy(nfs_wdata_mempool);
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002052 kmem_cache_destroy(nfs_wdata_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053}
2054
Anna Schumaker4a0de552014-05-06 09:12:30 -04002055static const struct nfs_rw_ops nfs_rw_write_ops = {
Anna Schumakera4cdda52014-05-06 09:12:31 -04002056 .rw_mode = FMODE_WRITE,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002057 .rw_alloc_header = nfs_writehdr_alloc,
2058 .rw_free_header = nfs_writehdr_free,
Anna Schumaker0eecb212014-05-06 09:12:32 -04002059 .rw_done = nfs_writeback_done,
2060 .rw_result = nfs_writeback_result,
Anna Schumaker1ed26f32014-05-06 09:12:37 -04002061 .rw_initiate = nfs_initiate_write,
Anna Schumaker4a0de552014-05-06 09:12:30 -04002062};