Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/pagelist.c
4 *
5 * A set of helper functions for managing NFS read and write requests.
6 * The main purpose of these routines is to provide support for the
7 * coalescing of several requests into a single RPC call.
8 *
9 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
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#include "nfstrace.h"
28
29#define NFSDBG_FACILITY NFSDBG_PAGECACHE
30
31static struct kmem_cache *nfs_page_cachep;
32static const struct rpc_call_ops nfs_pgio_common_ops;
33
34static struct nfs_pgio_mirror *
35nfs_pgio_get_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
36{
37 if (desc->pg_ops->pg_get_mirror)
38 return desc->pg_ops->pg_get_mirror(desc, idx);
39 return &desc->pg_mirrors[0];
40}
41
42struct nfs_pgio_mirror *
43nfs_pgio_current_mirror(struct nfs_pageio_descriptor *desc)
44{
45 return nfs_pgio_get_mirror(desc, desc->pg_mirror_idx);
46}
47EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
48
49static u32
50nfs_pgio_set_current_mirror(struct nfs_pageio_descriptor *desc, u32 idx)
51{
52 if (desc->pg_ops->pg_set_mirror)
53 return desc->pg_ops->pg_set_mirror(desc, idx);
54 return desc->pg_mirror_idx;
55}
56
57void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
58 struct nfs_pgio_header *hdr,
59 void (*release)(struct nfs_pgio_header *hdr))
60{
61 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
62
63
64 hdr->req = nfs_list_entry(mirror->pg_list.next);
65 hdr->inode = desc->pg_inode;
66 hdr->cred = nfs_req_openctx(hdr->req)->cred;
67 hdr->io_start = req_offset(hdr->req);
68 hdr->good_bytes = mirror->pg_count;
69 hdr->io_completion = desc->pg_io_completion;
70 hdr->dreq = desc->pg_dreq;
71 hdr->release = release;
72 hdr->completion_ops = desc->pg_completion_ops;
73 if (hdr->completion_ops->init_hdr)
74 hdr->completion_ops->init_hdr(hdr);
75
76 hdr->pgio_mirror_idx = desc->pg_mirror_idx;
77}
78EXPORT_SYMBOL_GPL(nfs_pgheader_init);
79
80void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
81{
82 unsigned int new = pos - hdr->io_start;
83
84 trace_nfs_pgio_error(hdr, error, pos);
85 if (hdr->good_bytes > new) {
86 hdr->good_bytes = new;
87 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
88 if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
89 hdr->error = error;
90 }
91}
92
93static inline struct nfs_page *nfs_page_alloc(void)
94{
95 struct nfs_page *p =
96 kmem_cache_zalloc(nfs_page_cachep, nfs_io_gfp_mask());
97 if (p)
98 INIT_LIST_HEAD(&p->wb_list);
99 return p;
100}
101
102static inline void
103nfs_page_free(struct nfs_page *p)
104{
105 kmem_cache_free(nfs_page_cachep, p);
106}
107
108/**
109 * nfs_iocounter_wait - wait for i/o to complete
110 * @l_ctx: nfs_lock_context with io_counter to use
111 *
112 * returns -ERESTARTSYS if interrupted by a fatal signal.
113 * Otherwise returns 0 once the io_count hits 0.
114 */
115int
116nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
117{
118 return wait_var_event_killable(&l_ctx->io_count,
119 !atomic_read(&l_ctx->io_count));
120}
121
122/**
123 * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
124 * to complete
125 * @task: the rpc_task that should wait
126 * @l_ctx: nfs_lock_context with io_counter to check
127 *
128 * Returns true if there is outstanding I/O to wait on and the
129 * task has been put to sleep.
130 */
131bool
132nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
133{
134 struct inode *inode = d_inode(l_ctx->open_context->dentry);
135 bool ret = false;
136
137 if (atomic_read(&l_ctx->io_count) > 0) {
138 rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
139 ret = true;
140 }
141
142 if (atomic_read(&l_ctx->io_count) == 0) {
143 rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
144 ret = false;
145 }
146
147 return ret;
148}
149EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
150
151/*
152 * nfs_page_lock_head_request - page lock the head of the page group
153 * @req: any member of the page group
154 */
155struct nfs_page *
156nfs_page_group_lock_head(struct nfs_page *req)
157{
158 struct nfs_page *head = req->wb_head;
159
160 while (!nfs_lock_request(head)) {
161 int ret = nfs_wait_on_request(head);
162 if (ret < 0)
163 return ERR_PTR(ret);
164 }
165 if (head != req)
166 kref_get(&head->wb_kref);
167 return head;
168}
169
170/*
171 * nfs_unroll_locks - unlock all newly locked reqs and wait on @req
172 * @head: head request of page group, must be holding head lock
173 * @req: request that couldn't lock and needs to wait on the req bit lock
174 *
175 * This is a helper function for nfs_lock_and_join_requests
176 * returns 0 on success, < 0 on error.
177 */
178static void
179nfs_unroll_locks(struct nfs_page *head, struct nfs_page *req)
180{
181 struct nfs_page *tmp;
182
183 /* relinquish all the locks successfully grabbed this run */
184 for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) {
185 if (!kref_read(&tmp->wb_kref))
186 continue;
187 nfs_unlock_and_release_request(tmp);
188 }
189}
190
191/*
192 * nfs_page_group_lock_subreq - try to lock a subrequest
193 * @head: head request of page group
194 * @subreq: request to lock
195 *
196 * This is a helper function for nfs_lock_and_join_requests which
197 * must be called with the head request and page group both locked.
198 * On error, it returns with the page group unlocked.
199 */
200static int
201nfs_page_group_lock_subreq(struct nfs_page *head, struct nfs_page *subreq)
202{
203 int ret;
204
205 if (!kref_get_unless_zero(&subreq->wb_kref))
206 return 0;
207 while (!nfs_lock_request(subreq)) {
208 nfs_page_group_unlock(head);
209 ret = nfs_wait_on_request(subreq);
210 if (!ret)
211 ret = nfs_page_group_lock(head);
212 if (ret < 0) {
213 nfs_unroll_locks(head, subreq);
214 nfs_release_request(subreq);
215 return ret;
216 }
217 }
218 return 0;
219}
220
221/*
222 * nfs_page_group_lock_subrequests - try to lock the subrequests
223 * @head: head request of page group
224 *
225 * This is a helper function for nfs_lock_and_join_requests which
226 * must be called with the head request locked.
227 */
228int nfs_page_group_lock_subrequests(struct nfs_page *head)
229{
230 struct nfs_page *subreq;
231 int ret;
232
233 ret = nfs_page_group_lock(head);
234 if (ret < 0)
235 return ret;
236 /* lock each request in the page group */
237 for (subreq = head->wb_this_page; subreq != head;
238 subreq = subreq->wb_this_page) {
239 ret = nfs_page_group_lock_subreq(head, subreq);
240 if (ret < 0)
241 return ret;
242 }
243 nfs_page_group_unlock(head);
244 return 0;
245}
246
247/*
248 * nfs_page_set_headlock - set the request PG_HEADLOCK
249 * @req: request that is to be locked
250 *
251 * this lock must be held when modifying req->wb_head
252 *
253 * return 0 on success, < 0 on error
254 */
255int
256nfs_page_set_headlock(struct nfs_page *req)
257{
258 if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
259 return 0;
260
261 set_bit(PG_CONTENDED1, &req->wb_flags);
262 smp_mb__after_atomic();
263 return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
264 TASK_UNINTERRUPTIBLE);
265}
266
267/*
268 * nfs_page_clear_headlock - clear the request PG_HEADLOCK
269 * @req: request that is to be locked
270 */
271void
272nfs_page_clear_headlock(struct nfs_page *req)
273{
274 clear_bit_unlock(PG_HEADLOCK, &req->wb_flags);
275 smp_mb__after_atomic();
276 if (!test_bit(PG_CONTENDED1, &req->wb_flags))
277 return;
278 wake_up_bit(&req->wb_flags, PG_HEADLOCK);
279}
280
281/*
282 * nfs_page_group_lock - lock the head of the page group
283 * @req: request in group that is to be locked
284 *
285 * this lock must be held when traversing or modifying the page
286 * group list
287 *
288 * return 0 on success, < 0 on error
289 */
290int
291nfs_page_group_lock(struct nfs_page *req)
292{
293 int ret;
294
295 ret = nfs_page_set_headlock(req);
296 if (ret || req->wb_head == req)
297 return ret;
298 return nfs_page_set_headlock(req->wb_head);
299}
300
301/*
302 * nfs_page_group_unlock - unlock the head of the page group
303 * @req: request in group that is to be unlocked
304 */
305void
306nfs_page_group_unlock(struct nfs_page *req)
307{
308 if (req != req->wb_head)
309 nfs_page_clear_headlock(req->wb_head);
310 nfs_page_clear_headlock(req);
311}
312
313/*
314 * nfs_page_group_sync_on_bit_locked
315 *
316 * must be called with page group lock held
317 */
318static bool
319nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
320{
321 struct nfs_page *head = req->wb_head;
322 struct nfs_page *tmp;
323
324 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
325 WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
326
327 tmp = req->wb_this_page;
328 while (tmp != req) {
329 if (!test_bit(bit, &tmp->wb_flags))
330 return false;
331 tmp = tmp->wb_this_page;
332 }
333
334 /* true! reset all bits */
335 tmp = req;
336 do {
337 clear_bit(bit, &tmp->wb_flags);
338 tmp = tmp->wb_this_page;
339 } while (tmp != req);
340
341 return true;
342}
343
344/*
345 * nfs_page_group_sync_on_bit - set bit on current request, but only
346 * return true if the bit is set for all requests in page group
347 * @req - request in page group
348 * @bit - PG_* bit that is used to sync page group
349 */
350bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
351{
352 bool ret;
353
354 nfs_page_group_lock(req);
355 ret = nfs_page_group_sync_on_bit_locked(req, bit);
356 nfs_page_group_unlock(req);
357
358 return ret;
359}
360
361/*
362 * nfs_page_group_init - Initialize the page group linkage for @req
363 * @req - a new nfs request
364 * @prev - the previous request in page group, or NULL if @req is the first
365 * or only request in the group (the head).
366 */
367static inline void
368nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
369{
370 struct inode *inode;
371 WARN_ON_ONCE(prev == req);
372
373 if (!prev) {
374 /* a head request */
375 req->wb_head = req;
376 req->wb_this_page = req;
377 } else {
378 /* a subrequest */
379 WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
380 WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
381 req->wb_head = prev->wb_head;
382 req->wb_this_page = prev->wb_this_page;
383 prev->wb_this_page = req;
384
385 /* All subrequests take a ref on the head request until
386 * nfs_page_group_destroy is called */
387 kref_get(&req->wb_head->wb_kref);
388
389 /* grab extra ref and bump the request count if head request
390 * has extra ref from the write/commit path to handle handoff
391 * between write and commit lists. */
392 if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
393 inode = page_file_mapping(req->wb_page)->host;
394 set_bit(PG_INODE_REF, &req->wb_flags);
395 kref_get(&req->wb_kref);
396 atomic_long_inc(&NFS_I(inode)->nrequests);
397 }
398 }
399}
400
401/*
402 * nfs_page_group_destroy - sync the destruction of page groups
403 * @req - request that no longer needs the page group
404 *
405 * releases the page group reference from each member once all
406 * members have called this function.
407 */
408static void
409nfs_page_group_destroy(struct kref *kref)
410{
411 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
412 struct nfs_page *head = req->wb_head;
413 struct nfs_page *tmp, *next;
414
415 if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
416 goto out;
417
418 tmp = req;
419 do {
420 next = tmp->wb_this_page;
421 /* unlink and free */
422 tmp->wb_this_page = tmp;
423 tmp->wb_head = tmp;
424 nfs_free_request(tmp);
425 tmp = next;
426 } while (tmp != req);
427out:
428 /* subrequests must release the ref on the head request */
429 if (head != req)
430 nfs_release_request(head);
431}
432
433static struct nfs_page *
434__nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
435 unsigned int pgbase, unsigned int offset,
436 unsigned int count)
437{
438 struct nfs_page *req;
439 struct nfs_open_context *ctx = l_ctx->open_context;
440
441 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
442 return ERR_PTR(-EBADF);
443 /* try to allocate the request struct */
444 req = nfs_page_alloc();
445 if (req == NULL)
446 return ERR_PTR(-ENOMEM);
447
448 req->wb_lock_context = l_ctx;
449 refcount_inc(&l_ctx->count);
450 atomic_inc(&l_ctx->io_count);
451
452 /* Initialize the request struct. Initially, we assume a
453 * long write-back delay. This will be adjusted in
454 * update_nfs_request below if the region is not locked. */
455 req->wb_page = page;
456 if (page) {
457 req->wb_index = page_index(page);
458 get_page(page);
459 }
460 req->wb_offset = offset;
461 req->wb_pgbase = pgbase;
462 req->wb_bytes = count;
463 kref_init(&req->wb_kref);
464 req->wb_nio = 0;
465 return req;
466}
467
468/**
469 * nfs_create_request - Create an NFS read/write request.
470 * @ctx: open context to use
471 * @page: page to write
472 * @offset: starting offset within the page for the write
473 * @count: number of bytes to read/write
474 *
475 * The page must be locked by the caller. This makes sure we never
476 * create two different requests for the same page.
477 * User should ensure it is safe to sleep in this function.
478 */
479struct nfs_page *
480nfs_create_request(struct nfs_open_context *ctx, struct page *page,
481 unsigned int offset, unsigned int count)
482{
483 struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
484 struct nfs_page *ret;
485
486 if (IS_ERR(l_ctx))
487 return ERR_CAST(l_ctx);
488 ret = __nfs_create_request(l_ctx, page, offset, offset, count);
489 if (!IS_ERR(ret))
490 nfs_page_group_init(ret, NULL);
491 nfs_put_lock_context(l_ctx);
492 return ret;
493}
494
495static struct nfs_page *
496nfs_create_subreq(struct nfs_page *req,
497 unsigned int pgbase,
498 unsigned int offset,
499 unsigned int count)
500{
501 struct nfs_page *last;
502 struct nfs_page *ret;
503
504 ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
505 pgbase, offset, count);
506 if (!IS_ERR(ret)) {
507 /* find the last request */
508 for (last = req->wb_head;
509 last->wb_this_page != req->wb_head;
510 last = last->wb_this_page)
511 ;
512
513 nfs_lock_request(ret);
514 ret->wb_index = req->wb_index;
515 nfs_page_group_init(ret, last);
516 ret->wb_nio = req->wb_nio;
517 }
518 return ret;
519}
520
521/**
522 * nfs_unlock_request - Unlock request and wake up sleepers.
523 * @req: pointer to request
524 */
525void nfs_unlock_request(struct nfs_page *req)
526{
527 clear_bit_unlock(PG_BUSY, &req->wb_flags);
528 smp_mb__after_atomic();
529 if (!test_bit(PG_CONTENDED2, &req->wb_flags))
530 return;
531 wake_up_bit(&req->wb_flags, PG_BUSY);
532}
533
534/**
535 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
536 * @req: pointer to request
537 */
538void nfs_unlock_and_release_request(struct nfs_page *req)
539{
540 nfs_unlock_request(req);
541 nfs_release_request(req);
542}
543
544/*
545 * nfs_clear_request - Free up all resources allocated to the request
546 * @req:
547 *
548 * Release page and open context resources associated with a read/write
549 * request after it has completed.
550 */
551static void nfs_clear_request(struct nfs_page *req)
552{
553 struct page *page = req->wb_page;
554 struct nfs_lock_context *l_ctx = req->wb_lock_context;
555 struct nfs_open_context *ctx;
556
557 if (page != NULL) {
558 put_page(page);
559 req->wb_page = NULL;
560 }
561 if (l_ctx != NULL) {
562 if (atomic_dec_and_test(&l_ctx->io_count)) {
563 wake_up_var(&l_ctx->io_count);
564 ctx = l_ctx->open_context;
565 if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
566 rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
567 }
568 nfs_put_lock_context(l_ctx);
569 req->wb_lock_context = NULL;
570 }
571}
572
573/**
574 * nfs_free_request - Release the count on an NFS read/write request
575 * @req: request to release
576 *
577 * Note: Should never be called with the spinlock held!
578 */
579void nfs_free_request(struct nfs_page *req)
580{
581 WARN_ON_ONCE(req->wb_this_page != req);
582
583 /* extra debug: make sure no sync bits are still set */
584 WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
585 WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
586 WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
587 WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
588 WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
589
590 /* Release struct file and open context */
591 nfs_clear_request(req);
592 nfs_page_free(req);
593}
594
595void nfs_release_request(struct nfs_page *req)
596{
597 kref_put(&req->wb_kref, nfs_page_group_destroy);
598}
599EXPORT_SYMBOL_GPL(nfs_release_request);
600
601/**
602 * nfs_wait_on_request - Wait for a request to complete.
603 * @req: request to wait upon.
604 *
605 * Interruptible by fatal signals only.
606 * The user is responsible for holding a count on the request.
607 */
608int
609nfs_wait_on_request(struct nfs_page *req)
610{
611 if (!test_bit(PG_BUSY, &req->wb_flags))
612 return 0;
613 set_bit(PG_CONTENDED2, &req->wb_flags);
614 smp_mb__after_atomic();
615 return wait_on_bit_io(&req->wb_flags, PG_BUSY,
616 TASK_UNINTERRUPTIBLE);
617}
618EXPORT_SYMBOL_GPL(nfs_wait_on_request);
619
620/*
621 * nfs_generic_pg_test - determine if requests can be coalesced
622 * @desc: pointer to descriptor
623 * @prev: previous request in desc, or NULL
624 * @req: this request
625 *
626 * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
627 * the size of the request.
628 */
629size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
630 struct nfs_page *prev, struct nfs_page *req)
631{
632 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
633
634
635 if (mirror->pg_count > mirror->pg_bsize) {
636 /* should never happen */
637 WARN_ON_ONCE(1);
638 return 0;
639 }
640
641 /*
642 * Limit the request size so that we can still allocate a page array
643 * for it without upsetting the slab allocator.
644 */
645 if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
646 sizeof(struct page *) > PAGE_SIZE)
647 return 0;
648
649 return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
650}
651EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
652
653struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
654{
655 struct nfs_pgio_header *hdr = ops->rw_alloc_header();
656
657 if (hdr) {
658 INIT_LIST_HEAD(&hdr->pages);
659 hdr->rw_ops = ops;
660 }
661 return hdr;
662}
663EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
664
665/**
666 * nfs_pgio_data_destroy - make @hdr suitable for reuse
667 *
668 * Frees memory and releases refs from nfs_generic_pgio, so that it may
669 * be called again.
670 *
671 * @hdr: A header that has had nfs_generic_pgio called
672 */
673static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
674{
675 if (hdr->args.context)
676 put_nfs_open_context(hdr->args.context);
677 if (hdr->page_array.pagevec != hdr->page_array.page_array)
678 kfree(hdr->page_array.pagevec);
679}
680
681/*
682 * nfs_pgio_header_free - Free a read or write header
683 * @hdr: The header to free
684 */
685void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
686{
687 nfs_pgio_data_destroy(hdr);
688 hdr->rw_ops->rw_free_header(hdr);
689}
690EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
691
692/**
693 * nfs_pgio_rpcsetup - Set up arguments for a pageio call
694 * @hdr: The pageio hdr
695 * @count: Number of bytes to read
696 * @how: How to commit data (writes only)
697 * @cinfo: Commit information for the call (writes only)
698 */
699static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
700 unsigned int count,
701 int how, struct nfs_commit_info *cinfo)
702{
703 struct nfs_page *req = hdr->req;
704
705 /* Set up the RPC argument and reply structs
706 * NB: take care not to mess about with hdr->commit et al. */
707
708 hdr->args.fh = NFS_FH(hdr->inode);
709 hdr->args.offset = req_offset(req);
710 /* pnfs_set_layoutcommit needs this */
711 hdr->mds_offset = hdr->args.offset;
712 hdr->args.pgbase = req->wb_pgbase;
713 hdr->args.pages = hdr->page_array.pagevec;
714 hdr->args.count = count;
715 hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
716 hdr->args.lock_context = req->wb_lock_context;
717 hdr->args.stable = NFS_UNSTABLE;
718 switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
719 case 0:
720 break;
721 case FLUSH_COND_STABLE:
722 if (nfs_reqs_to_commit(cinfo))
723 break;
724 fallthrough;
725 default:
726 hdr->args.stable = NFS_FILE_SYNC;
727 }
728
729 hdr->res.fattr = &hdr->fattr;
730 hdr->res.count = 0;
731 hdr->res.eof = 0;
732 hdr->res.verf = &hdr->verf;
733 nfs_fattr_init(&hdr->fattr);
734}
735
736/**
737 * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
738 * @task: The current task
739 * @calldata: pageio header to prepare
740 */
741static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
742{
743 struct nfs_pgio_header *hdr = calldata;
744 int err;
745 err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
746 if (err)
747 rpc_exit(task, err);
748}
749
750int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
751 const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
752 const struct rpc_call_ops *call_ops, int how, int flags)
753{
754 struct rpc_task *task;
755 struct rpc_message msg = {
756 .rpc_argp = &hdr->args,
757 .rpc_resp = &hdr->res,
758 .rpc_cred = cred,
759 };
760 struct rpc_task_setup task_setup_data = {
761 .rpc_client = clnt,
762 .task = &hdr->task,
763 .rpc_message = &msg,
764 .callback_ops = call_ops,
765 .callback_data = hdr,
766 .workqueue = nfsiod_workqueue,
767 .flags = RPC_TASK_ASYNC | flags,
768 };
769
770 if (nfs_server_capable(hdr->inode, NFS_CAP_MOVEABLE))
771 task_setup_data.flags |= RPC_TASK_MOVEABLE;
772
773 hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
774
775 dprintk("NFS: initiated pgio call "
776 "(req %s/%llu, %u bytes @ offset %llu)\n",
777 hdr->inode->i_sb->s_id,
778 (unsigned long long)NFS_FILEID(hdr->inode),
779 hdr->args.count,
780 (unsigned long long)hdr->args.offset);
781
782 task = rpc_run_task(&task_setup_data);
783 if (IS_ERR(task))
784 return PTR_ERR(task);
785 rpc_put_task(task);
786 return 0;
787}
788EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
789
790/**
791 * nfs_pgio_error - Clean up from a pageio error
792 * @hdr: pageio header
793 */
794static void nfs_pgio_error(struct nfs_pgio_header *hdr)
795{
796 set_bit(NFS_IOHDR_REDO, &hdr->flags);
797 hdr->completion_ops->completion(hdr);
798}
799
800/**
801 * nfs_pgio_release - Release pageio data
802 * @calldata: The pageio header to release
803 */
804static void nfs_pgio_release(void *calldata)
805{
806 struct nfs_pgio_header *hdr = calldata;
807 hdr->completion_ops->completion(hdr);
808}
809
810static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
811 unsigned int bsize)
812{
813 INIT_LIST_HEAD(&mirror->pg_list);
814 mirror->pg_bytes_written = 0;
815 mirror->pg_count = 0;
816 mirror->pg_bsize = bsize;
817 mirror->pg_base = 0;
818 mirror->pg_recoalesce = 0;
819}
820
821/**
822 * nfs_pageio_init - initialise a page io descriptor
823 * @desc: pointer to descriptor
824 * @inode: pointer to inode
825 * @pg_ops: pointer to pageio operations
826 * @compl_ops: pointer to pageio completion operations
827 * @rw_ops: pointer to nfs read/write operations
828 * @bsize: io block size
829 * @io_flags: extra parameters for the io function
830 */
831void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
832 struct inode *inode,
833 const struct nfs_pageio_ops *pg_ops,
834 const struct nfs_pgio_completion_ops *compl_ops,
835 const struct nfs_rw_ops *rw_ops,
836 size_t bsize,
837 int io_flags)
838{
839 desc->pg_moreio = 0;
840 desc->pg_inode = inode;
841 desc->pg_ops = pg_ops;
842 desc->pg_completion_ops = compl_ops;
843 desc->pg_rw_ops = rw_ops;
844 desc->pg_ioflags = io_flags;
845 desc->pg_error = 0;
846 desc->pg_lseg = NULL;
847 desc->pg_io_completion = NULL;
848 desc->pg_dreq = NULL;
849 desc->pg_bsize = bsize;
850
851 desc->pg_mirror_count = 1;
852 desc->pg_mirror_idx = 0;
853
854 desc->pg_mirrors_dynamic = NULL;
855 desc->pg_mirrors = desc->pg_mirrors_static;
856 nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
857 desc->pg_maxretrans = 0;
858}
859
860/**
861 * nfs_pgio_result - Basic pageio error handling
862 * @task: The task that ran
863 * @calldata: Pageio header to check
864 */
865static void nfs_pgio_result(struct rpc_task *task, void *calldata)
866{
867 struct nfs_pgio_header *hdr = calldata;
868 struct inode *inode = hdr->inode;
869
870 if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
871 return;
872 if (task->tk_status < 0)
873 nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
874 else
875 hdr->rw_ops->rw_result(task, hdr);
876}
877
878/*
879 * Create an RPC task for the given read or write request and kick it.
880 * The page must have been locked by the caller.
881 *
882 * It may happen that the page we're passed is not marked dirty.
883 * This is the case if nfs_updatepage detects a conflicting request
884 * that has been written but not committed.
885 */
886int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
887 struct nfs_pgio_header *hdr)
888{
889 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
890
891 struct nfs_page *req;
892 struct page **pages,
893 *last_page;
894 struct list_head *head = &mirror->pg_list;
895 struct nfs_commit_info cinfo;
896 struct nfs_page_array *pg_array = &hdr->page_array;
897 unsigned int pagecount, pageused;
898 gfp_t gfp_flags = nfs_io_gfp_mask();
899
900 pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
901 pg_array->npages = pagecount;
902
903 if (pagecount <= ARRAY_SIZE(pg_array->page_array))
904 pg_array->pagevec = pg_array->page_array;
905 else {
906 pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
907 if (!pg_array->pagevec) {
908 pg_array->npages = 0;
909 nfs_pgio_error(hdr);
910 desc->pg_error = -ENOMEM;
911 return desc->pg_error;
912 }
913 }
914
915 nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
916 pages = hdr->page_array.pagevec;
917 last_page = NULL;
918 pageused = 0;
919 while (!list_empty(head)) {
920 req = nfs_list_entry(head->next);
921 nfs_list_move_request(req, &hdr->pages);
922
923 if (!last_page || last_page != req->wb_page) {
924 pageused++;
925 if (pageused > pagecount)
926 break;
927 *pages++ = last_page = req->wb_page;
928 }
929 }
930 if (WARN_ON_ONCE(pageused != pagecount)) {
931 nfs_pgio_error(hdr);
932 desc->pg_error = -EINVAL;
933 return desc->pg_error;
934 }
935
936 if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
937 (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
938 desc->pg_ioflags &= ~FLUSH_COND_STABLE;
939
940 /* Set up the argument struct */
941 nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
942 desc->pg_rpc_callops = &nfs_pgio_common_ops;
943 return 0;
944}
945EXPORT_SYMBOL_GPL(nfs_generic_pgio);
946
947static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
948{
949 struct nfs_pgio_header *hdr;
950 int ret;
951 unsigned short task_flags = 0;
952
953 hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
954 if (!hdr) {
955 desc->pg_error = -ENOMEM;
956 return desc->pg_error;
957 }
958 nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
959 ret = nfs_generic_pgio(desc, hdr);
960 if (ret == 0) {
961 if (NFS_SERVER(hdr->inode)->nfs_client->cl_minorversion)
962 task_flags = RPC_TASK_MOVEABLE;
963 ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
964 hdr,
965 hdr->cred,
966 NFS_PROTO(hdr->inode),
967 desc->pg_rpc_callops,
968 desc->pg_ioflags,
969 RPC_TASK_CRED_NOREF | task_flags);
970 }
971 return ret;
972}
973
974static struct nfs_pgio_mirror *
975nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
976 unsigned int mirror_count)
977{
978 struct nfs_pgio_mirror *ret;
979 unsigned int i;
980
981 kfree(desc->pg_mirrors_dynamic);
982 desc->pg_mirrors_dynamic = NULL;
983 if (mirror_count == 1)
984 return desc->pg_mirrors_static;
985 ret = kmalloc_array(mirror_count, sizeof(*ret), nfs_io_gfp_mask());
986 if (ret != NULL) {
987 for (i = 0; i < mirror_count; i++)
988 nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
989 desc->pg_mirrors_dynamic = ret;
990 }
991 return ret;
992}
993
994/*
995 * nfs_pageio_setup_mirroring - determine if mirroring is to be used
996 * by calling the pg_get_mirror_count op
997 */
998static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
999 struct nfs_page *req)
1000{
1001 unsigned int mirror_count = 1;
1002
1003 if (pgio->pg_ops->pg_get_mirror_count)
1004 mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
1005 if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
1006 return;
1007
1008 if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
1009 pgio->pg_error = -EINVAL;
1010 return;
1011 }
1012
1013 pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
1014 if (pgio->pg_mirrors == NULL) {
1015 pgio->pg_error = -ENOMEM;
1016 pgio->pg_mirrors = pgio->pg_mirrors_static;
1017 mirror_count = 1;
1018 }
1019 pgio->pg_mirror_count = mirror_count;
1020}
1021
1022static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1023{
1024 pgio->pg_mirror_count = 1;
1025 pgio->pg_mirror_idx = 0;
1026 pgio->pg_mirrors = pgio->pg_mirrors_static;
1027 kfree(pgio->pg_mirrors_dynamic);
1028 pgio->pg_mirrors_dynamic = NULL;
1029}
1030
1031static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1032 const struct nfs_lock_context *l2)
1033{
1034 return l1->lockowner == l2->lockowner;
1035}
1036
1037/**
1038 * nfs_coalesce_size - test two requests for compatibility
1039 * @prev: pointer to nfs_page
1040 * @req: pointer to nfs_page
1041 * @pgio: pointer to nfs_pagio_descriptor
1042 *
1043 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1044 * page data area they describe is contiguous, and that their RPC
1045 * credentials, NFSv4 open state, and lockowners are the same.
1046 *
1047 * Returns size of the request that can be coalesced
1048 */
1049static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1050 struct nfs_page *req,
1051 struct nfs_pageio_descriptor *pgio)
1052{
1053 struct file_lock_context *flctx;
1054
1055 if (prev) {
1056 if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1057 return 0;
1058 flctx = locks_inode_context(d_inode(nfs_req_openctx(req)->dentry));
1059 if (flctx != NULL &&
1060 !(list_empty_careful(&flctx->flc_posix) &&
1061 list_empty_careful(&flctx->flc_flock)) &&
1062 !nfs_match_lock_context(req->wb_lock_context,
1063 prev->wb_lock_context))
1064 return 0;
1065 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1066 return 0;
1067 if (req->wb_page == prev->wb_page) {
1068 if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
1069 return 0;
1070 } else {
1071 if (req->wb_pgbase != 0 ||
1072 prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
1073 return 0;
1074 }
1075 }
1076 return pgio->pg_ops->pg_test(pgio, prev, req);
1077}
1078
1079/**
1080 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1081 * @desc: destination io descriptor
1082 * @req: request
1083 *
1084 * If the request 'req' was successfully coalesced into the existing list
1085 * of pages 'desc', it returns the size of req.
1086 */
1087static unsigned int
1088nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1089 struct nfs_page *req)
1090{
1091 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1092 struct nfs_page *prev = NULL;
1093 unsigned int size;
1094
1095 if (list_empty(&mirror->pg_list)) {
1096 if (desc->pg_ops->pg_init)
1097 desc->pg_ops->pg_init(desc, req);
1098 if (desc->pg_error < 0)
1099 return 0;
1100 mirror->pg_base = req->wb_pgbase;
1101 mirror->pg_count = 0;
1102 mirror->pg_recoalesce = 0;
1103 } else
1104 prev = nfs_list_entry(mirror->pg_list.prev);
1105
1106 if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1107 if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1108 desc->pg_error = -ETIMEDOUT;
1109 else
1110 desc->pg_error = -EIO;
1111 return 0;
1112 }
1113
1114 size = nfs_coalesce_size(prev, req, desc);
1115 if (size < req->wb_bytes)
1116 return size;
1117 nfs_list_move_request(req, &mirror->pg_list);
1118 mirror->pg_count += req->wb_bytes;
1119 return req->wb_bytes;
1120}
1121
1122/*
1123 * Helper for nfs_pageio_add_request and nfs_pageio_complete
1124 */
1125static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1126{
1127 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1128
1129 if (!list_empty(&mirror->pg_list)) {
1130 int error = desc->pg_ops->pg_doio(desc);
1131 if (error < 0)
1132 desc->pg_error = error;
1133 if (list_empty(&mirror->pg_list))
1134 mirror->pg_bytes_written += mirror->pg_count;
1135 }
1136}
1137
1138static void
1139nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1140 struct nfs_page *req)
1141{
1142 LIST_HEAD(head);
1143
1144 nfs_list_move_request(req, &head);
1145 desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1146}
1147
1148/**
1149 * __nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1150 * @desc: destination io descriptor
1151 * @req: request
1152 *
1153 * This may split a request into subrequests which are all part of the
1154 * same page group. If so, it will submit @req as the last one, to ensure
1155 * the pointer to @req is still valid in case of failure.
1156 *
1157 * Returns true if the request 'req' was successfully coalesced into the
1158 * existing list of pages 'desc'.
1159 */
1160static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1161 struct nfs_page *req)
1162{
1163 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1164 struct nfs_page *subreq;
1165 unsigned int size, subreq_size;
1166
1167 nfs_page_group_lock(req);
1168
1169 subreq = req;
1170 subreq_size = subreq->wb_bytes;
1171 for(;;) {
1172 size = nfs_pageio_do_add_request(desc, subreq);
1173 if (size == subreq_size) {
1174 /* We successfully submitted a request */
1175 if (subreq == req)
1176 break;
1177 req->wb_pgbase += size;
1178 req->wb_bytes -= size;
1179 req->wb_offset += size;
1180 subreq_size = req->wb_bytes;
1181 subreq = req;
1182 continue;
1183 }
1184 if (WARN_ON_ONCE(subreq != req)) {
1185 nfs_page_group_unlock(req);
1186 nfs_pageio_cleanup_request(desc, subreq);
1187 subreq = req;
1188 subreq_size = req->wb_bytes;
1189 nfs_page_group_lock(req);
1190 }
1191 if (!size) {
1192 /* Can't coalesce any more, so do I/O */
1193 nfs_page_group_unlock(req);
1194 desc->pg_moreio = 1;
1195 nfs_pageio_doio(desc);
1196 if (desc->pg_error < 0 || mirror->pg_recoalesce)
1197 return 0;
1198 /* retry add_request for this subreq */
1199 nfs_page_group_lock(req);
1200 continue;
1201 }
1202 subreq = nfs_create_subreq(req, req->wb_pgbase,
1203 req->wb_offset, size);
1204 if (IS_ERR(subreq))
1205 goto err_ptr;
1206 subreq_size = size;
1207 }
1208
1209 nfs_page_group_unlock(req);
1210 return 1;
1211err_ptr:
1212 desc->pg_error = PTR_ERR(subreq);
1213 nfs_page_group_unlock(req);
1214 return 0;
1215}
1216
1217static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1218{
1219 struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1220 LIST_HEAD(head);
1221
1222 do {
1223 list_splice_init(&mirror->pg_list, &head);
1224 mirror->pg_recoalesce = 0;
1225
1226 while (!list_empty(&head)) {
1227 struct nfs_page *req;
1228
1229 req = list_first_entry(&head, struct nfs_page, wb_list);
1230 if (__nfs_pageio_add_request(desc, req))
1231 continue;
1232 if (desc->pg_error < 0) {
1233 list_splice_tail(&head, &mirror->pg_list);
1234 mirror->pg_recoalesce = 1;
1235 return 0;
1236 }
1237 break;
1238 }
1239 } while (mirror->pg_recoalesce);
1240 return 1;
1241}
1242
1243static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1244 struct nfs_page *req)
1245{
1246 int ret;
1247
1248 do {
1249 ret = __nfs_pageio_add_request(desc, req);
1250 if (ret)
1251 break;
1252 if (desc->pg_error < 0)
1253 break;
1254 ret = nfs_do_recoalesce(desc);
1255 } while (ret);
1256
1257 return ret;
1258}
1259
1260static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1261{
1262 u32 midx;
1263 struct nfs_pgio_mirror *mirror;
1264
1265 if (!desc->pg_error)
1266 return;
1267
1268 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1269 mirror = nfs_pgio_get_mirror(desc, midx);
1270 desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1271 desc->pg_error);
1272 }
1273}
1274
1275int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1276 struct nfs_page *req)
1277{
1278 u32 midx;
1279 unsigned int pgbase, offset, bytes;
1280 struct nfs_page *dupreq;
1281
1282 pgbase = req->wb_pgbase;
1283 offset = req->wb_offset;
1284 bytes = req->wb_bytes;
1285
1286 nfs_pageio_setup_mirroring(desc, req);
1287 if (desc->pg_error < 0)
1288 goto out_failed;
1289
1290 /* Create the mirror instances first, and fire them off */
1291 for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1292 nfs_page_group_lock(req);
1293
1294 dupreq = nfs_create_subreq(req,
1295 pgbase, offset, bytes);
1296
1297 nfs_page_group_unlock(req);
1298 if (IS_ERR(dupreq)) {
1299 desc->pg_error = PTR_ERR(dupreq);
1300 goto out_failed;
1301 }
1302
1303 nfs_pgio_set_current_mirror(desc, midx);
1304 if (!nfs_pageio_add_request_mirror(desc, dupreq))
1305 goto out_cleanup_subreq;
1306 }
1307
1308 nfs_pgio_set_current_mirror(desc, 0);
1309 if (!nfs_pageio_add_request_mirror(desc, req))
1310 goto out_failed;
1311
1312 return 1;
1313
1314out_cleanup_subreq:
1315 nfs_pageio_cleanup_request(desc, dupreq);
1316out_failed:
1317 nfs_pageio_error_cleanup(desc);
1318 return 0;
1319}
1320
1321/*
1322 * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1323 * nfs_pageio_descriptor
1324 * @desc: pointer to io descriptor
1325 * @mirror_idx: pointer to mirror index
1326 */
1327static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1328 u32 mirror_idx)
1329{
1330 struct nfs_pgio_mirror *mirror;
1331 u32 restore_idx;
1332
1333 restore_idx = nfs_pgio_set_current_mirror(desc, mirror_idx);
1334 mirror = nfs_pgio_current_mirror(desc);
1335
1336 for (;;) {
1337 nfs_pageio_doio(desc);
1338 if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1339 break;
1340 if (!nfs_do_recoalesce(desc))
1341 break;
1342 }
1343 nfs_pgio_set_current_mirror(desc, restore_idx);
1344}
1345
1346/*
1347 * nfs_pageio_resend - Transfer requests to new descriptor and resend
1348 * @hdr - the pgio header to move request from
1349 * @desc - the pageio descriptor to add requests to
1350 *
1351 * Try to move each request (nfs_page) from @hdr to @desc then attempt
1352 * to send them.
1353 *
1354 * Returns 0 on success and < 0 on error.
1355 */
1356int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1357 struct nfs_pgio_header *hdr)
1358{
1359 LIST_HEAD(pages);
1360
1361 desc->pg_io_completion = hdr->io_completion;
1362 desc->pg_dreq = hdr->dreq;
1363 list_splice_init(&hdr->pages, &pages);
1364 while (!list_empty(&pages)) {
1365 struct nfs_page *req = nfs_list_entry(pages.next);
1366
1367 if (!nfs_pageio_add_request(desc, req))
1368 break;
1369 }
1370 nfs_pageio_complete(desc);
1371 if (!list_empty(&pages)) {
1372 int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1373 hdr->completion_ops->error_cleanup(&pages, err);
1374 nfs_set_pgio_error(hdr, err, hdr->io_start);
1375 return err;
1376 }
1377 return 0;
1378}
1379EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1380
1381/**
1382 * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1383 * @desc: pointer to io descriptor
1384 */
1385void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1386{
1387 u32 midx;
1388
1389 for (midx = 0; midx < desc->pg_mirror_count; midx++)
1390 nfs_pageio_complete_mirror(desc, midx);
1391
1392 if (desc->pg_error < 0)
1393 nfs_pageio_error_cleanup(desc);
1394 if (desc->pg_ops->pg_cleanup)
1395 desc->pg_ops->pg_cleanup(desc);
1396 nfs_pageio_cleanup_mirroring(desc);
1397}
1398
1399/**
1400 * nfs_pageio_cond_complete - Conditional I/O completion
1401 * @desc: pointer to io descriptor
1402 * @index: page index
1403 *
1404 * It is important to ensure that processes don't try to take locks
1405 * on non-contiguous ranges of pages as that might deadlock. This
1406 * function should be called before attempting to wait on a locked
1407 * nfs_page. It will complete the I/O if the page index 'index'
1408 * is not contiguous with the existing list of pages in 'desc'.
1409 */
1410void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1411{
1412 struct nfs_pgio_mirror *mirror;
1413 struct nfs_page *prev;
1414 u32 midx;
1415
1416 for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1417 mirror = nfs_pgio_get_mirror(desc, midx);
1418 if (!list_empty(&mirror->pg_list)) {
1419 prev = nfs_list_entry(mirror->pg_list.prev);
1420 if (index != prev->wb_index + 1) {
1421 nfs_pageio_complete(desc);
1422 break;
1423 }
1424 }
1425 }
1426}
1427
1428/*
1429 * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1430 */
1431void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1432{
1433 nfs_pageio_complete(pgio);
1434}
1435
1436int __init nfs_init_nfspagecache(void)
1437{
1438 nfs_page_cachep = kmem_cache_create("nfs_page",
1439 sizeof(struct nfs_page),
1440 0, SLAB_HWCACHE_ALIGN,
1441 NULL);
1442 if (nfs_page_cachep == NULL)
1443 return -ENOMEM;
1444
1445 return 0;
1446}
1447
1448void nfs_destroy_nfspagecache(void)
1449{
1450 kmem_cache_destroy(nfs_page_cachep);
1451}
1452
1453static const struct rpc_call_ops nfs_pgio_common_ops = {
1454 .rpc_call_prepare = nfs_pgio_prepare,
1455 .rpc_call_done = nfs_pgio_result,
1456 .rpc_release = nfs_pgio_release,
1457};
1458
1459const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1460 .pg_test = nfs_generic_pg_test,
1461 .pg_doio = nfs_generic_pg_pgios,
1462};
1/*
2 * linux/fs/nfs/pagelist.c
3 *
4 * A set of helper functions for managing NFS read and write requests.
5 * The main purpose of these routines is to provide support for the
6 * coalescing of several requests into a single RPC call.
7 *
8 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
9 *
10 */
11
12#include <linux/slab.h>
13#include <linux/file.h>
14#include <linux/sched.h>
15#include <linux/sunrpc/clnt.h>
16#include <linux/nfs.h>
17#include <linux/nfs3.h>
18#include <linux/nfs4.h>
19#include <linux/nfs_page.h>
20#include <linux/nfs_fs.h>
21#include <linux/nfs_mount.h>
22#include <linux/export.h>
23
24#include "internal.h"
25#include "pnfs.h"
26
27static struct kmem_cache *nfs_page_cachep;
28
29bool nfs_pgarray_set(struct nfs_page_array *p, unsigned int pagecount)
30{
31 p->npages = pagecount;
32 if (pagecount <= ARRAY_SIZE(p->page_array))
33 p->pagevec = p->page_array;
34 else {
35 p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
36 if (!p->pagevec)
37 p->npages = 0;
38 }
39 return p->pagevec != NULL;
40}
41
42void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
43 struct nfs_pgio_header *hdr,
44 void (*release)(struct nfs_pgio_header *hdr))
45{
46 hdr->req = nfs_list_entry(desc->pg_list.next);
47 hdr->inode = desc->pg_inode;
48 hdr->cred = hdr->req->wb_context->cred;
49 hdr->io_start = req_offset(hdr->req);
50 hdr->good_bytes = desc->pg_count;
51 hdr->dreq = desc->pg_dreq;
52 hdr->layout_private = desc->pg_layout_private;
53 hdr->release = release;
54 hdr->completion_ops = desc->pg_completion_ops;
55 if (hdr->completion_ops->init_hdr)
56 hdr->completion_ops->init_hdr(hdr);
57}
58EXPORT_SYMBOL_GPL(nfs_pgheader_init);
59
60void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
61{
62 spin_lock(&hdr->lock);
63 if (pos < hdr->io_start + hdr->good_bytes) {
64 set_bit(NFS_IOHDR_ERROR, &hdr->flags);
65 clear_bit(NFS_IOHDR_EOF, &hdr->flags);
66 hdr->good_bytes = pos - hdr->io_start;
67 hdr->error = error;
68 }
69 spin_unlock(&hdr->lock);
70}
71
72static inline struct nfs_page *
73nfs_page_alloc(void)
74{
75 struct nfs_page *p = kmem_cache_zalloc(nfs_page_cachep, GFP_NOIO);
76 if (p)
77 INIT_LIST_HEAD(&p->wb_list);
78 return p;
79}
80
81static inline void
82nfs_page_free(struct nfs_page *p)
83{
84 kmem_cache_free(nfs_page_cachep, p);
85}
86
87static void
88nfs_iocounter_inc(struct nfs_io_counter *c)
89{
90 atomic_inc(&c->io_count);
91}
92
93static void
94nfs_iocounter_dec(struct nfs_io_counter *c)
95{
96 if (atomic_dec_and_test(&c->io_count)) {
97 clear_bit(NFS_IO_INPROGRESS, &c->flags);
98 smp_mb__after_clear_bit();
99 wake_up_bit(&c->flags, NFS_IO_INPROGRESS);
100 }
101}
102
103static int
104__nfs_iocounter_wait(struct nfs_io_counter *c)
105{
106 wait_queue_head_t *wq = bit_waitqueue(&c->flags, NFS_IO_INPROGRESS);
107 DEFINE_WAIT_BIT(q, &c->flags, NFS_IO_INPROGRESS);
108 int ret = 0;
109
110 do {
111 prepare_to_wait(wq, &q.wait, TASK_KILLABLE);
112 set_bit(NFS_IO_INPROGRESS, &c->flags);
113 if (atomic_read(&c->io_count) == 0)
114 break;
115 ret = nfs_wait_bit_killable(&c->flags);
116 } while (atomic_read(&c->io_count) != 0);
117 finish_wait(wq, &q.wait);
118 return ret;
119}
120
121/**
122 * nfs_iocounter_wait - wait for i/o to complete
123 * @c: nfs_io_counter to use
124 *
125 * returns -ERESTARTSYS if interrupted by a fatal signal.
126 * Otherwise returns 0 once the io_count hits 0.
127 */
128int
129nfs_iocounter_wait(struct nfs_io_counter *c)
130{
131 if (atomic_read(&c->io_count) == 0)
132 return 0;
133 return __nfs_iocounter_wait(c);
134}
135
136/**
137 * nfs_create_request - Create an NFS read/write request.
138 * @ctx: open context to use
139 * @inode: inode to which the request is attached
140 * @page: page to write
141 * @offset: starting offset within the page for the write
142 * @count: number of bytes to read/write
143 *
144 * The page must be locked by the caller. This makes sure we never
145 * create two different requests for the same page.
146 * User should ensure it is safe to sleep in this function.
147 */
148struct nfs_page *
149nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
150 struct page *page,
151 unsigned int offset, unsigned int count)
152{
153 struct nfs_page *req;
154 struct nfs_lock_context *l_ctx;
155
156 if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
157 return ERR_PTR(-EBADF);
158 /* try to allocate the request struct */
159 req = nfs_page_alloc();
160 if (req == NULL)
161 return ERR_PTR(-ENOMEM);
162
163 /* get lock context early so we can deal with alloc failures */
164 l_ctx = nfs_get_lock_context(ctx);
165 if (IS_ERR(l_ctx)) {
166 nfs_page_free(req);
167 return ERR_CAST(l_ctx);
168 }
169 req->wb_lock_context = l_ctx;
170 nfs_iocounter_inc(&l_ctx->io_count);
171
172 /* Initialize the request struct. Initially, we assume a
173 * long write-back delay. This will be adjusted in
174 * update_nfs_request below if the region is not locked. */
175 req->wb_page = page;
176 req->wb_index = page_file_index(page);
177 page_cache_get(page);
178 req->wb_offset = offset;
179 req->wb_pgbase = offset;
180 req->wb_bytes = count;
181 req->wb_context = get_nfs_open_context(ctx);
182 kref_init(&req->wb_kref);
183 return req;
184}
185
186/**
187 * nfs_unlock_request - Unlock request and wake up sleepers.
188 * @req:
189 */
190void nfs_unlock_request(struct nfs_page *req)
191{
192 if (!NFS_WBACK_BUSY(req)) {
193 printk(KERN_ERR "NFS: Invalid unlock attempted\n");
194 BUG();
195 }
196 smp_mb__before_clear_bit();
197 clear_bit(PG_BUSY, &req->wb_flags);
198 smp_mb__after_clear_bit();
199 wake_up_bit(&req->wb_flags, PG_BUSY);
200}
201
202/**
203 * nfs_unlock_and_release_request - Unlock request and release the nfs_page
204 * @req:
205 */
206void nfs_unlock_and_release_request(struct nfs_page *req)
207{
208 nfs_unlock_request(req);
209 nfs_release_request(req);
210}
211
212/*
213 * nfs_clear_request - Free up all resources allocated to the request
214 * @req:
215 *
216 * Release page and open context resources associated with a read/write
217 * request after it has completed.
218 */
219static void nfs_clear_request(struct nfs_page *req)
220{
221 struct page *page = req->wb_page;
222 struct nfs_open_context *ctx = req->wb_context;
223 struct nfs_lock_context *l_ctx = req->wb_lock_context;
224
225 if (page != NULL) {
226 page_cache_release(page);
227 req->wb_page = NULL;
228 }
229 if (l_ctx != NULL) {
230 nfs_iocounter_dec(&l_ctx->io_count);
231 nfs_put_lock_context(l_ctx);
232 req->wb_lock_context = NULL;
233 }
234 if (ctx != NULL) {
235 put_nfs_open_context(ctx);
236 req->wb_context = NULL;
237 }
238}
239
240
241/**
242 * nfs_release_request - Release the count on an NFS read/write request
243 * @req: request to release
244 *
245 * Note: Should never be called with the spinlock held!
246 */
247static void nfs_free_request(struct kref *kref)
248{
249 struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
250
251 /* Release struct file and open context */
252 nfs_clear_request(req);
253 nfs_page_free(req);
254}
255
256void nfs_release_request(struct nfs_page *req)
257{
258 kref_put(&req->wb_kref, nfs_free_request);
259}
260
261static int nfs_wait_bit_uninterruptible(void *word)
262{
263 io_schedule();
264 return 0;
265}
266
267/**
268 * nfs_wait_on_request - Wait for a request to complete.
269 * @req: request to wait upon.
270 *
271 * Interruptible by fatal signals only.
272 * The user is responsible for holding a count on the request.
273 */
274int
275nfs_wait_on_request(struct nfs_page *req)
276{
277 return wait_on_bit(&req->wb_flags, PG_BUSY,
278 nfs_wait_bit_uninterruptible,
279 TASK_UNINTERRUPTIBLE);
280}
281
282bool nfs_generic_pg_test(struct nfs_pageio_descriptor *desc, struct nfs_page *prev, struct nfs_page *req)
283{
284 /*
285 * FIXME: ideally we should be able to coalesce all requests
286 * that are not block boundary aligned, but currently this
287 * is problematic for the case of bsize < PAGE_CACHE_SIZE,
288 * since nfs_flush_multi and nfs_pagein_multi assume you
289 * can have only one struct nfs_page.
290 */
291 if (desc->pg_bsize < PAGE_SIZE)
292 return 0;
293
294 return desc->pg_count + req->wb_bytes <= desc->pg_bsize;
295}
296EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
297
298/**
299 * nfs_pageio_init - initialise a page io descriptor
300 * @desc: pointer to descriptor
301 * @inode: pointer to inode
302 * @doio: pointer to io function
303 * @bsize: io block size
304 * @io_flags: extra parameters for the io function
305 */
306void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
307 struct inode *inode,
308 const struct nfs_pageio_ops *pg_ops,
309 const struct nfs_pgio_completion_ops *compl_ops,
310 size_t bsize,
311 int io_flags)
312{
313 INIT_LIST_HEAD(&desc->pg_list);
314 desc->pg_bytes_written = 0;
315 desc->pg_count = 0;
316 desc->pg_bsize = bsize;
317 desc->pg_base = 0;
318 desc->pg_moreio = 0;
319 desc->pg_recoalesce = 0;
320 desc->pg_inode = inode;
321 desc->pg_ops = pg_ops;
322 desc->pg_completion_ops = compl_ops;
323 desc->pg_ioflags = io_flags;
324 desc->pg_error = 0;
325 desc->pg_lseg = NULL;
326 desc->pg_dreq = NULL;
327 desc->pg_layout_private = NULL;
328}
329EXPORT_SYMBOL_GPL(nfs_pageio_init);
330
331static bool nfs_match_open_context(const struct nfs_open_context *ctx1,
332 const struct nfs_open_context *ctx2)
333{
334 return ctx1->cred == ctx2->cred && ctx1->state == ctx2->state;
335}
336
337static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
338 const struct nfs_lock_context *l2)
339{
340 return l1->lockowner.l_owner == l2->lockowner.l_owner
341 && l1->lockowner.l_pid == l2->lockowner.l_pid;
342}
343
344/**
345 * nfs_can_coalesce_requests - test two requests for compatibility
346 * @prev: pointer to nfs_page
347 * @req: pointer to nfs_page
348 *
349 * The nfs_page structures 'prev' and 'req' are compared to ensure that the
350 * page data area they describe is contiguous, and that their RPC
351 * credentials, NFSv4 open state, and lockowners are the same.
352 *
353 * Return 'true' if this is the case, else return 'false'.
354 */
355static bool nfs_can_coalesce_requests(struct nfs_page *prev,
356 struct nfs_page *req,
357 struct nfs_pageio_descriptor *pgio)
358{
359 if (!nfs_match_open_context(req->wb_context, prev->wb_context))
360 return false;
361 if (req->wb_context->dentry->d_inode->i_flock != NULL &&
362 !nfs_match_lock_context(req->wb_lock_context, prev->wb_lock_context))
363 return false;
364 if (req->wb_pgbase != 0)
365 return false;
366 if (prev->wb_pgbase + prev->wb_bytes != PAGE_CACHE_SIZE)
367 return false;
368 if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
369 return false;
370 return pgio->pg_ops->pg_test(pgio, prev, req);
371}
372
373/**
374 * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
375 * @desc: destination io descriptor
376 * @req: request
377 *
378 * Returns true if the request 'req' was successfully coalesced into the
379 * existing list of pages 'desc'.
380 */
381static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
382 struct nfs_page *req)
383{
384 if (desc->pg_count != 0) {
385 struct nfs_page *prev;
386
387 prev = nfs_list_entry(desc->pg_list.prev);
388 if (!nfs_can_coalesce_requests(prev, req, desc))
389 return 0;
390 } else {
391 if (desc->pg_ops->pg_init)
392 desc->pg_ops->pg_init(desc, req);
393 desc->pg_base = req->wb_pgbase;
394 }
395 nfs_list_remove_request(req);
396 nfs_list_add_request(req, &desc->pg_list);
397 desc->pg_count += req->wb_bytes;
398 return 1;
399}
400
401/*
402 * Helper for nfs_pageio_add_request and nfs_pageio_complete
403 */
404static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
405{
406 if (!list_empty(&desc->pg_list)) {
407 int error = desc->pg_ops->pg_doio(desc);
408 if (error < 0)
409 desc->pg_error = error;
410 else
411 desc->pg_bytes_written += desc->pg_count;
412 }
413 if (list_empty(&desc->pg_list)) {
414 desc->pg_count = 0;
415 desc->pg_base = 0;
416 }
417}
418
419/**
420 * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
421 * @desc: destination io descriptor
422 * @req: request
423 *
424 * Returns true if the request 'req' was successfully coalesced into the
425 * existing list of pages 'desc'.
426 */
427static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
428 struct nfs_page *req)
429{
430 while (!nfs_pageio_do_add_request(desc, req)) {
431 desc->pg_moreio = 1;
432 nfs_pageio_doio(desc);
433 if (desc->pg_error < 0)
434 return 0;
435 desc->pg_moreio = 0;
436 if (desc->pg_recoalesce)
437 return 0;
438 }
439 return 1;
440}
441
442static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
443{
444 LIST_HEAD(head);
445
446 do {
447 list_splice_init(&desc->pg_list, &head);
448 desc->pg_bytes_written -= desc->pg_count;
449 desc->pg_count = 0;
450 desc->pg_base = 0;
451 desc->pg_recoalesce = 0;
452
453 while (!list_empty(&head)) {
454 struct nfs_page *req;
455
456 req = list_first_entry(&head, struct nfs_page, wb_list);
457 nfs_list_remove_request(req);
458 if (__nfs_pageio_add_request(desc, req))
459 continue;
460 if (desc->pg_error < 0)
461 return 0;
462 break;
463 }
464 } while (desc->pg_recoalesce);
465 return 1;
466}
467
468int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
469 struct nfs_page *req)
470{
471 int ret;
472
473 do {
474 ret = __nfs_pageio_add_request(desc, req);
475 if (ret)
476 break;
477 if (desc->pg_error < 0)
478 break;
479 ret = nfs_do_recoalesce(desc);
480 } while (ret);
481 return ret;
482}
483EXPORT_SYMBOL_GPL(nfs_pageio_add_request);
484
485/**
486 * nfs_pageio_complete - Complete I/O on an nfs_pageio_descriptor
487 * @desc: pointer to io descriptor
488 */
489void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
490{
491 for (;;) {
492 nfs_pageio_doio(desc);
493 if (!desc->pg_recoalesce)
494 break;
495 if (!nfs_do_recoalesce(desc))
496 break;
497 }
498}
499EXPORT_SYMBOL_GPL(nfs_pageio_complete);
500
501/**
502 * nfs_pageio_cond_complete - Conditional I/O completion
503 * @desc: pointer to io descriptor
504 * @index: page index
505 *
506 * It is important to ensure that processes don't try to take locks
507 * on non-contiguous ranges of pages as that might deadlock. This
508 * function should be called before attempting to wait on a locked
509 * nfs_page. It will complete the I/O if the page index 'index'
510 * is not contiguous with the existing list of pages in 'desc'.
511 */
512void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
513{
514 if (!list_empty(&desc->pg_list)) {
515 struct nfs_page *prev = nfs_list_entry(desc->pg_list.prev);
516 if (index != prev->wb_index + 1)
517 nfs_pageio_complete(desc);
518 }
519}
520
521int __init nfs_init_nfspagecache(void)
522{
523 nfs_page_cachep = kmem_cache_create("nfs_page",
524 sizeof(struct nfs_page),
525 0, SLAB_HWCACHE_ALIGN,
526 NULL);
527 if (nfs_page_cachep == NULL)
528 return -ENOMEM;
529
530 return 0;
531}
532
533void nfs_destroy_nfspagecache(void)
534{
535 kmem_cache_destroy(nfs_page_cachep);
536}
537