This source file includes following definitions.
- nfs_pgio_current_mirror
- nfs_pgheader_init
- nfs_set_pgio_error
- nfs_page_alloc
- nfs_page_free
- nfs_iocounter_wait
- nfs_async_iocounter_wait
- nfs_page_group_lock
- nfs_page_group_unlock
- nfs_page_group_sync_on_bit_locked
- nfs_page_group_sync_on_bit
- nfs_page_group_init
- nfs_page_group_destroy
- __nfs_create_request
- nfs_create_request
- nfs_create_subreq
- nfs_unlock_request
- nfs_unlock_and_release_request
- nfs_clear_request
- nfs_free_request
- nfs_release_request
- nfs_wait_on_request
- nfs_generic_pg_test
- nfs_pgio_header_alloc
- nfs_pgio_data_destroy
- nfs_pgio_header_free
- nfs_pgio_rpcsetup
- nfs_pgio_prepare
- nfs_initiate_pgio
- nfs_pgio_error
- nfs_pgio_release
- nfs_pageio_mirror_init
- nfs_pageio_init
- nfs_pgio_result
- nfs_generic_pgio
- nfs_generic_pg_pgios
- nfs_pageio_alloc_mirrors
- nfs_pageio_setup_mirroring
- nfs_pageio_cleanup_mirroring
- nfs_match_lock_context
- nfs_can_coalesce_requests
- nfs_pageio_do_add_request
- nfs_pageio_doio
- nfs_pageio_cleanup_request
- __nfs_pageio_add_request
- nfs_do_recoalesce
- nfs_pageio_add_request_mirror
- nfs_pageio_error_cleanup
- nfs_pageio_add_request
- nfs_pageio_complete_mirror
- nfs_pageio_resend
- nfs_pageio_complete
- nfs_pageio_cond_complete
- nfs_pageio_stop_mirroring
- nfs_init_nfspagecache
- nfs_destroy_nfspagecache
1
2
3
4
5
6
7
8
9
10
11
12
13 #include <linux/slab.h>
14 #include <linux/file.h>
15 #include <linux/sched.h>
16 #include <linux/sunrpc/clnt.h>
17 #include <linux/nfs.h>
18 #include <linux/nfs3.h>
19 #include <linux/nfs4.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_page.h>
22 #include <linux/nfs_mount.h>
23 #include <linux/export.h>
24
25 #include "internal.h"
26 #include "pnfs.h"
27
28 #define NFSDBG_FACILITY NFSDBG_PAGECACHE
29
30 static struct kmem_cache *nfs_page_cachep;
31 static const struct rpc_call_ops nfs_pgio_common_ops;
32
33 struct nfs_pgio_mirror *
34 nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
35 {
36 return nfs_pgio_has_mirroring(desc) ?
37 &desc->pg_mirrors[desc->pg_mirror_idx] :
38 &desc->pg_mirrors[0];
39 }
40 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
41
42 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
43 struct nfs_pgio_header *hdr,
44 void (*release)(struct nfs_pgio_header *hdr))
45 {
46 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
47
48
49 hdr->req = nfs_list_entry(mirror->pg_list.next);
50 hdr->inode = desc->pg_inode;
51 hdr->cred = nfs_req_openctx(hdr->req)->cred;
52 hdr->io_start = req_offset(hdr->req);
53 hdr->good_bytes = mirror->pg_count;
54 hdr->io_completion = desc->pg_io_completion;
55 hdr->dreq = desc->pg_dreq;
56 hdr->release = release;
57 hdr->completion_ops = desc->pg_completion_ops;
58 if (hdr->completion_ops->init_hdr)
59 hdr->completion_ops->init_hdr(hdr);
60
61 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
62 }
63 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
64
65 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
66 {
67 unsigned int new = pos - hdr->io_start;
68
69 if (hdr->good_bytes > new) {
70 hdr->good_bytes = new;
71 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
72 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
73 hdr->error = error;
74 }
75 }
76
77 static inline struct nfs_page *
78 nfs_page_alloc(void)
79 {
80 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
81 if (p)
82 INIT_LIST_HEAD(&p->wb_list);
83 return p;
84 }
85
86 static inline void
87 nfs_page_free(struct nfs_page *p)
88 {
89 kmem_cache_free(nfs_page_cachep, p);
90 }
91
92
93
94
95
96
97
98
99 int
100 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
101 {
102 return wait_var_event_killable(&l_ctx->io_count,
103 !atomic_read(&l_ctx->io_count));
104 }
105
106
107
108
109
110
111
112
113
114
115 bool
116 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
117 {
118 struct inode *inode = d_inode(l_ctx->open_context->dentry);
119 bool ret = false;
120
121 if (atomic_read(&l_ctx->io_count) > 0) {
122 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
123 ret = true;
124 }
125
126 if (atomic_read(&l_ctx->io_count) == 0) {
127 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
128 ret = false;
129 }
130
131 return ret;
132 }
133 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
134
135
136
137
138
139
140
141
142
143
144 int
145 nfs_page_group_lock(struct nfs_page *req)
146 {
147 struct nfs_page *head = req->wb_head;
148
149 WARN_ON_ONCE(head != head->wb_head);
150
151 if (!test_and_set_bit(PG_HEADLOCK, &head->wb_flags))
152 return 0;
153
154 set_bit(PG_CONTENDED1, &head->wb_flags);
155 smp_mb__after_atomic();
156 return wait_on_bit_lock(&head->wb_flags, PG_HEADLOCK,
157 TASK_UNINTERRUPTIBLE);
158 }
159
160
161
162
163
164 void
165 nfs_page_group_unlock(struct nfs_page *req)
166 {
167 struct nfs_page *head = req->wb_head;
168
169 WARN_ON_ONCE(head != head->wb_head);
170
171 smp_mb__before_atomic();
172 clear_bit(PG_HEADLOCK, &head->wb_flags);
173 smp_mb__after_atomic();
174 if (!test_bit(PG_CONTENDED1, &head->wb_flags))
175 return;
176 wake_up_bit(&head->wb_flags, PG_HEADLOCK);
177 }
178
179
180
181
182
183
184 static bool
185 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
186 {
187 struct nfs_page *head = req->wb_head;
188 struct nfs_page *tmp;
189
190 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
191 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
192
193 tmp = req->wb_this_page;
194 while (tmp != req) {
195 if (!test_bit(bit, &tmp->wb_flags))
196 return false;
197 tmp = tmp->wb_this_page;
198 }
199
200
201 tmp = req;
202 do {
203 clear_bit(bit, &tmp->wb_flags);
204 tmp = tmp->wb_this_page;
205 } while (tmp != req);
206
207 return true;
208 }
209
210
211
212
213
214
215
216 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
217 {
218 bool ret;
219
220 nfs_page_group_lock(req);
221 ret = nfs_page_group_sync_on_bit_locked(req, bit);
222 nfs_page_group_unlock(req);
223
224 return ret;
225 }
226
227
228
229
230
231
232
233 static inline void
234 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
235 {
236 struct inode *inode;
237 WARN_ON_ONCE(prev == req);
238
239 if (!prev) {
240
241 req->wb_head = req;
242 req->wb_this_page = req;
243 } else {
244
245 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
246 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
247 req->wb_head = prev->wb_head;
248 req->wb_this_page = prev->wb_this_page;
249 prev->wb_this_page = req;
250
251
252
253 kref_get(&req->wb_head->wb_kref);
254
255
256
257
258 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
259 inode = page_file_mapping(req->wb_page)->host;
260 set_bit(PG_INODE_REF, &req->wb_flags);
261 kref_get(&req->wb_kref);
262 atomic_long_inc(&NFS_I(inode)->nrequests);
263 }
264 }
265 }
266
267
268
269
270
271
272
273
274 static void
275 nfs_page_group_destroy(struct kref *kref)
276 {
277 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
278 struct nfs_page *head = req->wb_head;
279 struct nfs_page *tmp, *next;
280
281 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
282 goto out;
283
284 tmp = req;
285 do {
286 next = tmp->wb_this_page;
287
288 tmp->wb_this_page = tmp;
289 tmp->wb_head = tmp;
290 nfs_free_request(tmp);
291 tmp = next;
292 } while (tmp != req);
293 out:
294
295 if (head != req)
296 nfs_release_request(head);
297 }
298
299 static struct nfs_page *
300 __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
301 unsigned int pgbase, unsigned int offset,
302 unsigned int count)
303 {
304 struct nfs_page *req;
305 struct nfs_open_context *ctx = l_ctx->open_context;
306
307 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
308 return ERR_PTR(-EBADF);
309
310 req = nfs_page_alloc();
311 if (req == NULL)
312 return ERR_PTR(-ENOMEM);
313
314 req->wb_lock_context = l_ctx;
315 refcount_inc(&l_ctx->count);
316 atomic_inc(&l_ctx->io_count);
317
318
319
320
321 req->wb_page = page;
322 if (page) {
323 req->wb_index = page_index(page);
324 get_page(page);
325 }
326 req->wb_offset = offset;
327 req->wb_pgbase = pgbase;
328 req->wb_bytes = count;
329 kref_init(&req->wb_kref);
330 req->wb_nio = 0;
331 return req;
332 }
333
334
335
336
337
338
339
340
341
342
343
344
345 struct nfs_page *
346 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
347 unsigned int offset, unsigned int count)
348 {
349 struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
350 struct nfs_page *ret;
351
352 if (IS_ERR(l_ctx))
353 return ERR_CAST(l_ctx);
354 ret = __nfs_create_request(l_ctx, page, offset, offset, count);
355 if (!IS_ERR(ret))
356 nfs_page_group_init(ret, NULL);
357 nfs_put_lock_context(l_ctx);
358 return ret;
359 }
360
361 static struct nfs_page *
362 nfs_create_subreq(struct nfs_page *req, struct nfs_page *last,
363 unsigned int pgbase, unsigned int offset,
364 unsigned int count)
365 {
366 struct nfs_page *ret;
367
368 ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
369 pgbase, offset, count);
370 if (!IS_ERR(ret)) {
371 nfs_lock_request(ret);
372 ret->wb_index = req->wb_index;
373 nfs_page_group_init(ret, last);
374 ret->wb_nio = req->wb_nio;
375 }
376 return ret;
377 }
378
379
380
381
382
383 void nfs_unlock_request(struct nfs_page *req)
384 {
385 if (!NFS_WBACK_BUSY(req)) {
386 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
387 BUG();
388 }
389 smp_mb__before_atomic();
390 clear_bit(PG_BUSY, &req->wb_flags);
391 smp_mb__after_atomic();
392 if (!test_bit(PG_CONTENDED2, &req->wb_flags))
393 return;
394 wake_up_bit(&req->wb_flags, PG_BUSY);
395 }
396
397
398
399
400
401 void nfs_unlock_and_release_request(struct nfs_page *req)
402 {
403 nfs_unlock_request(req);
404 nfs_release_request(req);
405 }
406
407
408
409
410
411
412
413
414 static void nfs_clear_request(struct nfs_page *req)
415 {
416 struct page *page = req->wb_page;
417 struct nfs_lock_context *l_ctx = req->wb_lock_context;
418 struct nfs_open_context *ctx;
419
420 if (page != NULL) {
421 put_page(page);
422 req->wb_page = NULL;
423 }
424 if (l_ctx != NULL) {
425 if (atomic_dec_and_test(&l_ctx->io_count)) {
426 wake_up_var(&l_ctx->io_count);
427 ctx = l_ctx->open_context;
428 if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
429 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
430 }
431 nfs_put_lock_context(l_ctx);
432 req->wb_lock_context = NULL;
433 }
434 }
435
436
437
438
439
440
441
442 void nfs_free_request(struct nfs_page *req)
443 {
444 WARN_ON_ONCE(req->wb_this_page != req);
445
446
447 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
448 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
449 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
450 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
451 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
452
453
454 nfs_clear_request(req);
455 nfs_page_free(req);
456 }
457
458 void nfs_release_request(struct nfs_page *req)
459 {
460 kref_put(&req->wb_kref, nfs_page_group_destroy);
461 }
462 EXPORT_SYMBOL_GPL(nfs_release_request);
463
464
465
466
467
468
469
470
471 int
472 nfs_wait_on_request(struct nfs_page *req)
473 {
474 if (!test_bit(PG_BUSY, &req->wb_flags))
475 return 0;
476 set_bit(PG_CONTENDED2, &req->wb_flags);
477 smp_mb__after_atomic();
478 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
479 TASK_UNINTERRUPTIBLE);
480 }
481 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
482
483
484
485
486
487
488
489
490
491
492 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
493 struct nfs_page *prev, struct nfs_page *req)
494 {
495 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
496
497
498 if (mirror->pg_count > mirror->pg_bsize) {
499
500 WARN_ON_ONCE(1);
501 return 0;
502 }
503
504
505
506
507
508 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
509 sizeof(struct page *) > PAGE_SIZE)
510 return 0;
511
512 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
513 }
514 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
515
516 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
517 {
518 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
519
520 if (hdr) {
521 INIT_LIST_HEAD(&hdr->pages);
522 hdr->rw_ops = ops;
523 }
524 return hdr;
525 }
526 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
527
528
529
530
531
532
533
534
535
536 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
537 {
538 if (hdr->args.context)
539 put_nfs_open_context(hdr->args.context);
540 if (hdr->page_array.pagevec != hdr->page_array.page_array)
541 kfree(hdr->page_array.pagevec);
542 }
543
544
545
546
547
548 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
549 {
550 nfs_pgio_data_destroy(hdr);
551 hdr->rw_ops->rw_free_header(hdr);
552 }
553 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
554
555
556
557
558
559
560
561
562 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
563 unsigned int count,
564 int how, struct nfs_commit_info *cinfo)
565 {
566 struct nfs_page *req = hdr->req;
567
568
569
570
571 hdr->args.fh = NFS_FH(hdr->inode);
572 hdr->args.offset = req_offset(req);
573
574 hdr->mds_offset = hdr->args.offset;
575 hdr->args.pgbase = req->wb_pgbase;
576 hdr->args.pages = hdr->page_array.pagevec;
577 hdr->args.count = count;
578 hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
579 hdr->args.lock_context = req->wb_lock_context;
580 hdr->args.stable = NFS_UNSTABLE;
581 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
582 case 0:
583 break;
584 case FLUSH_COND_STABLE:
585 if (nfs_reqs_to_commit(cinfo))
586 break;
587
588 default:
589 hdr->args.stable = NFS_FILE_SYNC;
590 }
591
592 hdr->res.fattr = &hdr->fattr;
593 hdr->res.count = 0;
594 hdr->res.eof = 0;
595 hdr->res.verf = &hdr->verf;
596 nfs_fattr_init(&hdr->fattr);
597 }
598
599
600
601
602
603
604 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
605 {
606 struct nfs_pgio_header *hdr = calldata;
607 int err;
608 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
609 if (err)
610 rpc_exit(task, err);
611 }
612
613 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
614 const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
615 const struct rpc_call_ops *call_ops, int how, int flags)
616 {
617 struct rpc_task *task;
618 struct rpc_message msg = {
619 .rpc_argp = &hdr->args,
620 .rpc_resp = &hdr->res,
621 .rpc_cred = cred,
622 };
623 struct rpc_task_setup task_setup_data = {
624 .rpc_client = clnt,
625 .task = &hdr->task,
626 .rpc_message = &msg,
627 .callback_ops = call_ops,
628 .callback_data = hdr,
629 .workqueue = nfsiod_workqueue,
630 .flags = RPC_TASK_ASYNC | flags,
631 };
632 int ret = 0;
633
634 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
635
636 dprintk("NFS: initiated pgio call "
637 "(req %s/%llu, %u bytes @ offset %llu)\n",
638 hdr->inode->i_sb->s_id,
639 (unsigned long long)NFS_FILEID(hdr->inode),
640 hdr->args.count,
641 (unsigned long long)hdr->args.offset);
642
643 task = rpc_run_task(&task_setup_data);
644 if (IS_ERR(task)) {
645 ret = PTR_ERR(task);
646 goto out;
647 }
648 if (how & FLUSH_SYNC) {
649 ret = rpc_wait_for_completion_task(task);
650 if (ret == 0)
651 ret = task->tk_status;
652 }
653 rpc_put_task(task);
654 out:
655 return ret;
656 }
657 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
658
659
660
661
662
663 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
664 {
665 set_bit(NFS_IOHDR_REDO, &hdr->flags);
666 hdr->completion_ops->completion(hdr);
667 }
668
669
670
671
672
673 static void nfs_pgio_release(void *calldata)
674 {
675 struct nfs_pgio_header *hdr = calldata;
676 hdr->completion_ops->completion(hdr);
677 }
678
679 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
680 unsigned int bsize)
681 {
682 INIT_LIST_HEAD(&mirror->pg_list);
683 mirror->pg_bytes_written = 0;
684 mirror->pg_count = 0;
685 mirror->pg_bsize = bsize;
686 mirror->pg_base = 0;
687 mirror->pg_recoalesce = 0;
688 }
689
690
691
692
693
694
695
696
697
698
699
700 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
701 struct inode *inode,
702 const struct nfs_pageio_ops *pg_ops,
703 const struct nfs_pgio_completion_ops *compl_ops,
704 const struct nfs_rw_ops *rw_ops,
705 size_t bsize,
706 int io_flags)
707 {
708 desc->pg_moreio = 0;
709 desc->pg_inode = inode;
710 desc->pg_ops = pg_ops;
711 desc->pg_completion_ops = compl_ops;
712 desc->pg_rw_ops = rw_ops;
713 desc->pg_ioflags = io_flags;
714 desc->pg_error = 0;
715 desc->pg_lseg = NULL;
716 desc->pg_io_completion = NULL;
717 desc->pg_dreq = NULL;
718 desc->pg_bsize = bsize;
719
720 desc->pg_mirror_count = 1;
721 desc->pg_mirror_idx = 0;
722
723 desc->pg_mirrors_dynamic = NULL;
724 desc->pg_mirrors = desc->pg_mirrors_static;
725 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
726 desc->pg_maxretrans = 0;
727 }
728
729
730
731
732
733
734 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
735 {
736 struct nfs_pgio_header *hdr = calldata;
737 struct inode *inode = hdr->inode;
738
739 dprintk("NFS: %s: %5u, (status %d)\n", __func__,
740 task->tk_pid, task->tk_status);
741
742 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
743 return;
744 if (task->tk_status < 0)
745 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
746 else
747 hdr->rw_ops->rw_result(task, hdr);
748 }
749
750
751
752
753
754
755
756
757
758 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
759 struct nfs_pgio_header *hdr)
760 {
761 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
762
763 struct nfs_page *req;
764 struct page **pages,
765 *last_page;
766 struct list_head *head = &mirror->pg_list;
767 struct nfs_commit_info cinfo;
768 struct nfs_page_array *pg_array = &hdr->page_array;
769 unsigned int pagecount, pageused;
770 gfp_t gfp_flags = GFP_KERNEL;
771
772 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
773 pg_array->npages = pagecount;
774
775 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
776 pg_array->pagevec = pg_array->page_array;
777 else {
778 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
779 if (!pg_array->pagevec) {
780 pg_array->npages = 0;
781 nfs_pgio_error(hdr);
782 desc->pg_error = -ENOMEM;
783 return desc->pg_error;
784 }
785 }
786
787 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
788 pages = hdr->page_array.pagevec;
789 last_page = NULL;
790 pageused = 0;
791 while (!list_empty(head)) {
792 req = nfs_list_entry(head->next);
793 nfs_list_move_request(req, &hdr->pages);
794
795 if (!last_page || last_page != req->wb_page) {
796 pageused++;
797 if (pageused > pagecount)
798 break;
799 *pages++ = last_page = req->wb_page;
800 }
801 }
802 if (WARN_ON_ONCE(pageused != pagecount)) {
803 nfs_pgio_error(hdr);
804 desc->pg_error = -EINVAL;
805 return desc->pg_error;
806 }
807
808 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
809 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
810 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
811
812
813 nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
814 desc->pg_rpc_callops = &nfs_pgio_common_ops;
815 return 0;
816 }
817 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
818
819 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
820 {
821 struct nfs_pgio_header *hdr;
822 int ret;
823
824 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
825 if (!hdr) {
826 desc->pg_error = -ENOMEM;
827 return desc->pg_error;
828 }
829 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
830 ret = nfs_generic_pgio(desc, hdr);
831 if (ret == 0)
832 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
833 hdr,
834 hdr->cred,
835 NFS_PROTO(hdr->inode),
836 desc->pg_rpc_callops,
837 desc->pg_ioflags, 0);
838 return ret;
839 }
840
841 static struct nfs_pgio_mirror *
842 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
843 unsigned int mirror_count)
844 {
845 struct nfs_pgio_mirror *ret;
846 unsigned int i;
847
848 kfree(desc->pg_mirrors_dynamic);
849 desc->pg_mirrors_dynamic = NULL;
850 if (mirror_count == 1)
851 return desc->pg_mirrors_static;
852 ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_KERNEL);
853 if (ret != NULL) {
854 for (i = 0; i < mirror_count; i++)
855 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
856 desc->pg_mirrors_dynamic = ret;
857 }
858 return ret;
859 }
860
861
862
863
864
865 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
866 struct nfs_page *req)
867 {
868 unsigned int mirror_count = 1;
869
870 if (pgio->pg_ops->pg_get_mirror_count)
871 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
872 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
873 return;
874
875 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
876 pgio->pg_error = -EINVAL;
877 return;
878 }
879
880 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
881 if (pgio->pg_mirrors == NULL) {
882 pgio->pg_error = -ENOMEM;
883 pgio->pg_mirrors = pgio->pg_mirrors_static;
884 mirror_count = 1;
885 }
886 pgio->pg_mirror_count = mirror_count;
887 }
888
889 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
890 {
891 pgio->pg_mirror_count = 1;
892 pgio->pg_mirror_idx = 0;
893 pgio->pg_mirrors = pgio->pg_mirrors_static;
894 kfree(pgio->pg_mirrors_dynamic);
895 pgio->pg_mirrors_dynamic = NULL;
896 }
897
898 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
899 const struct nfs_lock_context *l2)
900 {
901 return l1->lockowner == l2->lockowner;
902 }
903
904
905
906
907
908
909
910
911
912
913
914
915
916 static bool nfs_can_coalesce_requests(struct nfs_page *prev,
917 struct nfs_page *req,
918 struct nfs_pageio_descriptor *pgio)
919 {
920 size_t size;
921 struct file_lock_context *flctx;
922
923 if (prev) {
924 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
925 return false;
926 flctx = d_inode(nfs_req_openctx(req)->dentry)->i_flctx;
927 if (flctx != NULL &&
928 !(list_empty_careful(&flctx->flc_posix) &&
929 list_empty_careful(&flctx->flc_flock)) &&
930 !nfs_match_lock_context(req->wb_lock_context,
931 prev->wb_lock_context))
932 return false;
933 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
934 return false;
935 if (req->wb_page == prev->wb_page) {
936 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
937 return false;
938 } else {
939 if (req->wb_pgbase != 0 ||
940 prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
941 return false;
942 }
943 }
944 size = pgio->pg_ops->pg_test(pgio, prev, req);
945 WARN_ON_ONCE(size > req->wb_bytes);
946 if (size && size < req->wb_bytes)
947 req->wb_bytes = size;
948 return size > 0;
949 }
950
951
952
953
954
955
956
957
958
959 static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
960 struct nfs_page *req)
961 {
962 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
963
964 struct nfs_page *prev = NULL;
965
966 if (mirror->pg_count != 0) {
967 prev = nfs_list_entry(mirror->pg_list.prev);
968 } else {
969 if (desc->pg_ops->pg_init)
970 desc->pg_ops->pg_init(desc, req);
971 if (desc->pg_error < 0)
972 return 0;
973 mirror->pg_base = req->wb_pgbase;
974 }
975
976 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
977 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
978 desc->pg_error = -ETIMEDOUT;
979 else
980 desc->pg_error = -EIO;
981 return 0;
982 }
983
984 if (!nfs_can_coalesce_requests(prev, req, desc))
985 return 0;
986 nfs_list_move_request(req, &mirror->pg_list);
987 mirror->pg_count += req->wb_bytes;
988 return 1;
989 }
990
991
992
993
994 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
995 {
996 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
997
998
999 if (!list_empty(&mirror->pg_list)) {
1000 int error = desc->pg_ops->pg_doio(desc);
1001 if (error < 0)
1002 desc->pg_error = error;
1003 else
1004 mirror->pg_bytes_written += mirror->pg_count;
1005 }
1006 if (list_empty(&mirror->pg_list)) {
1007 mirror->pg_count = 0;
1008 mirror->pg_base = 0;
1009 }
1010 }
1011
1012 static void
1013 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1014 struct nfs_page *req)
1015 {
1016 LIST_HEAD(head);
1017
1018 nfs_list_move_request(req, &head);
1019 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1020 }
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1034 struct nfs_page *req)
1035 {
1036 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1037
1038 struct nfs_page *subreq;
1039 unsigned int bytes_left = 0;
1040 unsigned int offset, pgbase;
1041
1042 nfs_page_group_lock(req);
1043
1044 subreq = req;
1045 bytes_left = subreq->wb_bytes;
1046 offset = subreq->wb_offset;
1047 pgbase = subreq->wb_pgbase;
1048
1049 do {
1050 if (!nfs_pageio_do_add_request(desc, subreq)) {
1051
1052 WARN_ON_ONCE(subreq->wb_bytes != bytes_left);
1053 WARN_ON_ONCE(subreq->wb_offset != offset);
1054 WARN_ON_ONCE(subreq->wb_pgbase != pgbase);
1055
1056 nfs_page_group_unlock(req);
1057 desc->pg_moreio = 1;
1058 nfs_pageio_doio(desc);
1059 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1060 goto out_cleanup_subreq;
1061
1062 nfs_page_group_lock(req);
1063 continue;
1064 }
1065
1066
1067 WARN_ON_ONCE(subreq->wb_bytes + subreq->wb_pgbase > PAGE_SIZE);
1068 WARN_ON_ONCE(subreq->wb_bytes > bytes_left);
1069 WARN_ON_ONCE(subreq->wb_bytes == 0);
1070
1071 bytes_left -= subreq->wb_bytes;
1072 offset += subreq->wb_bytes;
1073 pgbase += subreq->wb_bytes;
1074
1075 if (bytes_left) {
1076 subreq = nfs_create_subreq(req, subreq, pgbase,
1077 offset, bytes_left);
1078 if (IS_ERR(subreq))
1079 goto err_ptr;
1080 }
1081 } while (bytes_left > 0);
1082
1083 nfs_page_group_unlock(req);
1084 return 1;
1085 err_ptr:
1086 desc->pg_error = PTR_ERR(subreq);
1087 nfs_page_group_unlock(req);
1088 return 0;
1089 out_cleanup_subreq:
1090 if (req != subreq)
1091 nfs_pageio_cleanup_request(desc, subreq);
1092 return 0;
1093 }
1094
1095 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1096 {
1097 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1098 LIST_HEAD(head);
1099
1100 do {
1101 list_splice_init(&mirror->pg_list, &head);
1102 mirror->pg_bytes_written -= mirror->pg_count;
1103 mirror->pg_count = 0;
1104 mirror->pg_base = 0;
1105 mirror->pg_recoalesce = 0;
1106
1107 while (!list_empty(&head)) {
1108 struct nfs_page *req;
1109
1110 req = list_first_entry(&head, struct nfs_page, wb_list);
1111 if (__nfs_pageio_add_request(desc, req))
1112 continue;
1113 if (desc->pg_error < 0) {
1114 list_splice_tail(&head, &mirror->pg_list);
1115 mirror->pg_recoalesce = 1;
1116 return 0;
1117 }
1118 break;
1119 }
1120 } while (mirror->pg_recoalesce);
1121 return 1;
1122 }
1123
1124 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1125 struct nfs_page *req)
1126 {
1127 int ret;
1128
1129 do {
1130 ret = __nfs_pageio_add_request(desc, req);
1131 if (ret)
1132 break;
1133 if (desc->pg_error < 0)
1134 break;
1135 ret = nfs_do_recoalesce(desc);
1136 } while (ret);
1137
1138 return ret;
1139 }
1140
1141 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1142 {
1143 u32 midx;
1144 struct nfs_pgio_mirror *mirror;
1145
1146 if (!desc->pg_error)
1147 return;
1148
1149 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1150 mirror = &desc->pg_mirrors[midx];
1151 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1152 desc->pg_error);
1153 }
1154 }
1155
1156 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1157 struct nfs_page *req)
1158 {
1159 u32 midx;
1160 unsigned int pgbase, offset, bytes;
1161 struct nfs_page *dupreq, *lastreq;
1162
1163 pgbase = req->wb_pgbase;
1164 offset = req->wb_offset;
1165 bytes = req->wb_bytes;
1166
1167 nfs_pageio_setup_mirroring(desc, req);
1168 if (desc->pg_error < 0)
1169 goto out_failed;
1170
1171
1172 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1173 nfs_page_group_lock(req);
1174
1175
1176 for (lastreq = req->wb_head;
1177 lastreq->wb_this_page != req->wb_head;
1178 lastreq = lastreq->wb_this_page)
1179 ;
1180
1181 dupreq = nfs_create_subreq(req, lastreq,
1182 pgbase, offset, bytes);
1183
1184 nfs_page_group_unlock(req);
1185 if (IS_ERR(dupreq)) {
1186 desc->pg_error = PTR_ERR(dupreq);
1187 goto out_failed;
1188 }
1189
1190 desc->pg_mirror_idx = midx;
1191 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1192 goto out_cleanup_subreq;
1193 }
1194
1195 desc->pg_mirror_idx = 0;
1196 if (!nfs_pageio_add_request_mirror(desc, req))
1197 goto out_failed;
1198
1199 return 1;
1200
1201 out_cleanup_subreq:
1202 nfs_pageio_cleanup_request(desc, dupreq);
1203 out_failed:
1204 nfs_pageio_error_cleanup(desc);
1205 return 0;
1206 }
1207
1208
1209
1210
1211
1212
1213
1214 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1215 u32 mirror_idx)
1216 {
1217 struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1218 u32 restore_idx = desc->pg_mirror_idx;
1219
1220 if (nfs_pgio_has_mirroring(desc))
1221 desc->pg_mirror_idx = mirror_idx;
1222 for (;;) {
1223 nfs_pageio_doio(desc);
1224 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1225 break;
1226 if (!nfs_do_recoalesce(desc))
1227 break;
1228 }
1229 desc->pg_mirror_idx = restore_idx;
1230 }
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1243 struct nfs_pgio_header *hdr)
1244 {
1245 LIST_HEAD(pages);
1246
1247 desc->pg_io_completion = hdr->io_completion;
1248 desc->pg_dreq = hdr->dreq;
1249 list_splice_init(&hdr->pages, &pages);
1250 while (!list_empty(&pages)) {
1251 struct nfs_page *req = nfs_list_entry(pages.next);
1252
1253 if (!nfs_pageio_add_request(desc, req))
1254 break;
1255 }
1256 nfs_pageio_complete(desc);
1257 if (!list_empty(&pages)) {
1258 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1259 hdr->completion_ops->error_cleanup(&pages, err);
1260 nfs_set_pgio_error(hdr, err, hdr->io_start);
1261 return err;
1262 }
1263 return 0;
1264 }
1265 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1266
1267
1268
1269
1270
1271 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1272 {
1273 u32 midx;
1274
1275 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1276 nfs_pageio_complete_mirror(desc, midx);
1277
1278 if (desc->pg_error < 0)
1279 nfs_pageio_error_cleanup(desc);
1280 if (desc->pg_ops->pg_cleanup)
1281 desc->pg_ops->pg_cleanup(desc);
1282 nfs_pageio_cleanup_mirroring(desc);
1283 }
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1297 {
1298 struct nfs_pgio_mirror *mirror;
1299 struct nfs_page *prev;
1300 u32 midx;
1301
1302 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1303 mirror = &desc->pg_mirrors[midx];
1304 if (!list_empty(&mirror->pg_list)) {
1305 prev = nfs_list_entry(mirror->pg_list.prev);
1306 if (index != prev->wb_index + 1) {
1307 nfs_pageio_complete(desc);
1308 break;
1309 }
1310 }
1311 }
1312 }
1313
1314
1315
1316
1317 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1318 {
1319 nfs_pageio_complete(pgio);
1320 }
1321
1322 int __init nfs_init_nfspagecache(void)
1323 {
1324 nfs_page_cachep = kmem_cache_create("nfs_page",
1325 sizeof(struct nfs_page),
1326 0, SLAB_HWCACHE_ALIGN,
1327 NULL);
1328 if (nfs_page_cachep == NULL)
1329 return -ENOMEM;
1330
1331 return 0;
1332 }
1333
1334 void nfs_destroy_nfspagecache(void)
1335 {
1336 kmem_cache_destroy(nfs_page_cachep);
1337 }
1338
1339 static const struct rpc_call_ops nfs_pgio_common_ops = {
1340 .rpc_call_prepare = nfs_pgio_prepare,
1341 .rpc_call_done = nfs_pgio_result,
1342 .rpc_release = nfs_pgio_release,
1343 };
1344
1345 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1346 .pg_test = nfs_generic_pg_test,
1347 .pg_doio = nfs_generic_pg_pgios,
1348 };