1
2
3
4
5
6
7
8
9#include <linux/types.h>
10#include <linux/slab.h>
11#include <linux/mm.h>
12#include <linux/pagemap.h>
13#include <linux/file.h>
14#include <linux/writeback.h>
15#include <linux/swap.h>
16#include <linux/migrate.h>
17
18#include <linux/sunrpc/clnt.h>
19#include <linux/nfs_fs.h>
20#include <linux/nfs_mount.h>
21#include <linux/nfs_page.h>
22#include <linux/backing-dev.h>
23#include <linux/export.h>
24#include <linux/freezer.h>
25#include <linux/wait.h>
26
27#include <asm/uaccess.h>
28
29#include "delegation.h"
30#include "internal.h"
31#include "iostat.h"
32#include "nfs4_fs.h"
33#include "fscache.h"
34#include "pnfs.h"
35
36#include "nfstrace.h"
37
38#define NFSDBG_FACILITY NFSDBG_PAGECACHE
39
40#define MIN_POOL_WRITE (32)
41#define MIN_POOL_COMMIT (4)
42
43
44
45
46static void nfs_redirty_request(struct nfs_page *req);
47static const struct rpc_call_ops nfs_commit_ops;
48static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
49static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
50static const struct nfs_rw_ops nfs_rw_write_ops;
51static void nfs_clear_request_commit(struct nfs_page *req);
52static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
53 struct inode *inode);
54static struct nfs_page *
55nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
56 struct page *page);
57
58static struct kmem_cache *nfs_wdata_cachep;
59static mempool_t *nfs_wdata_mempool;
60static struct kmem_cache *nfs_cdata_cachep;
61static mempool_t *nfs_commit_mempool;
62
63struct nfs_commit_data *nfs_commitdata_alloc(void)
64{
65 struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
66
67 if (p) {
68 memset(p, 0, sizeof(*p));
69 INIT_LIST_HEAD(&p->pages);
70 }
71 return p;
72}
73EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
74
75void nfs_commit_free(struct nfs_commit_data *p)
76{
77 mempool_free(p, nfs_commit_mempool);
78}
79EXPORT_SYMBOL_GPL(nfs_commit_free);
80
81static struct nfs_pgio_header *nfs_writehdr_alloc(void)
82{
83 struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
84
85 if (p)
86 memset(p, 0, sizeof(*p));
87 return p;
88}
89
90static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
91{
92 mempool_free(hdr, nfs_wdata_mempool);
93}
94
95static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
96{
97 ctx->error = error;
98 smp_wmb();
99 set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
100}
101
102
103
104
105
106
107
108
109static struct nfs_page *
110nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
111{
112 struct nfs_page *req = NULL;
113
114 if (PagePrivate(page))
115 req = (struct nfs_page *)page_private(page);
116 else if (unlikely(PageSwapCache(page)))
117 req = nfs_page_search_commits_for_head_request_locked(nfsi,
118 page);
119
120 if (req) {
121 WARN_ON_ONCE(req->wb_head != req);
122 kref_get(&req->wb_kref);
123 }
124
125 return req;
126}
127
128
129
130
131
132
133static struct nfs_page *nfs_page_find_head_request(struct page *page)
134{
135 struct inode *inode = page_file_mapping(page)->host;
136 struct nfs_page *req = NULL;
137
138 spin_lock(&inode->i_lock);
139 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
140 spin_unlock(&inode->i_lock);
141 return req;
142}
143
144
145static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
146{
147 struct inode *inode = page_file_mapping(page)->host;
148 loff_t end, i_size;
149 pgoff_t end_index;
150
151 spin_lock(&inode->i_lock);
152 i_size = i_size_read(inode);
153 end_index = (i_size - 1) >> PAGE_SHIFT;
154 if (i_size > 0 && page_file_index(page) < end_index)
155 goto out;
156 end = page_file_offset(page) + ((loff_t)offset+count);
157 if (i_size >= end)
158 goto out;
159 i_size_write(inode, end);
160 nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
161out:
162 spin_unlock(&inode->i_lock);
163}
164
165
166static void nfs_set_pageerror(struct page *page)
167{
168 nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
169}
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184static struct nfs_page *
185nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
186{
187 struct nfs_page *req;
188
189 WARN_ON_ONCE(head != head->wb_head);
190 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
191
192 req = head;
193 do {
194 if (page_offset >= req->wb_pgbase &&
195 page_offset < (req->wb_pgbase + req->wb_bytes))
196 return req;
197
198 req = req->wb_this_page;
199 } while (req != head);
200
201 return NULL;
202}
203
204
205
206
207
208
209
210
211static bool nfs_page_group_covers_page(struct nfs_page *req)
212{
213 struct nfs_page *tmp;
214 unsigned int pos = 0;
215 unsigned int len = nfs_page_length(req->wb_page);
216
217 nfs_page_group_lock(req, false);
218
219 do {
220 tmp = nfs_page_group_search_locked(req->wb_head, pos);
221 if (tmp) {
222
223 WARN_ON_ONCE(tmp->wb_pgbase != pos);
224 pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
225 }
226 } while (tmp && pos < len);
227
228 nfs_page_group_unlock(req);
229 WARN_ON_ONCE(pos > len);
230 return pos == len;
231}
232
233
234
235
236static void nfs_mark_uptodate(struct nfs_page *req)
237{
238 if (PageUptodate(req->wb_page))
239 return;
240 if (!nfs_page_group_covers_page(req))
241 return;
242 SetPageUptodate(req->wb_page);
243}
244
245static int wb_priority(struct writeback_control *wbc)
246{
247 int ret = 0;
248 if (wbc->for_reclaim)
249 return FLUSH_HIGHPRI | FLUSH_COND_STABLE;
250 if (wbc->sync_mode == WB_SYNC_ALL)
251 ret = FLUSH_COND_STABLE;
252 return ret;
253}
254
255
256
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
265static void nfs_set_page_writeback(struct page *page)
266{
267 struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
268 int ret = test_set_page_writeback(page);
269
270 WARN_ON_ONCE(ret != 0);
271
272 if (atomic_long_inc_return(&nfss->writeback) >
273 NFS_CONGESTION_ON_THRESH) {
274 set_bdi_congested(&nfss->backing_dev_info,
275 BLK_RW_ASYNC);
276 }
277}
278
279static void nfs_end_page_writeback(struct nfs_page *req)
280{
281 struct inode *inode = page_file_mapping(req->wb_page)->host;
282 struct nfs_server *nfss = NFS_SERVER(inode);
283
284 if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
285 return;
286
287 end_page_writeback(req->wb_page);
288 if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
289 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
290}
291
292
293
294
295
296
297static void
298nfs_page_group_clear_bits(struct nfs_page *req)
299{
300 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
310
311
312
313
314
315
316
317
318
319
320
321
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;
329 int ret;
330
331
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
338 kref_get(&req->wb_kref);
339
340 nfs_page_group_unlock(head);
341 spin_unlock(&inode->i_lock);
342
343
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
357
358
359
360
361
362
363
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
378 subreq->wb_head = subreq;
379 subreq->wb_this_page = subreq;
380
381
382
383 nfs_unlock_request(subreq);
384
385 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
386
387 nfs_release_request(old_head);
388
389 nfs_page_group_clear_bits(subreq);
390
391
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
399
400
401
402 nfs_page_group_clear_bits(subreq);
403 nfs_free_request(subreq);
404 }
405 }
406}
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
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
445
446
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
455
456
457 ret = nfs_page_group_lock(head, true);
458 if (ret < 0) {
459 spin_unlock(&inode->i_lock);
460
461 if (!nonblock && ret == -EAGAIN) {
462 nfs_page_group_lock_wait(head);
463 nfs_release_request(head);
464 goto try_again;
465 }
466
467 nfs_release_request(head);
468 return ERR_PTR(ret);
469 }
470
471
472 subreq = head;
473 do {
474
475
476
477
478 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
479
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 }
488
489 if (!nfs_lock_request(subreq)) {
490
491
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
505
506 subreq = head;
507 do {
508 nfs_clear_request_commit(subreq);
509 subreq = subreq->wb_this_page;
510 } while (subreq != head);
511
512
513 if (head->wb_this_page != head) {
514
515 destroy_list = head->wb_this_page;
516 head->wb_this_page = head;
517
518
519
520 head->wb_bytes = total_bytes;
521 }
522
523
524
525
526 nfs_page_group_clear_bits(head);
527
528
529
530
531
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
538
539 spin_unlock(&inode->i_lock);
540
541 nfs_destroy_unlinked_subrequests(destroy_list, head);
542
543
544
545 return head;
546}
547
548static 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
557
558
559
560
561static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
562 struct page *page, bool nonblock,
563 bool launder)
564{
565 struct nfs_page *req;
566 int ret = 0;
567
568 req = nfs_lock_and_join_requests(page, nonblock);
569 if (!req)
570 goto out;
571 ret = PTR_ERR(req);
572 if (IS_ERR(req))
573 goto out;
574
575 nfs_set_page_writeback(page);
576 WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
577
578 ret = 0;
579 if (!nfs_pageio_add_request(pgio, req)) {
580 ret = pgio->pg_error;
581
582
583
584
585
586 if (nfs_error_is_fatal(ret)) {
587 nfs_context_set_write_error(req->wb_context, ret);
588 if (launder) {
589 nfs_write_error_remove_page(req);
590 goto out;
591 }
592 }
593 nfs_redirty_request(req);
594 ret = -EAGAIN;
595 } else
596 nfs_add_stats(page_file_mapping(page)->host,
597 NFSIOS_WRITEPAGES, 1);
598out:
599 return ret;
600}
601
602static int nfs_do_writepage(struct page *page, struct writeback_control *wbc,
603 struct nfs_pageio_descriptor *pgio, bool launder)
604{
605 int ret;
606
607 nfs_pageio_cond_complete(pgio, page_file_index(page));
608 ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE,
609 launder);
610 if (ret == -EAGAIN) {
611 redirty_page_for_writepage(wbc, page);
612 ret = 0;
613 }
614 return ret;
615}
616
617
618
619
620static int nfs_writepage_locked(struct page *page,
621 struct writeback_control *wbc,
622 bool launder)
623{
624 struct nfs_pageio_descriptor pgio;
625 struct inode *inode = page_file_mapping(page)->host;
626 int err;
627
628 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
629 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
630 false, &nfs_async_write_completion_ops);
631 err = nfs_do_writepage(page, wbc, &pgio, launder);
632 nfs_pageio_complete(&pgio);
633 if (err < 0)
634 return err;
635 if (pgio.pg_error < 0)
636 return pgio.pg_error;
637 return 0;
638}
639
640int nfs_writepage(struct page *page, struct writeback_control *wbc)
641{
642 int ret;
643
644 ret = nfs_writepage_locked(page, wbc, false);
645 unlock_page(page);
646 return ret;
647}
648
649static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
650{
651 int ret;
652
653 ret = nfs_do_writepage(page, wbc, data, false);
654 unlock_page(page);
655 return ret;
656}
657
658int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
659{
660 struct inode *inode = mapping->host;
661 unsigned long *bitlock = &NFS_I(inode)->flags;
662 struct nfs_pageio_descriptor pgio;
663 int err;
664
665
666 err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
667 nfs_wait_bit_killable, TASK_KILLABLE);
668 if (err)
669 goto out_err;
670
671 nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
672
673 nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
674 &nfs_async_write_completion_ops);
675 err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
676 nfs_pageio_complete(&pgio);
677
678 clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
679 smp_mb__after_atomic();
680 wake_up_bit(bitlock, NFS_INO_FLUSHING);
681
682 if (err < 0)
683 goto out_err;
684 err = pgio.pg_error;
685 if (err < 0)
686 goto out_err;
687 return 0;
688out_err:
689 return err;
690}
691
692
693
694
695static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
696{
697 struct nfs_inode *nfsi = NFS_I(inode);
698
699 WARN_ON_ONCE(req->wb_this_page != req);
700
701
702 nfs_lock_request(req);
703
704 spin_lock(&inode->i_lock);
705 if (!nfsi->nrequests &&
706 NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
707 inode->i_version++;
708
709
710
711
712 if (likely(!PageSwapCache(req->wb_page))) {
713 set_bit(PG_MAPPED, &req->wb_flags);
714 SetPagePrivate(req->wb_page);
715 set_page_private(req->wb_page, (unsigned long)req);
716 }
717 nfsi->nrequests++;
718
719
720
721
722 WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
723 kref_get(&req->wb_kref);
724 spin_unlock(&inode->i_lock);
725}
726
727
728
729
730static void nfs_inode_remove_request(struct nfs_page *req)
731{
732 struct inode *inode = d_inode(req->wb_context->dentry);
733 struct nfs_inode *nfsi = NFS_I(inode);
734 struct nfs_page *head;
735
736 if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
737 head = req->wb_head;
738
739 spin_lock(&inode->i_lock);
740 if (likely(!PageSwapCache(head->wb_page))) {
741 set_page_private(head->wb_page, 0);
742 ClearPagePrivate(head->wb_page);
743 smp_mb__after_atomic();
744 wake_up_page(head->wb_page, PG_private);
745 clear_bit(PG_MAPPED, &head->wb_flags);
746 }
747 nfsi->nrequests--;
748 spin_unlock(&inode->i_lock);
749 } else {
750 spin_lock(&inode->i_lock);
751 nfsi->nrequests--;
752 spin_unlock(&inode->i_lock);
753 }
754
755 if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
756 nfs_release_request(req);
757}
758
759static void
760nfs_mark_request_dirty(struct nfs_page *req)
761{
762 __set_page_dirty_nobuffers(req->wb_page);
763}
764
765
766
767
768
769
770
771
772
773static struct nfs_page *
774nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
775 struct page *page)
776{
777 struct nfs_page *freq, *t;
778 struct nfs_commit_info cinfo;
779 struct inode *inode = &nfsi->vfs_inode;
780
781 nfs_init_cinfo_from_inode(&cinfo, inode);
782
783
784 freq = pnfs_search_commit_reqs(inode, &cinfo, page);
785 if (freq)
786 return freq->wb_head;
787
788
789 list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
790 if (freq->wb_page == page)
791 return freq->wb_head;
792 }
793
794 return NULL;
795}
796
797
798
799
800
801
802
803
804
805
806
807
808
809void
810nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
811 struct nfs_commit_info *cinfo)
812{
813 set_bit(PG_CLEAN, &req->wb_flags);
814 nfs_list_add_request(req, dst);
815 cinfo->mds->ncommit++;
816}
817EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832void
833nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo)
834{
835 spin_lock(cinfo->lock);
836 nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo);
837 spin_unlock(cinfo->lock);
838 nfs_mark_page_unstable(req->wb_page, cinfo);
839}
840EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
841
842
843
844
845
846
847
848
849
850
851
852
853void
854nfs_request_remove_commit_list(struct nfs_page *req,
855 struct nfs_commit_info *cinfo)
856{
857 if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
858 return;
859 nfs_list_remove_request(req);
860 cinfo->mds->ncommit--;
861}
862EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
863
864static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
865 struct inode *inode)
866{
867 cinfo->lock = &inode->i_lock;
868 cinfo->mds = &NFS_I(inode)->commit_info;
869 cinfo->ds = pnfs_get_ds_info(inode);
870 cinfo->dreq = NULL;
871 cinfo->completion_ops = &nfs_commit_completion_ops;
872}
873
874void nfs_init_cinfo(struct nfs_commit_info *cinfo,
875 struct inode *inode,
876 struct nfs_direct_req *dreq)
877{
878 if (dreq)
879 nfs_init_cinfo_from_dreq(cinfo, dreq);
880 else
881 nfs_init_cinfo_from_inode(cinfo, inode);
882}
883EXPORT_SYMBOL_GPL(nfs_init_cinfo);
884
885
886
887
888void
889nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
890 struct nfs_commit_info *cinfo, u32 ds_commit_idx)
891{
892 if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
893 return;
894 nfs_request_add_commit_list(req, cinfo);
895}
896
897static void
898nfs_clear_page_commit(struct page *page)
899{
900 dec_zone_page_state(page, NR_UNSTABLE_NFS);
901 dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
902 WB_RECLAIMABLE);
903}
904
905
906static void
907nfs_clear_request_commit(struct nfs_page *req)
908{
909 if (test_bit(PG_CLEAN, &req->wb_flags)) {
910 struct inode *inode = d_inode(req->wb_context->dentry);
911 struct nfs_commit_info cinfo;
912
913 nfs_init_cinfo_from_inode(&cinfo, inode);
914 if (!pnfs_clear_request_commit(req, &cinfo)) {
915 nfs_request_remove_commit_list(req, &cinfo);
916 }
917 nfs_clear_page_commit(req->wb_page);
918 }
919}
920
921int nfs_write_need_commit(struct nfs_pgio_header *hdr)
922{
923 if (hdr->verf.committed == NFS_DATA_SYNC)
924 return hdr->lseg == NULL;
925 return hdr->verf.committed != NFS_FILE_SYNC;
926}
927
928static void nfs_write_completion(struct nfs_pgio_header *hdr)
929{
930 struct nfs_commit_info cinfo;
931 unsigned long bytes = 0;
932
933 if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
934 goto out;
935 nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
936 while (!list_empty(&hdr->pages)) {
937 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
938
939 bytes += req->wb_bytes;
940 nfs_list_remove_request(req);
941 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
942 (hdr->good_bytes < bytes)) {
943 nfs_set_pageerror(req->wb_page);
944 nfs_context_set_write_error(req->wb_context, hdr->error);
945 goto remove_req;
946 }
947 if (nfs_write_need_commit(hdr)) {
948 memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
949 nfs_mark_request_commit(req, hdr->lseg, &cinfo,
950 hdr->pgio_mirror_idx);
951 goto next;
952 }
953remove_req:
954 nfs_inode_remove_request(req);
955next:
956 nfs_unlock_request(req);
957 nfs_end_page_writeback(req);
958 nfs_release_request(req);
959 }
960out:
961 hdr->release(hdr);
962}
963
964unsigned long
965nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
966{
967 return cinfo->mds->ncommit;
968}
969
970
971int
972nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
973 struct nfs_commit_info *cinfo, int max)
974{
975 struct nfs_page *req, *tmp;
976 int ret = 0;
977
978 list_for_each_entry_safe(req, tmp, src, wb_list) {
979 if (!nfs_lock_request(req))
980 continue;
981 kref_get(&req->wb_kref);
982 if (cond_resched_lock(cinfo->lock))
983 list_safe_reset_next(req, tmp, wb_list);
984 nfs_request_remove_commit_list(req, cinfo);
985 nfs_list_add_request(req, dst);
986 ret++;
987 if ((ret == max) && !cinfo->dreq)
988 break;
989 }
990 return ret;
991}
992
993
994
995
996
997
998
999
1000
1001
1002int
1003nfs_scan_commit(struct inode *inode, struct list_head *dst,
1004 struct nfs_commit_info *cinfo)
1005{
1006 int ret = 0;
1007
1008 spin_lock(cinfo->lock);
1009 if (cinfo->mds->ncommit > 0) {
1010 const int max = INT_MAX;
1011
1012 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1013 cinfo, max);
1014 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
1015 }
1016 spin_unlock(cinfo->lock);
1017 return ret;
1018}
1019
1020
1021
1022
1023
1024
1025
1026
1027static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1028 struct page *page,
1029 unsigned int offset,
1030 unsigned int bytes)
1031{
1032 struct nfs_page *req;
1033 unsigned int rqend;
1034 unsigned int end;
1035 int error;
1036
1037 if (!PagePrivate(page))
1038 return NULL;
1039
1040 end = offset + bytes;
1041 spin_lock(&inode->i_lock);
1042
1043 for (;;) {
1044 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
1045 if (req == NULL)
1046 goto out_unlock;
1047
1048
1049 WARN_ON_ONCE(req->wb_head != req);
1050 WARN_ON_ONCE(req->wb_this_page != req);
1051
1052 rqend = req->wb_offset + req->wb_bytes;
1053
1054
1055
1056
1057
1058
1059 if (offset > rqend
1060 || end < req->wb_offset)
1061 goto out_flushme;
1062
1063 if (nfs_lock_request(req))
1064 break;
1065
1066
1067 spin_unlock(&inode->i_lock);
1068 error = nfs_wait_on_request(req);
1069 nfs_release_request(req);
1070 if (error != 0)
1071 goto out_err;
1072 spin_lock(&inode->i_lock);
1073 }
1074
1075
1076 if (offset < req->wb_offset) {
1077 req->wb_offset = offset;
1078 req->wb_pgbase = offset;
1079 }
1080 if (end > rqend)
1081 req->wb_bytes = end - req->wb_offset;
1082 else
1083 req->wb_bytes = rqend - req->wb_offset;
1084out_unlock:
1085 if (req)
1086 nfs_clear_request_commit(req);
1087 spin_unlock(&inode->i_lock);
1088 return req;
1089out_flushme:
1090 spin_unlock(&inode->i_lock);
1091 nfs_release_request(req);
1092 error = nfs_wb_page(inode, page);
1093out_err:
1094 return ERR_PTR(error);
1095}
1096
1097
1098
1099
1100
1101
1102
1103
1104static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1105 struct page *page, unsigned int offset, unsigned int bytes)
1106{
1107 struct inode *inode = page_file_mapping(page)->host;
1108 struct nfs_page *req;
1109
1110 req = nfs_try_to_update_request(inode, page, offset, bytes);
1111 if (req != NULL)
1112 goto out;
1113 req = nfs_create_request(ctx, page, NULL, offset, bytes);
1114 if (IS_ERR(req))
1115 goto out;
1116 nfs_inode_add_request(inode, req);
1117out:
1118 return req;
1119}
1120
1121static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1122 unsigned int offset, unsigned int count)
1123{
1124 struct nfs_page *req;
1125
1126 req = nfs_setup_write_request(ctx, page, offset, count);
1127 if (IS_ERR(req))
1128 return PTR_ERR(req);
1129
1130 nfs_grow_file(page, offset, count);
1131 nfs_mark_uptodate(req);
1132 nfs_mark_request_dirty(req);
1133 nfs_unlock_and_release_request(req);
1134 return 0;
1135}
1136
1137int nfs_flush_incompatible(struct file *file, struct page *page)
1138{
1139 struct nfs_open_context *ctx = nfs_file_open_context(file);
1140 struct nfs_lock_context *l_ctx;
1141 struct file_lock_context *flctx = file_inode(file)->i_flctx;
1142 struct nfs_page *req;
1143 int do_flush, status;
1144
1145
1146
1147
1148
1149
1150
1151
1152 do {
1153 req = nfs_page_find_head_request(page);
1154 if (req == NULL)
1155 return 0;
1156 l_ctx = req->wb_lock_context;
1157 do_flush = req->wb_page != page ||
1158 !nfs_match_open_context(req->wb_context, ctx);
1159
1160 do_flush |= req->wb_this_page != req;
1161 if (l_ctx && flctx &&
1162 !(list_empty_careful(&flctx->flc_posix) &&
1163 list_empty_careful(&flctx->flc_flock))) {
1164 do_flush |= l_ctx->lockowner.l_owner != current->files
1165 || l_ctx->lockowner.l_pid != current->tgid;
1166 }
1167 nfs_release_request(req);
1168 if (!do_flush)
1169 return 0;
1170 status = nfs_wb_page(page_file_mapping(page)->host, page);
1171 } while (status == 0);
1172 return status;
1173}
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185int
1186nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1187{
1188 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1189 struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1190
1191 return rpcauth_key_timeout_notify(auth, ctx->cred);
1192}
1193
1194
1195
1196
1197bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1198{
1199 return rpcauth_cred_key_to_expire(ctx->cred);
1200}
1201
1202
1203
1204
1205
1206
1207static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
1208{
1209 struct nfs_inode *nfsi = NFS_I(inode);
1210
1211 if (nfs_have_delegated_attributes(inode))
1212 goto out;
1213 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1214 return false;
1215 smp_rmb();
1216 if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
1217 return false;
1218out:
1219 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1220 return false;
1221 return PageUptodate(page) != 0;
1222}
1223
1224static bool
1225is_whole_file_wrlock(struct file_lock *fl)
1226{
1227 return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1228 fl->fl_type == F_WRLCK;
1229}
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1240{
1241 int ret;
1242 struct file_lock_context *flctx = inode->i_flctx;
1243 struct file_lock *fl;
1244
1245 if (file->f_flags & O_DSYNC)
1246 return 0;
1247 if (!nfs_write_pageuptodate(page, inode))
1248 return 0;
1249 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1250 return 1;
1251 if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1252 list_empty_careful(&flctx->flc_posix)))
1253 return 1;
1254
1255
1256 ret = 0;
1257 spin_lock(&flctx->flc_lock);
1258 if (!list_empty(&flctx->flc_posix)) {
1259 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1260 fl_list);
1261 if (is_whole_file_wrlock(fl))
1262 ret = 1;
1263 } else if (!list_empty(&flctx->flc_flock)) {
1264 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1265 fl_list);
1266 if (fl->fl_type == F_WRLCK)
1267 ret = 1;
1268 }
1269 spin_unlock(&flctx->flc_lock);
1270 return ret;
1271}
1272
1273
1274
1275
1276
1277
1278
1279int nfs_updatepage(struct file *file, struct page *page,
1280 unsigned int offset, unsigned int count)
1281{
1282 struct nfs_open_context *ctx = nfs_file_open_context(file);
1283 struct inode *inode = page_file_mapping(page)->host;
1284 int status = 0;
1285
1286 nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1287
1288 dprintk("NFS: nfs_updatepage(%pD2 %d@%lld)\n",
1289 file, count, (long long)(page_file_offset(page) + offset));
1290
1291 if (nfs_can_extend_write(file, page, inode)) {
1292 count = max(count + offset, nfs_page_length(page));
1293 offset = 0;
1294 }
1295
1296 status = nfs_writepage_setup(ctx, page, offset, count);
1297 if (status < 0)
1298 nfs_set_pageerror(page);
1299 else
1300 __set_page_dirty_nobuffers(page);
1301
1302 dprintk("NFS: nfs_updatepage returns %d (isize %lld)\n",
1303 status, (long long)i_size_read(inode));
1304 return status;
1305}
1306
1307static int flush_task_priority(int how)
1308{
1309 switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1310 case FLUSH_HIGHPRI:
1311 return RPC_PRIORITY_HIGH;
1312 case FLUSH_LOWPRI:
1313 return RPC_PRIORITY_LOW;
1314 }
1315 return RPC_PRIORITY_NORMAL;
1316}
1317
1318static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1319 struct rpc_message *msg,
1320 const struct nfs_rpc_ops *rpc_ops,
1321 struct rpc_task_setup *task_setup_data, int how)
1322{
1323 int priority = flush_task_priority(how);
1324
1325 task_setup_data->priority = priority;
1326 rpc_ops->write_setup(hdr, msg);
1327
1328 nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
1329 &task_setup_data->rpc_client, msg, hdr);
1330}
1331
1332
1333
1334
1335
1336static void nfs_redirty_request(struct nfs_page *req)
1337{
1338 nfs_mark_request_dirty(req);
1339 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1340 nfs_unlock_request(req);
1341 nfs_end_page_writeback(req);
1342 nfs_release_request(req);
1343}
1344
1345static void nfs_async_write_error(struct list_head *head)
1346{
1347 struct nfs_page *req;
1348
1349 while (!list_empty(head)) {
1350 req = nfs_list_entry(head->next);
1351 nfs_list_remove_request(req);
1352 nfs_redirty_request(req);
1353 }
1354}
1355
1356static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
1357{
1358 nfs_async_write_error(&hdr->pages);
1359}
1360
1361static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1362 .error_cleanup = nfs_async_write_error,
1363 .completion = nfs_write_completion,
1364 .reschedule_io = nfs_async_write_reschedule_io,
1365};
1366
1367void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1368 struct inode *inode, int ioflags, bool force_mds,
1369 const struct nfs_pgio_completion_ops *compl_ops)
1370{
1371 struct nfs_server *server = NFS_SERVER(inode);
1372 const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1373
1374#ifdef CONFIG_NFS_V4_1
1375 if (server->pnfs_curr_ld && !force_mds)
1376 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1377#endif
1378 nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1379 server->wsize, ioflags);
1380}
1381EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1382
1383void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1384{
1385 struct nfs_pgio_mirror *mirror;
1386
1387 if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1388 pgio->pg_ops->pg_cleanup(pgio);
1389
1390 pgio->pg_ops = &nfs_pgio_rw_ops;
1391
1392 nfs_pageio_stop_mirroring(pgio);
1393
1394 mirror = &pgio->pg_mirrors[0];
1395 mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1396}
1397EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1398
1399
1400void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1401{
1402 struct nfs_commit_data *data = calldata;
1403
1404 NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1405}
1406
1407
1408
1409
1410static int nfs_should_remove_suid(const struct inode *inode)
1411{
1412 umode_t mode = inode->i_mode;
1413 int kill = 0;
1414
1415
1416 if (unlikely(mode & S_ISUID))
1417 kill = ATTR_KILL_SUID;
1418
1419
1420
1421
1422
1423 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1424 kill |= ATTR_KILL_SGID;
1425
1426 if (unlikely(kill && S_ISREG(mode)))
1427 return kill;
1428
1429 return 0;
1430}
1431
1432static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1433 struct nfs_fattr *fattr)
1434{
1435 struct nfs_pgio_args *argp = &hdr->args;
1436 struct nfs_pgio_res *resp = &hdr->res;
1437 u64 size = argp->offset + resp->count;
1438
1439 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1440 fattr->size = size;
1441 if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1442 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
1443 return;
1444 }
1445 if (size != fattr->size)
1446 return;
1447
1448 nfs_fattr_set_barrier(fattr);
1449
1450 fattr->valid |= NFS_ATTR_FATTR_SIZE;
1451}
1452
1453void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1454{
1455 struct nfs_fattr *fattr = &hdr->fattr;
1456 struct inode *inode = hdr->inode;
1457
1458 spin_lock(&inode->i_lock);
1459 nfs_writeback_check_extend(hdr, fattr);
1460 nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1461 spin_unlock(&inode->i_lock);
1462}
1463EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1464
1465
1466
1467
1468static int nfs_writeback_done(struct rpc_task *task,
1469 struct nfs_pgio_header *hdr,
1470 struct inode *inode)
1471{
1472 int status;
1473
1474
1475
1476
1477
1478
1479
1480
1481 status = NFS_PROTO(inode)->write_done(task, hdr);
1482 if (status != 0)
1483 return status;
1484 nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1485
1486 if (hdr->res.verf->committed < hdr->args.stable &&
1487 task->tk_status >= 0) {
1488
1489
1490
1491
1492
1493
1494
1495
1496 static unsigned long complain;
1497
1498
1499 if (time_before(complain, jiffies)) {
1500 dprintk("NFS: faulty NFS server %s:"
1501 " (committed = %d) != (stable = %d)\n",
1502 NFS_SERVER(inode)->nfs_client->cl_hostname,
1503 hdr->res.verf->committed, hdr->args.stable);
1504 complain = jiffies + 300 * HZ;
1505 }
1506 }
1507
1508
1509 if (nfs_should_remove_suid(inode))
1510 nfs_mark_for_revalidate(inode);
1511 return 0;
1512}
1513
1514
1515
1516
1517static void nfs_writeback_result(struct rpc_task *task,
1518 struct nfs_pgio_header *hdr)
1519{
1520 struct nfs_pgio_args *argp = &hdr->args;
1521 struct nfs_pgio_res *resp = &hdr->res;
1522
1523 if (resp->count < argp->count) {
1524 static unsigned long complain;
1525
1526
1527 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1528
1529
1530 if (resp->count == 0) {
1531 if (time_before(complain, jiffies)) {
1532 printk(KERN_WARNING
1533 "NFS: Server wrote zero bytes, expected %u.\n",
1534 argp->count);
1535 complain = jiffies + 300 * HZ;
1536 }
1537 nfs_set_pgio_error(hdr, -EIO, argp->offset);
1538 task->tk_status = -EIO;
1539 return;
1540 }
1541
1542
1543 if (!task->tk_ops) {
1544 hdr->pnfs_error = -EAGAIN;
1545 return;
1546 }
1547
1548
1549 if (resp->verf->committed != NFS_UNSTABLE) {
1550
1551 hdr->mds_offset += resp->count;
1552 argp->offset += resp->count;
1553 argp->pgbase += resp->count;
1554 argp->count -= resp->count;
1555 } else {
1556
1557
1558
1559 argp->stable = NFS_FILE_SYNC;
1560 }
1561 rpc_restart_call_prepare(task);
1562 }
1563}
1564
1565static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
1566{
1567 return wait_on_atomic_t(&cinfo->rpcs_out,
1568 nfs_wait_atomic_killable, TASK_KILLABLE);
1569}
1570
1571static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
1572{
1573 atomic_inc(&cinfo->rpcs_out);
1574}
1575
1576static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
1577{
1578 if (atomic_dec_and_test(&cinfo->rpcs_out))
1579 wake_up_atomic_t(&cinfo->rpcs_out);
1580}
1581
1582void nfs_commitdata_release(struct nfs_commit_data *data)
1583{
1584 put_nfs_open_context(data->context);
1585 nfs_commit_free(data);
1586}
1587EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1588
1589int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1590 const struct nfs_rpc_ops *nfs_ops,
1591 const struct rpc_call_ops *call_ops,
1592 int how, int flags)
1593{
1594 struct rpc_task *task;
1595 int priority = flush_task_priority(how);
1596 struct rpc_message msg = {
1597 .rpc_argp = &data->args,
1598 .rpc_resp = &data->res,
1599 .rpc_cred = data->cred,
1600 };
1601 struct rpc_task_setup task_setup_data = {
1602 .task = &data->task,
1603 .rpc_client = clnt,
1604 .rpc_message = &msg,
1605 .callback_ops = call_ops,
1606 .callback_data = data,
1607 .workqueue = nfsiod_workqueue,
1608 .flags = RPC_TASK_ASYNC | flags,
1609 .priority = priority,
1610 };
1611
1612 nfs_ops->commit_setup(data, &msg);
1613
1614 dprintk("NFS: initiated commit call\n");
1615
1616 nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1617 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1618
1619 task = rpc_run_task(&task_setup_data);
1620 if (IS_ERR(task))
1621 return PTR_ERR(task);
1622 if (how & FLUSH_SYNC)
1623 rpc_wait_for_completion_task(task);
1624 rpc_put_task(task);
1625 return 0;
1626}
1627EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1628
1629static loff_t nfs_get_lwb(struct list_head *head)
1630{
1631 loff_t lwb = 0;
1632 struct nfs_page *req;
1633
1634 list_for_each_entry(req, head, wb_list)
1635 if (lwb < (req_offset(req) + req->wb_bytes))
1636 lwb = req_offset(req) + req->wb_bytes;
1637
1638 return lwb;
1639}
1640
1641
1642
1643
1644void nfs_init_commit(struct nfs_commit_data *data,
1645 struct list_head *head,
1646 struct pnfs_layout_segment *lseg,
1647 struct nfs_commit_info *cinfo)
1648{
1649 struct nfs_page *first = nfs_list_entry(head->next);
1650 struct inode *inode = d_inode(first->wb_context->dentry);
1651
1652
1653
1654
1655 list_splice_init(head, &data->pages);
1656
1657 data->inode = inode;
1658 data->cred = first->wb_context->cred;
1659 data->lseg = lseg;
1660
1661 if (lseg)
1662 data->lwb = nfs_get_lwb(&data->pages);
1663 data->mds_ops = &nfs_commit_ops;
1664 data->completion_ops = cinfo->completion_ops;
1665 data->dreq = cinfo->dreq;
1666
1667 data->args.fh = NFS_FH(data->inode);
1668
1669 data->args.offset = 0;
1670 data->args.count = 0;
1671 data->context = get_nfs_open_context(first->wb_context);
1672 data->res.fattr = &data->fattr;
1673 data->res.verf = &data->verf;
1674 nfs_fattr_init(&data->fattr);
1675}
1676EXPORT_SYMBOL_GPL(nfs_init_commit);
1677
1678void nfs_retry_commit(struct list_head *page_list,
1679 struct pnfs_layout_segment *lseg,
1680 struct nfs_commit_info *cinfo,
1681 u32 ds_commit_idx)
1682{
1683 struct nfs_page *req;
1684
1685 while (!list_empty(page_list)) {
1686 req = nfs_list_entry(page_list->next);
1687 nfs_list_remove_request(req);
1688 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
1689 if (!cinfo->dreq)
1690 nfs_clear_page_commit(req->wb_page);
1691 nfs_unlock_and_release_request(req);
1692 }
1693}
1694EXPORT_SYMBOL_GPL(nfs_retry_commit);
1695
1696static void
1697nfs_commit_resched_write(struct nfs_commit_info *cinfo,
1698 struct nfs_page *req)
1699{
1700 __set_page_dirty_nobuffers(req->wb_page);
1701}
1702
1703
1704
1705
1706static int
1707nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1708 struct nfs_commit_info *cinfo)
1709{
1710 struct nfs_commit_data *data;
1711
1712 data = nfs_commitdata_alloc();
1713
1714 if (!data)
1715 goto out_bad;
1716
1717
1718 nfs_init_commit(data, head, NULL, cinfo);
1719 atomic_inc(&cinfo->mds->rpcs_out);
1720 return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1721 data->mds_ops, how, 0);
1722 out_bad:
1723 nfs_retry_commit(head, NULL, cinfo, 0);
1724 return -ENOMEM;
1725}
1726
1727
1728
1729
1730static void nfs_commit_done(struct rpc_task *task, void *calldata)
1731{
1732 struct nfs_commit_data *data = calldata;
1733
1734 dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1735 task->tk_pid, task->tk_status);
1736
1737
1738 NFS_PROTO(data->inode)->commit_done(task, data);
1739}
1740
1741static void nfs_commit_release_pages(struct nfs_commit_data *data)
1742{
1743 struct nfs_page *req;
1744 int status = data->task.tk_status;
1745 struct nfs_commit_info cinfo;
1746 struct nfs_server *nfss;
1747
1748 while (!list_empty(&data->pages)) {
1749 req = nfs_list_entry(data->pages.next);
1750 nfs_list_remove_request(req);
1751 nfs_clear_page_commit(req->wb_page);
1752
1753 dprintk("NFS: commit (%s/%llu %d@%lld)",
1754 req->wb_context->dentry->d_sb->s_id,
1755 (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
1756 req->wb_bytes,
1757 (long long)req_offset(req));
1758 if (status < 0) {
1759 nfs_context_set_write_error(req->wb_context, status);
1760 nfs_inode_remove_request(req);
1761 dprintk(", error = %d\n", status);
1762 goto next;
1763 }
1764
1765
1766
1767 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
1768
1769 nfs_inode_remove_request(req);
1770 dprintk(" OK\n");
1771 goto next;
1772 }
1773
1774 dprintk(" mismatch\n");
1775 nfs_mark_request_dirty(req);
1776 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1777 next:
1778 nfs_unlock_and_release_request(req);
1779 }
1780 nfss = NFS_SERVER(data->inode);
1781 if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1782 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1783
1784 nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1785 nfs_commit_end(cinfo.mds);
1786}
1787
1788static void nfs_commit_release(void *calldata)
1789{
1790 struct nfs_commit_data *data = calldata;
1791
1792 data->completion_ops->completion(data);
1793 nfs_commitdata_release(calldata);
1794}
1795
1796static const struct rpc_call_ops nfs_commit_ops = {
1797 .rpc_call_prepare = nfs_commit_prepare,
1798 .rpc_call_done = nfs_commit_done,
1799 .rpc_release = nfs_commit_release,
1800};
1801
1802static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1803 .completion = nfs_commit_release_pages,
1804 .resched_write = nfs_commit_resched_write,
1805};
1806
1807int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1808 int how, struct nfs_commit_info *cinfo)
1809{
1810 int status;
1811
1812 status = pnfs_commit_list(inode, head, how, cinfo);
1813 if (status == PNFS_NOT_ATTEMPTED)
1814 status = nfs_commit_list(inode, head, how, cinfo);
1815 return status;
1816}
1817
1818int nfs_commit_inode(struct inode *inode, int how)
1819{
1820 LIST_HEAD(head);
1821 struct nfs_commit_info cinfo;
1822 int may_wait = how & FLUSH_SYNC;
1823 int error = 0;
1824 int res;
1825
1826 nfs_init_cinfo_from_inode(&cinfo, inode);
1827 nfs_commit_begin(cinfo.mds);
1828 res = nfs_scan_commit(inode, &head, &cinfo);
1829 if (res)
1830 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1831 nfs_commit_end(cinfo.mds);
1832 if (error < 0)
1833 goto out_error;
1834 if (!may_wait)
1835 goto out_mark_dirty;
1836 error = wait_on_commit(cinfo.mds);
1837 if (error < 0)
1838 return error;
1839 return res;
1840out_error:
1841 res = error;
1842
1843
1844
1845
1846
1847out_mark_dirty:
1848 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1849 return res;
1850}
1851EXPORT_SYMBOL_GPL(nfs_commit_inode);
1852
1853int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1854{
1855 struct nfs_inode *nfsi = NFS_I(inode);
1856 int flags = FLUSH_SYNC;
1857 int ret = 0;
1858
1859
1860 if (!nfsi->commit_info.ncommit)
1861 return ret;
1862
1863 if (wbc->sync_mode == WB_SYNC_NONE) {
1864
1865
1866
1867 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
1868 goto out_mark_dirty;
1869
1870
1871 flags = 0;
1872 }
1873
1874 ret = nfs_commit_inode(inode, flags);
1875 if (ret >= 0) {
1876 if (wbc->sync_mode == WB_SYNC_NONE) {
1877 if (ret < wbc->nr_to_write)
1878 wbc->nr_to_write -= ret;
1879 else
1880 wbc->nr_to_write = 0;
1881 }
1882 return 0;
1883 }
1884out_mark_dirty:
1885 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1886 return ret;
1887}
1888EXPORT_SYMBOL_GPL(nfs_write_inode);
1889
1890
1891
1892
1893int nfs_wb_all(struct inode *inode)
1894{
1895 int ret;
1896
1897 trace_nfs_writeback_inode_enter(inode);
1898
1899 ret = filemap_write_and_wait(inode->i_mapping);
1900 if (ret)
1901 goto out;
1902 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1903 if (ret < 0)
1904 goto out;
1905 pnfs_sync_inode(inode, true);
1906 ret = 0;
1907
1908out:
1909 trace_nfs_writeback_inode_exit(inode, ret);
1910 return ret;
1911}
1912EXPORT_SYMBOL_GPL(nfs_wb_all);
1913
1914int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1915{
1916 struct nfs_page *req;
1917 int ret = 0;
1918
1919 wait_on_page_writeback(page);
1920
1921
1922
1923 req = nfs_lock_and_join_requests(page, false);
1924
1925 if (IS_ERR(req)) {
1926 ret = PTR_ERR(req);
1927 } else if (req) {
1928
1929
1930
1931
1932 nfs_inode_remove_request(req);
1933 nfs_unlock_and_release_request(req);
1934 }
1935
1936 return ret;
1937}
1938
1939
1940
1941
1942int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder)
1943{
1944 loff_t range_start = page_file_offset(page);
1945 loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
1946 struct writeback_control wbc = {
1947 .sync_mode = WB_SYNC_ALL,
1948 .nr_to_write = 0,
1949 .range_start = range_start,
1950 .range_end = range_end,
1951 };
1952 int ret;
1953
1954 trace_nfs_writeback_page_enter(inode);
1955
1956 for (;;) {
1957 wait_on_page_writeback(page);
1958 if (clear_page_dirty_for_io(page)) {
1959 ret = nfs_writepage_locked(page, &wbc, launder);
1960 if (ret < 0)
1961 goto out_error;
1962 continue;
1963 }
1964 ret = 0;
1965 if (!PagePrivate(page))
1966 break;
1967 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1968 if (ret < 0)
1969 goto out_error;
1970 }
1971out_error:
1972 trace_nfs_writeback_page_exit(inode, ret);
1973 return ret;
1974}
1975
1976#ifdef CONFIG_MIGRATION
1977int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1978 struct page *page, enum migrate_mode mode)
1979{
1980
1981
1982
1983
1984
1985
1986
1987
1988 if (PagePrivate(page))
1989 return -EBUSY;
1990
1991 if (!nfs_fscache_release_page(page, GFP_KERNEL))
1992 return -EBUSY;
1993
1994 return migrate_page(mapping, newpage, page, mode);
1995}
1996#endif
1997
1998int __init nfs_init_writepagecache(void)
1999{
2000 nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
2001 sizeof(struct nfs_pgio_header),
2002 0, SLAB_HWCACHE_ALIGN,
2003 NULL);
2004 if (nfs_wdata_cachep == NULL)
2005 return -ENOMEM;
2006
2007 nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2008 nfs_wdata_cachep);
2009 if (nfs_wdata_mempool == NULL)
2010 goto out_destroy_write_cache;
2011
2012 nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2013 sizeof(struct nfs_commit_data),
2014 0, SLAB_HWCACHE_ALIGN,
2015 NULL);
2016 if (nfs_cdata_cachep == NULL)
2017 goto out_destroy_write_mempool;
2018
2019 nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
2020 nfs_cdata_cachep);
2021 if (nfs_commit_mempool == NULL)
2022 goto out_destroy_commit_cache;
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040 nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2041 if (nfs_congestion_kb > 256*1024)
2042 nfs_congestion_kb = 256*1024;
2043
2044 return 0;
2045
2046out_destroy_commit_cache:
2047 kmem_cache_destroy(nfs_cdata_cachep);
2048out_destroy_write_mempool:
2049 mempool_destroy(nfs_wdata_mempool);
2050out_destroy_write_cache:
2051 kmem_cache_destroy(nfs_wdata_cachep);
2052 return -ENOMEM;
2053}
2054
2055void nfs_destroy_writepagecache(void)
2056{
2057 mempool_destroy(nfs_commit_mempool);
2058 kmem_cache_destroy(nfs_cdata_cachep);
2059 mempool_destroy(nfs_wdata_mempool);
2060 kmem_cache_destroy(nfs_wdata_cachep);
2061}
2062
2063static const struct nfs_rw_ops nfs_rw_write_ops = {
2064 .rw_mode = FMODE_WRITE,
2065 .rw_alloc_header = nfs_writehdr_alloc,
2066 .rw_free_header = nfs_writehdr_free,
2067 .rw_done = nfs_writeback_done,
2068 .rw_result = nfs_writeback_result,
2069 .rw_initiate = nfs_initiate_write,
2070};
2071