xref: /openbmc/linux/fs/nfs/pagelist.c (revision 3381df09)
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 
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 &desc->pg_mirrors[desc->pg_mirror_idx];
37 }
38 EXPORT_SYMBOL_GPL(nfs_pgio_current_mirror);
39 
40 void nfs_pgheader_init(struct nfs_pageio_descriptor *desc,
41 		       struct nfs_pgio_header *hdr,
42 		       void (*release)(struct nfs_pgio_header *hdr))
43 {
44 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
45 
46 
47 	hdr->req = nfs_list_entry(mirror->pg_list.next);
48 	hdr->inode = desc->pg_inode;
49 	hdr->cred = nfs_req_openctx(hdr->req)->cred;
50 	hdr->io_start = req_offset(hdr->req);
51 	hdr->good_bytes = mirror->pg_count;
52 	hdr->io_completion = desc->pg_io_completion;
53 	hdr->dreq = desc->pg_dreq;
54 	hdr->release = release;
55 	hdr->completion_ops = desc->pg_completion_ops;
56 	if (hdr->completion_ops->init_hdr)
57 		hdr->completion_ops->init_hdr(hdr);
58 
59 	hdr->pgio_mirror_idx = desc->pg_mirror_idx;
60 }
61 EXPORT_SYMBOL_GPL(nfs_pgheader_init);
62 
63 void nfs_set_pgio_error(struct nfs_pgio_header *hdr, int error, loff_t pos)
64 {
65 	unsigned int new = pos - hdr->io_start;
66 
67 	if (hdr->good_bytes > new) {
68 		hdr->good_bytes = new;
69 		clear_bit(NFS_IOHDR_EOF, &hdr->flags);
70 		if (!test_and_set_bit(NFS_IOHDR_ERROR, &hdr->flags))
71 			hdr->error = error;
72 	}
73 }
74 
75 static inline struct nfs_page *
76 nfs_page_alloc(void)
77 {
78 	struct nfs_page	*p = kmem_cache_zalloc(nfs_page_cachep, GFP_KERNEL);
79 	if (p)
80 		INIT_LIST_HEAD(&p->wb_list);
81 	return p;
82 }
83 
84 static inline void
85 nfs_page_free(struct nfs_page *p)
86 {
87 	kmem_cache_free(nfs_page_cachep, p);
88 }
89 
90 /**
91  * nfs_iocounter_wait - wait for i/o to complete
92  * @l_ctx: nfs_lock_context with io_counter to use
93  *
94  * returns -ERESTARTSYS if interrupted by a fatal signal.
95  * Otherwise returns 0 once the io_count hits 0.
96  */
97 int
98 nfs_iocounter_wait(struct nfs_lock_context *l_ctx)
99 {
100 	return wait_var_event_killable(&l_ctx->io_count,
101 				       !atomic_read(&l_ctx->io_count));
102 }
103 
104 /**
105  * nfs_async_iocounter_wait - wait on a rpc_waitqueue for I/O
106  * to complete
107  * @task: the rpc_task that should wait
108  * @l_ctx: nfs_lock_context with io_counter to check
109  *
110  * Returns true if there is outstanding I/O to wait on and the
111  * task has been put to sleep.
112  */
113 bool
114 nfs_async_iocounter_wait(struct rpc_task *task, struct nfs_lock_context *l_ctx)
115 {
116 	struct inode *inode = d_inode(l_ctx->open_context->dentry);
117 	bool ret = false;
118 
119 	if (atomic_read(&l_ctx->io_count) > 0) {
120 		rpc_sleep_on(&NFS_SERVER(inode)->uoc_rpcwaitq, task, NULL);
121 		ret = true;
122 	}
123 
124 	if (atomic_read(&l_ctx->io_count) == 0) {
125 		rpc_wake_up_queued_task(&NFS_SERVER(inode)->uoc_rpcwaitq, task);
126 		ret = false;
127 	}
128 
129 	return ret;
130 }
131 EXPORT_SYMBOL_GPL(nfs_async_iocounter_wait);
132 
133 /*
134  * nfs_page_lock_head_request - page lock the head of the page group
135  * @req: any member of the page group
136  */
137 struct nfs_page *
138 nfs_page_group_lock_head(struct nfs_page *req)
139 {
140 	struct nfs_page *head = req->wb_head;
141 
142 	while (!nfs_lock_request(head)) {
143 		int ret = nfs_wait_on_request(head);
144 		if (ret < 0)
145 			return ERR_PTR(ret);
146 	}
147 	if (head != req)
148 		kref_get(&head->wb_kref);
149 	return head;
150 }
151 
152 /*
153  * nfs_unroll_locks -  unlock all newly locked reqs and wait on @req
154  * @head: head request of page group, must be holding head lock
155  * @req: request that couldn't lock and needs to wait on the req bit lock
156  *
157  * This is a helper function for nfs_lock_and_join_requests
158  * returns 0 on success, < 0 on error.
159  */
160 static void
161 nfs_unroll_locks(struct nfs_page *head, struct nfs_page *req)
162 {
163 	struct nfs_page *tmp;
164 
165 	/* relinquish all the locks successfully grabbed this run */
166 	for (tmp = head->wb_this_page ; tmp != req; tmp = tmp->wb_this_page) {
167 		if (!kref_read(&tmp->wb_kref))
168 			continue;
169 		nfs_unlock_and_release_request(tmp);
170 	}
171 }
172 
173 /*
174  * nfs_page_group_lock_subreq -  try to lock a subrequest
175  * @head: head request of page group
176  * @subreq: request to lock
177  *
178  * This is a helper function for nfs_lock_and_join_requests which
179  * must be called with the head request and page group both locked.
180  * On error, it returns with the page group unlocked.
181  */
182 static int
183 nfs_page_group_lock_subreq(struct nfs_page *head, struct nfs_page *subreq)
184 {
185 	int ret;
186 
187 	if (!kref_get_unless_zero(&subreq->wb_kref))
188 		return 0;
189 	while (!nfs_lock_request(subreq)) {
190 		nfs_page_group_unlock(head);
191 		ret = nfs_wait_on_request(subreq);
192 		if (!ret)
193 			ret = nfs_page_group_lock(head);
194 		if (ret < 0) {
195 			nfs_unroll_locks(head, subreq);
196 			nfs_release_request(subreq);
197 			return ret;
198 		}
199 	}
200 	return 0;
201 }
202 
203 /*
204  * nfs_page_group_lock_subrequests -  try to lock the subrequests
205  * @head: head request of page group
206  *
207  * This is a helper function for nfs_lock_and_join_requests which
208  * must be called with the head request locked.
209  */
210 int nfs_page_group_lock_subrequests(struct nfs_page *head)
211 {
212 	struct nfs_page *subreq;
213 	int ret;
214 
215 	ret = nfs_page_group_lock(head);
216 	if (ret < 0)
217 		return ret;
218 	/* lock each request in the page group */
219 	for (subreq = head->wb_this_page; subreq != head;
220 			subreq = subreq->wb_this_page) {
221 		ret = nfs_page_group_lock_subreq(head, subreq);
222 		if (ret < 0)
223 			return ret;
224 	}
225 	nfs_page_group_unlock(head);
226 	return 0;
227 }
228 
229 /*
230  * nfs_page_set_headlock - set the request PG_HEADLOCK
231  * @req: request that is to be locked
232  *
233  * this lock must be held when modifying req->wb_head
234  *
235  * return 0 on success, < 0 on error
236  */
237 int
238 nfs_page_set_headlock(struct nfs_page *req)
239 {
240 	if (!test_and_set_bit(PG_HEADLOCK, &req->wb_flags))
241 		return 0;
242 
243 	set_bit(PG_CONTENDED1, &req->wb_flags);
244 	smp_mb__after_atomic();
245 	return wait_on_bit_lock(&req->wb_flags, PG_HEADLOCK,
246 				TASK_UNINTERRUPTIBLE);
247 }
248 
249 /*
250  * nfs_page_clear_headlock - clear the request PG_HEADLOCK
251  * @req: request that is to be locked
252  */
253 void
254 nfs_page_clear_headlock(struct nfs_page *req)
255 {
256 	smp_mb__before_atomic();
257 	clear_bit(PG_HEADLOCK, &req->wb_flags);
258 	smp_mb__after_atomic();
259 	if (!test_bit(PG_CONTENDED1, &req->wb_flags))
260 		return;
261 	wake_up_bit(&req->wb_flags, PG_HEADLOCK);
262 }
263 
264 /*
265  * nfs_page_group_lock - lock the head of the page group
266  * @req: request in group that is to be locked
267  *
268  * this lock must be held when traversing or modifying the page
269  * group list
270  *
271  * return 0 on success, < 0 on error
272  */
273 int
274 nfs_page_group_lock(struct nfs_page *req)
275 {
276 	int ret;
277 
278 	ret = nfs_page_set_headlock(req);
279 	if (ret || req->wb_head == req)
280 		return ret;
281 	return nfs_page_set_headlock(req->wb_head);
282 }
283 
284 /*
285  * nfs_page_group_unlock - unlock the head of the page group
286  * @req: request in group that is to be unlocked
287  */
288 void
289 nfs_page_group_unlock(struct nfs_page *req)
290 {
291 	if (req != req->wb_head)
292 		nfs_page_clear_headlock(req->wb_head);
293 	nfs_page_clear_headlock(req);
294 }
295 
296 /*
297  * nfs_page_group_sync_on_bit_locked
298  *
299  * must be called with page group lock held
300  */
301 static bool
302 nfs_page_group_sync_on_bit_locked(struct nfs_page *req, unsigned int bit)
303 {
304 	struct nfs_page *head = req->wb_head;
305 	struct nfs_page *tmp;
306 
307 	WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_flags));
308 	WARN_ON_ONCE(test_and_set_bit(bit, &req->wb_flags));
309 
310 	tmp = req->wb_this_page;
311 	while (tmp != req) {
312 		if (!test_bit(bit, &tmp->wb_flags))
313 			return false;
314 		tmp = tmp->wb_this_page;
315 	}
316 
317 	/* true! reset all bits */
318 	tmp = req;
319 	do {
320 		clear_bit(bit, &tmp->wb_flags);
321 		tmp = tmp->wb_this_page;
322 	} while (tmp != req);
323 
324 	return true;
325 }
326 
327 /*
328  * nfs_page_group_sync_on_bit - set bit on current request, but only
329  *   return true if the bit is set for all requests in page group
330  * @req - request in page group
331  * @bit - PG_* bit that is used to sync page group
332  */
333 bool nfs_page_group_sync_on_bit(struct nfs_page *req, unsigned int bit)
334 {
335 	bool ret;
336 
337 	nfs_page_group_lock(req);
338 	ret = nfs_page_group_sync_on_bit_locked(req, bit);
339 	nfs_page_group_unlock(req);
340 
341 	return ret;
342 }
343 
344 /*
345  * nfs_page_group_init - Initialize the page group linkage for @req
346  * @req - a new nfs request
347  * @prev - the previous request in page group, or NULL if @req is the first
348  *         or only request in the group (the head).
349  */
350 static inline void
351 nfs_page_group_init(struct nfs_page *req, struct nfs_page *prev)
352 {
353 	struct inode *inode;
354 	WARN_ON_ONCE(prev == req);
355 
356 	if (!prev) {
357 		/* a head request */
358 		req->wb_head = req;
359 		req->wb_this_page = req;
360 	} else {
361 		/* a subrequest */
362 		WARN_ON_ONCE(prev->wb_this_page != prev->wb_head);
363 		WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &prev->wb_head->wb_flags));
364 		req->wb_head = prev->wb_head;
365 		req->wb_this_page = prev->wb_this_page;
366 		prev->wb_this_page = req;
367 
368 		/* All subrequests take a ref on the head request until
369 		 * nfs_page_group_destroy is called */
370 		kref_get(&req->wb_head->wb_kref);
371 
372 		/* grab extra ref and bump the request count if head request
373 		 * has extra ref from the write/commit path to handle handoff
374 		 * between write and commit lists. */
375 		if (test_bit(PG_INODE_REF, &prev->wb_head->wb_flags)) {
376 			inode = page_file_mapping(req->wb_page)->host;
377 			set_bit(PG_INODE_REF, &req->wb_flags);
378 			kref_get(&req->wb_kref);
379 			atomic_long_inc(&NFS_I(inode)->nrequests);
380 		}
381 	}
382 }
383 
384 /*
385  * nfs_page_group_destroy - sync the destruction of page groups
386  * @req - request that no longer needs the page group
387  *
388  * releases the page group reference from each member once all
389  * members have called this function.
390  */
391 static void
392 nfs_page_group_destroy(struct kref *kref)
393 {
394 	struct nfs_page *req = container_of(kref, struct nfs_page, wb_kref);
395 	struct nfs_page *head = req->wb_head;
396 	struct nfs_page *tmp, *next;
397 
398 	if (!nfs_page_group_sync_on_bit(req, PG_TEARDOWN))
399 		goto out;
400 
401 	tmp = req;
402 	do {
403 		next = tmp->wb_this_page;
404 		/* unlink and free */
405 		tmp->wb_this_page = tmp;
406 		tmp->wb_head = tmp;
407 		nfs_free_request(tmp);
408 		tmp = next;
409 	} while (tmp != req);
410 out:
411 	/* subrequests must release the ref on the head request */
412 	if (head != req)
413 		nfs_release_request(head);
414 }
415 
416 static struct nfs_page *
417 __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
418 		   unsigned int pgbase, unsigned int offset,
419 		   unsigned int count)
420 {
421 	struct nfs_page		*req;
422 	struct nfs_open_context *ctx = l_ctx->open_context;
423 
424 	if (test_bit(NFS_CONTEXT_BAD, &ctx->flags))
425 		return ERR_PTR(-EBADF);
426 	/* try to allocate the request struct */
427 	req = nfs_page_alloc();
428 	if (req == NULL)
429 		return ERR_PTR(-ENOMEM);
430 
431 	req->wb_lock_context = l_ctx;
432 	refcount_inc(&l_ctx->count);
433 	atomic_inc(&l_ctx->io_count);
434 
435 	/* Initialize the request struct. Initially, we assume a
436 	 * long write-back delay. This will be adjusted in
437 	 * update_nfs_request below if the region is not locked. */
438 	req->wb_page    = page;
439 	if (page) {
440 		req->wb_index = page_index(page);
441 		get_page(page);
442 	}
443 	req->wb_offset  = offset;
444 	req->wb_pgbase	= pgbase;
445 	req->wb_bytes   = count;
446 	kref_init(&req->wb_kref);
447 	req->wb_nio = 0;
448 	return req;
449 }
450 
451 /**
452  * nfs_create_request - Create an NFS read/write request.
453  * @ctx: open context to use
454  * @page: page to write
455  * @offset: starting offset within the page for the write
456  * @count: number of bytes to read/write
457  *
458  * The page must be locked by the caller. This makes sure we never
459  * create two different requests for the same page.
460  * User should ensure it is safe to sleep in this function.
461  */
462 struct nfs_page *
463 nfs_create_request(struct nfs_open_context *ctx, struct page *page,
464 		   unsigned int offset, unsigned int count)
465 {
466 	struct nfs_lock_context *l_ctx = nfs_get_lock_context(ctx);
467 	struct nfs_page *ret;
468 
469 	if (IS_ERR(l_ctx))
470 		return ERR_CAST(l_ctx);
471 	ret = __nfs_create_request(l_ctx, page, offset, offset, count);
472 	if (!IS_ERR(ret))
473 		nfs_page_group_init(ret, NULL);
474 	nfs_put_lock_context(l_ctx);
475 	return ret;
476 }
477 
478 static struct nfs_page *
479 nfs_create_subreq(struct nfs_page *req,
480 		  unsigned int pgbase,
481 		  unsigned int offset,
482 		  unsigned int count)
483 {
484 	struct nfs_page *last;
485 	struct nfs_page *ret;
486 
487 	ret = __nfs_create_request(req->wb_lock_context, req->wb_page,
488 			pgbase, offset, count);
489 	if (!IS_ERR(ret)) {
490 		/* find the last request */
491 		for (last = req->wb_head;
492 		     last->wb_this_page != req->wb_head;
493 		     last = last->wb_this_page)
494 			;
495 
496 		nfs_lock_request(ret);
497 		ret->wb_index = req->wb_index;
498 		nfs_page_group_init(ret, last);
499 		ret->wb_nio = req->wb_nio;
500 	}
501 	return ret;
502 }
503 
504 /**
505  * nfs_unlock_request - Unlock request and wake up sleepers.
506  * @req: pointer to request
507  */
508 void nfs_unlock_request(struct nfs_page *req)
509 {
510 	if (!NFS_WBACK_BUSY(req)) {
511 		printk(KERN_ERR "NFS: Invalid unlock attempted\n");
512 		BUG();
513 	}
514 	smp_mb__before_atomic();
515 	clear_bit(PG_BUSY, &req->wb_flags);
516 	smp_mb__after_atomic();
517 	if (!test_bit(PG_CONTENDED2, &req->wb_flags))
518 		return;
519 	wake_up_bit(&req->wb_flags, PG_BUSY);
520 }
521 
522 /**
523  * nfs_unlock_and_release_request - Unlock request and release the nfs_page
524  * @req: pointer to request
525  */
526 void nfs_unlock_and_release_request(struct nfs_page *req)
527 {
528 	nfs_unlock_request(req);
529 	nfs_release_request(req);
530 }
531 
532 /*
533  * nfs_clear_request - Free up all resources allocated to the request
534  * @req:
535  *
536  * Release page and open context resources associated with a read/write
537  * request after it has completed.
538  */
539 static void nfs_clear_request(struct nfs_page *req)
540 {
541 	struct page *page = req->wb_page;
542 	struct nfs_lock_context *l_ctx = req->wb_lock_context;
543 	struct nfs_open_context *ctx;
544 
545 	if (page != NULL) {
546 		put_page(page);
547 		req->wb_page = NULL;
548 	}
549 	if (l_ctx != NULL) {
550 		if (atomic_dec_and_test(&l_ctx->io_count)) {
551 			wake_up_var(&l_ctx->io_count);
552 			ctx = l_ctx->open_context;
553 			if (test_bit(NFS_CONTEXT_UNLOCK, &ctx->flags))
554 				rpc_wake_up(&NFS_SERVER(d_inode(ctx->dentry))->uoc_rpcwaitq);
555 		}
556 		nfs_put_lock_context(l_ctx);
557 		req->wb_lock_context = NULL;
558 	}
559 }
560 
561 /**
562  * nfs_release_request - Release the count on an NFS read/write request
563  * @req: request to release
564  *
565  * Note: Should never be called with the spinlock held!
566  */
567 void nfs_free_request(struct nfs_page *req)
568 {
569 	WARN_ON_ONCE(req->wb_this_page != req);
570 
571 	/* extra debug: make sure no sync bits are still set */
572 	WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
573 	WARN_ON_ONCE(test_bit(PG_UNLOCKPAGE, &req->wb_flags));
574 	WARN_ON_ONCE(test_bit(PG_UPTODATE, &req->wb_flags));
575 	WARN_ON_ONCE(test_bit(PG_WB_END, &req->wb_flags));
576 	WARN_ON_ONCE(test_bit(PG_REMOVE, &req->wb_flags));
577 
578 	/* Release struct file and open context */
579 	nfs_clear_request(req);
580 	nfs_page_free(req);
581 }
582 
583 void nfs_release_request(struct nfs_page *req)
584 {
585 	kref_put(&req->wb_kref, nfs_page_group_destroy);
586 }
587 EXPORT_SYMBOL_GPL(nfs_release_request);
588 
589 /**
590  * nfs_wait_on_request - Wait for a request to complete.
591  * @req: request to wait upon.
592  *
593  * Interruptible by fatal signals only.
594  * The user is responsible for holding a count on the request.
595  */
596 int
597 nfs_wait_on_request(struct nfs_page *req)
598 {
599 	if (!test_bit(PG_BUSY, &req->wb_flags))
600 		return 0;
601 	set_bit(PG_CONTENDED2, &req->wb_flags);
602 	smp_mb__after_atomic();
603 	return wait_on_bit_io(&req->wb_flags, PG_BUSY,
604 			      TASK_UNINTERRUPTIBLE);
605 }
606 EXPORT_SYMBOL_GPL(nfs_wait_on_request);
607 
608 /*
609  * nfs_generic_pg_test - determine if requests can be coalesced
610  * @desc: pointer to descriptor
611  * @prev: previous request in desc, or NULL
612  * @req: this request
613  *
614  * Returns zero if @req cannot be coalesced into @desc, otherwise it returns
615  * the size of the request.
616  */
617 size_t nfs_generic_pg_test(struct nfs_pageio_descriptor *desc,
618 			   struct nfs_page *prev, struct nfs_page *req)
619 {
620 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
621 
622 
623 	if (mirror->pg_count > mirror->pg_bsize) {
624 		/* should never happen */
625 		WARN_ON_ONCE(1);
626 		return 0;
627 	}
628 
629 	/*
630 	 * Limit the request size so that we can still allocate a page array
631 	 * for it without upsetting the slab allocator.
632 	 */
633 	if (((mirror->pg_count + req->wb_bytes) >> PAGE_SHIFT) *
634 			sizeof(struct page *) > PAGE_SIZE)
635 		return 0;
636 
637 	return min(mirror->pg_bsize - mirror->pg_count, (size_t)req->wb_bytes);
638 }
639 EXPORT_SYMBOL_GPL(nfs_generic_pg_test);
640 
641 struct nfs_pgio_header *nfs_pgio_header_alloc(const struct nfs_rw_ops *ops)
642 {
643 	struct nfs_pgio_header *hdr = ops->rw_alloc_header();
644 
645 	if (hdr) {
646 		INIT_LIST_HEAD(&hdr->pages);
647 		hdr->rw_ops = ops;
648 	}
649 	return hdr;
650 }
651 EXPORT_SYMBOL_GPL(nfs_pgio_header_alloc);
652 
653 /**
654  * nfs_pgio_data_destroy - make @hdr suitable for reuse
655  *
656  * Frees memory and releases refs from nfs_generic_pgio, so that it may
657  * be called again.
658  *
659  * @hdr: A header that has had nfs_generic_pgio called
660  */
661 static void nfs_pgio_data_destroy(struct nfs_pgio_header *hdr)
662 {
663 	if (hdr->args.context)
664 		put_nfs_open_context(hdr->args.context);
665 	if (hdr->page_array.pagevec != hdr->page_array.page_array)
666 		kfree(hdr->page_array.pagevec);
667 }
668 
669 /*
670  * nfs_pgio_header_free - Free a read or write header
671  * @hdr: The header to free
672  */
673 void nfs_pgio_header_free(struct nfs_pgio_header *hdr)
674 {
675 	nfs_pgio_data_destroy(hdr);
676 	hdr->rw_ops->rw_free_header(hdr);
677 }
678 EXPORT_SYMBOL_GPL(nfs_pgio_header_free);
679 
680 /**
681  * nfs_pgio_rpcsetup - Set up arguments for a pageio call
682  * @hdr: The pageio hdr
683  * @count: Number of bytes to read
684  * @how: How to commit data (writes only)
685  * @cinfo: Commit information for the call (writes only)
686  */
687 static void nfs_pgio_rpcsetup(struct nfs_pgio_header *hdr,
688 			      unsigned int count,
689 			      int how, struct nfs_commit_info *cinfo)
690 {
691 	struct nfs_page *req = hdr->req;
692 
693 	/* Set up the RPC argument and reply structs
694 	 * NB: take care not to mess about with hdr->commit et al. */
695 
696 	hdr->args.fh     = NFS_FH(hdr->inode);
697 	hdr->args.offset = req_offset(req);
698 	/* pnfs_set_layoutcommit needs this */
699 	hdr->mds_offset = hdr->args.offset;
700 	hdr->args.pgbase = req->wb_pgbase;
701 	hdr->args.pages  = hdr->page_array.pagevec;
702 	hdr->args.count  = count;
703 	hdr->args.context = get_nfs_open_context(nfs_req_openctx(req));
704 	hdr->args.lock_context = req->wb_lock_context;
705 	hdr->args.stable  = NFS_UNSTABLE;
706 	switch (how & (FLUSH_STABLE | FLUSH_COND_STABLE)) {
707 	case 0:
708 		break;
709 	case FLUSH_COND_STABLE:
710 		if (nfs_reqs_to_commit(cinfo))
711 			break;
712 		/* fall through */
713 	default:
714 		hdr->args.stable = NFS_FILE_SYNC;
715 	}
716 
717 	hdr->res.fattr   = &hdr->fattr;
718 	hdr->res.count   = 0;
719 	hdr->res.eof     = 0;
720 	hdr->res.verf    = &hdr->verf;
721 	nfs_fattr_init(&hdr->fattr);
722 }
723 
724 /**
725  * nfs_pgio_prepare - Prepare pageio hdr to go over the wire
726  * @task: The current task
727  * @calldata: pageio header to prepare
728  */
729 static void nfs_pgio_prepare(struct rpc_task *task, void *calldata)
730 {
731 	struct nfs_pgio_header *hdr = calldata;
732 	int err;
733 	err = NFS_PROTO(hdr->inode)->pgio_rpc_prepare(task, hdr);
734 	if (err)
735 		rpc_exit(task, err);
736 }
737 
738 int nfs_initiate_pgio(struct rpc_clnt *clnt, struct nfs_pgio_header *hdr,
739 		      const struct cred *cred, const struct nfs_rpc_ops *rpc_ops,
740 		      const struct rpc_call_ops *call_ops, int how, int flags)
741 {
742 	struct rpc_task *task;
743 	struct rpc_message msg = {
744 		.rpc_argp = &hdr->args,
745 		.rpc_resp = &hdr->res,
746 		.rpc_cred = cred,
747 	};
748 	struct rpc_task_setup task_setup_data = {
749 		.rpc_client = clnt,
750 		.task = &hdr->task,
751 		.rpc_message = &msg,
752 		.callback_ops = call_ops,
753 		.callback_data = hdr,
754 		.workqueue = nfsiod_workqueue,
755 		.flags = RPC_TASK_ASYNC | RPC_TASK_CRED_NOREF | flags,
756 	};
757 
758 	hdr->rw_ops->rw_initiate(hdr, &msg, rpc_ops, &task_setup_data, how);
759 
760 	dprintk("NFS: initiated pgio call "
761 		"(req %s/%llu, %u bytes @ offset %llu)\n",
762 		hdr->inode->i_sb->s_id,
763 		(unsigned long long)NFS_FILEID(hdr->inode),
764 		hdr->args.count,
765 		(unsigned long long)hdr->args.offset);
766 
767 	task = rpc_run_task(&task_setup_data);
768 	if (IS_ERR(task))
769 		return PTR_ERR(task);
770 	rpc_put_task(task);
771 	return 0;
772 }
773 EXPORT_SYMBOL_GPL(nfs_initiate_pgio);
774 
775 /**
776  * nfs_pgio_error - Clean up from a pageio error
777  * @hdr: pageio header
778  */
779 static void nfs_pgio_error(struct nfs_pgio_header *hdr)
780 {
781 	set_bit(NFS_IOHDR_REDO, &hdr->flags);
782 	hdr->completion_ops->completion(hdr);
783 }
784 
785 /**
786  * nfs_pgio_release - Release pageio data
787  * @calldata: The pageio header to release
788  */
789 static void nfs_pgio_release(void *calldata)
790 {
791 	struct nfs_pgio_header *hdr = calldata;
792 	hdr->completion_ops->completion(hdr);
793 }
794 
795 static void nfs_pageio_mirror_init(struct nfs_pgio_mirror *mirror,
796 				   unsigned int bsize)
797 {
798 	INIT_LIST_HEAD(&mirror->pg_list);
799 	mirror->pg_bytes_written = 0;
800 	mirror->pg_count = 0;
801 	mirror->pg_bsize = bsize;
802 	mirror->pg_base = 0;
803 	mirror->pg_recoalesce = 0;
804 }
805 
806 /**
807  * nfs_pageio_init - initialise a page io descriptor
808  * @desc: pointer to descriptor
809  * @inode: pointer to inode
810  * @pg_ops: pointer to pageio operations
811  * @compl_ops: pointer to pageio completion operations
812  * @rw_ops: pointer to nfs read/write operations
813  * @bsize: io block size
814  * @io_flags: extra parameters for the io function
815  */
816 void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
817 		     struct inode *inode,
818 		     const struct nfs_pageio_ops *pg_ops,
819 		     const struct nfs_pgio_completion_ops *compl_ops,
820 		     const struct nfs_rw_ops *rw_ops,
821 		     size_t bsize,
822 		     int io_flags)
823 {
824 	desc->pg_moreio = 0;
825 	desc->pg_inode = inode;
826 	desc->pg_ops = pg_ops;
827 	desc->pg_completion_ops = compl_ops;
828 	desc->pg_rw_ops = rw_ops;
829 	desc->pg_ioflags = io_flags;
830 	desc->pg_error = 0;
831 	desc->pg_lseg = NULL;
832 	desc->pg_io_completion = NULL;
833 	desc->pg_dreq = NULL;
834 	desc->pg_bsize = bsize;
835 
836 	desc->pg_mirror_count = 1;
837 	desc->pg_mirror_idx = 0;
838 
839 	desc->pg_mirrors_dynamic = NULL;
840 	desc->pg_mirrors = desc->pg_mirrors_static;
841 	nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
842 	desc->pg_maxretrans = 0;
843 }
844 
845 /**
846  * nfs_pgio_result - Basic pageio error handling
847  * @task: The task that ran
848  * @calldata: Pageio header to check
849  */
850 static void nfs_pgio_result(struct rpc_task *task, void *calldata)
851 {
852 	struct nfs_pgio_header *hdr = calldata;
853 	struct inode *inode = hdr->inode;
854 
855 	dprintk("NFS: %s: %5u, (status %d)\n", __func__,
856 		task->tk_pid, task->tk_status);
857 
858 	if (hdr->rw_ops->rw_done(task, hdr, inode) != 0)
859 		return;
860 	if (task->tk_status < 0)
861 		nfs_set_pgio_error(hdr, task->tk_status, hdr->args.offset);
862 	else
863 		hdr->rw_ops->rw_result(task, hdr);
864 }
865 
866 /*
867  * Create an RPC task for the given read or write request and kick it.
868  * The page must have been locked by the caller.
869  *
870  * It may happen that the page we're passed is not marked dirty.
871  * This is the case if nfs_updatepage detects a conflicting request
872  * that has been written but not committed.
873  */
874 int nfs_generic_pgio(struct nfs_pageio_descriptor *desc,
875 		     struct nfs_pgio_header *hdr)
876 {
877 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
878 
879 	struct nfs_page		*req;
880 	struct page		**pages,
881 				*last_page;
882 	struct list_head *head = &mirror->pg_list;
883 	struct nfs_commit_info cinfo;
884 	struct nfs_page_array *pg_array = &hdr->page_array;
885 	unsigned int pagecount, pageused;
886 	gfp_t gfp_flags = GFP_KERNEL;
887 
888 	pagecount = nfs_page_array_len(mirror->pg_base, mirror->pg_count);
889 	pg_array->npages = pagecount;
890 
891 	if (pagecount <= ARRAY_SIZE(pg_array->page_array))
892 		pg_array->pagevec = pg_array->page_array;
893 	else {
894 		pg_array->pagevec = kcalloc(pagecount, sizeof(struct page *), gfp_flags);
895 		if (!pg_array->pagevec) {
896 			pg_array->npages = 0;
897 			nfs_pgio_error(hdr);
898 			desc->pg_error = -ENOMEM;
899 			return desc->pg_error;
900 		}
901 	}
902 
903 	nfs_init_cinfo(&cinfo, desc->pg_inode, desc->pg_dreq);
904 	pages = hdr->page_array.pagevec;
905 	last_page = NULL;
906 	pageused = 0;
907 	while (!list_empty(head)) {
908 		req = nfs_list_entry(head->next);
909 		nfs_list_move_request(req, &hdr->pages);
910 
911 		if (!last_page || last_page != req->wb_page) {
912 			pageused++;
913 			if (pageused > pagecount)
914 				break;
915 			*pages++ = last_page = req->wb_page;
916 		}
917 	}
918 	if (WARN_ON_ONCE(pageused != pagecount)) {
919 		nfs_pgio_error(hdr);
920 		desc->pg_error = -EINVAL;
921 		return desc->pg_error;
922 	}
923 
924 	if ((desc->pg_ioflags & FLUSH_COND_STABLE) &&
925 	    (desc->pg_moreio || nfs_reqs_to_commit(&cinfo)))
926 		desc->pg_ioflags &= ~FLUSH_COND_STABLE;
927 
928 	/* Set up the argument struct */
929 	nfs_pgio_rpcsetup(hdr, mirror->pg_count, desc->pg_ioflags, &cinfo);
930 	desc->pg_rpc_callops = &nfs_pgio_common_ops;
931 	return 0;
932 }
933 EXPORT_SYMBOL_GPL(nfs_generic_pgio);
934 
935 static int nfs_generic_pg_pgios(struct nfs_pageio_descriptor *desc)
936 {
937 	struct nfs_pgio_header *hdr;
938 	int ret;
939 
940 	hdr = nfs_pgio_header_alloc(desc->pg_rw_ops);
941 	if (!hdr) {
942 		desc->pg_error = -ENOMEM;
943 		return desc->pg_error;
944 	}
945 	nfs_pgheader_init(desc, hdr, nfs_pgio_header_free);
946 	ret = nfs_generic_pgio(desc, hdr);
947 	if (ret == 0)
948 		ret = nfs_initiate_pgio(NFS_CLIENT(hdr->inode),
949 					hdr,
950 					hdr->cred,
951 					NFS_PROTO(hdr->inode),
952 					desc->pg_rpc_callops,
953 					desc->pg_ioflags, 0);
954 	return ret;
955 }
956 
957 static struct nfs_pgio_mirror *
958 nfs_pageio_alloc_mirrors(struct nfs_pageio_descriptor *desc,
959 		unsigned int mirror_count)
960 {
961 	struct nfs_pgio_mirror *ret;
962 	unsigned int i;
963 
964 	kfree(desc->pg_mirrors_dynamic);
965 	desc->pg_mirrors_dynamic = NULL;
966 	if (mirror_count == 1)
967 		return desc->pg_mirrors_static;
968 	ret = kmalloc_array(mirror_count, sizeof(*ret), GFP_KERNEL);
969 	if (ret != NULL) {
970 		for (i = 0; i < mirror_count; i++)
971 			nfs_pageio_mirror_init(&ret[i], desc->pg_bsize);
972 		desc->pg_mirrors_dynamic = ret;
973 	}
974 	return ret;
975 }
976 
977 /*
978  * nfs_pageio_setup_mirroring - determine if mirroring is to be used
979  *				by calling the pg_get_mirror_count op
980  */
981 static void nfs_pageio_setup_mirroring(struct nfs_pageio_descriptor *pgio,
982 				       struct nfs_page *req)
983 {
984 	unsigned int mirror_count = 1;
985 
986 	if (pgio->pg_ops->pg_get_mirror_count)
987 		mirror_count = pgio->pg_ops->pg_get_mirror_count(pgio, req);
988 	if (mirror_count == pgio->pg_mirror_count || pgio->pg_error < 0)
989 		return;
990 
991 	if (!mirror_count || mirror_count > NFS_PAGEIO_DESCRIPTOR_MIRROR_MAX) {
992 		pgio->pg_error = -EINVAL;
993 		return;
994 	}
995 
996 	pgio->pg_mirrors = nfs_pageio_alloc_mirrors(pgio, mirror_count);
997 	if (pgio->pg_mirrors == NULL) {
998 		pgio->pg_error = -ENOMEM;
999 		pgio->pg_mirrors = pgio->pg_mirrors_static;
1000 		mirror_count = 1;
1001 	}
1002 	pgio->pg_mirror_count = mirror_count;
1003 }
1004 
1005 static void nfs_pageio_cleanup_mirroring(struct nfs_pageio_descriptor *pgio)
1006 {
1007 	pgio->pg_mirror_count = 1;
1008 	pgio->pg_mirror_idx = 0;
1009 	pgio->pg_mirrors = pgio->pg_mirrors_static;
1010 	kfree(pgio->pg_mirrors_dynamic);
1011 	pgio->pg_mirrors_dynamic = NULL;
1012 }
1013 
1014 static bool nfs_match_lock_context(const struct nfs_lock_context *l1,
1015 		const struct nfs_lock_context *l2)
1016 {
1017 	return l1->lockowner == l2->lockowner;
1018 }
1019 
1020 /**
1021  * nfs_coalesce_size - test two requests for compatibility
1022  * @prev: pointer to nfs_page
1023  * @req: pointer to nfs_page
1024  * @pgio: pointer to nfs_pagio_descriptor
1025  *
1026  * The nfs_page structures 'prev' and 'req' are compared to ensure that the
1027  * page data area they describe is contiguous, and that their RPC
1028  * credentials, NFSv4 open state, and lockowners are the same.
1029  *
1030  * Returns size of the request that can be coalesced
1031  */
1032 static unsigned int nfs_coalesce_size(struct nfs_page *prev,
1033 				      struct nfs_page *req,
1034 				      struct nfs_pageio_descriptor *pgio)
1035 {
1036 	struct file_lock_context *flctx;
1037 
1038 	if (prev) {
1039 		if (!nfs_match_open_context(nfs_req_openctx(req), nfs_req_openctx(prev)))
1040 			return 0;
1041 		flctx = d_inode(nfs_req_openctx(req)->dentry)->i_flctx;
1042 		if (flctx != NULL &&
1043 		    !(list_empty_careful(&flctx->flc_posix) &&
1044 		      list_empty_careful(&flctx->flc_flock)) &&
1045 		    !nfs_match_lock_context(req->wb_lock_context,
1046 					    prev->wb_lock_context))
1047 			return 0;
1048 		if (req_offset(req) != req_offset(prev) + prev->wb_bytes)
1049 			return 0;
1050 		if (req->wb_page == prev->wb_page) {
1051 			if (req->wb_pgbase != prev->wb_pgbase + prev->wb_bytes)
1052 				return 0;
1053 		} else {
1054 			if (req->wb_pgbase != 0 ||
1055 			    prev->wb_pgbase + prev->wb_bytes != PAGE_SIZE)
1056 				return 0;
1057 		}
1058 	}
1059 	return pgio->pg_ops->pg_test(pgio, prev, req);
1060 }
1061 
1062 /**
1063  * nfs_pageio_do_add_request - Attempt to coalesce a request into a page list.
1064  * @desc: destination io descriptor
1065  * @req: request
1066  *
1067  * If the request 'req' was successfully coalesced into the existing list
1068  * of pages 'desc', it returns the size of req.
1069  */
1070 static unsigned int
1071 nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
1072 		struct nfs_page *req)
1073 {
1074 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1075 	struct nfs_page *prev = NULL;
1076 	unsigned int size;
1077 
1078 	if (mirror->pg_count != 0) {
1079 		prev = nfs_list_entry(mirror->pg_list.prev);
1080 	} else {
1081 		if (desc->pg_ops->pg_init)
1082 			desc->pg_ops->pg_init(desc, req);
1083 		if (desc->pg_error < 0)
1084 			return 0;
1085 		mirror->pg_base = req->wb_pgbase;
1086 	}
1087 
1088 	if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
1089 		if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
1090 			desc->pg_error = -ETIMEDOUT;
1091 		else
1092 			desc->pg_error = -EIO;
1093 		return 0;
1094 	}
1095 
1096 	size = nfs_coalesce_size(prev, req, desc);
1097 	if (size < req->wb_bytes)
1098 		return size;
1099 	nfs_list_move_request(req, &mirror->pg_list);
1100 	mirror->pg_count += req->wb_bytes;
1101 	return req->wb_bytes;
1102 }
1103 
1104 /*
1105  * Helper for nfs_pageio_add_request and nfs_pageio_complete
1106  */
1107 static void nfs_pageio_doio(struct nfs_pageio_descriptor *desc)
1108 {
1109 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1110 
1111 
1112 	if (!list_empty(&mirror->pg_list)) {
1113 		int error = desc->pg_ops->pg_doio(desc);
1114 		if (error < 0)
1115 			desc->pg_error = error;
1116 		else
1117 			mirror->pg_bytes_written += mirror->pg_count;
1118 	}
1119 	if (list_empty(&mirror->pg_list)) {
1120 		mirror->pg_count = 0;
1121 		mirror->pg_base = 0;
1122 	}
1123 }
1124 
1125 static void
1126 nfs_pageio_cleanup_request(struct nfs_pageio_descriptor *desc,
1127 		struct nfs_page *req)
1128 {
1129 	LIST_HEAD(head);
1130 
1131 	nfs_list_move_request(req, &head);
1132 	desc->pg_completion_ops->error_cleanup(&head, desc->pg_error);
1133 }
1134 
1135 /**
1136  * nfs_pageio_add_request - Attempt to coalesce a request into a page list.
1137  * @desc: destination io descriptor
1138  * @req: request
1139  *
1140  * This may split a request into subrequests which are all part of the
1141  * same page group. If so, it will submit @req as the last one, to ensure
1142  * the pointer to @req is still valid in case of failure.
1143  *
1144  * Returns true if the request 'req' was successfully coalesced into the
1145  * existing list of pages 'desc'.
1146  */
1147 static int __nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1148 			   struct nfs_page *req)
1149 {
1150 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1151 	struct nfs_page *subreq;
1152 	unsigned int size, subreq_size;
1153 
1154 	nfs_page_group_lock(req);
1155 
1156 	subreq = req;
1157 	subreq_size = subreq->wb_bytes;
1158 	for(;;) {
1159 		size = nfs_pageio_do_add_request(desc, subreq);
1160 		if (size == subreq_size) {
1161 			/* We successfully submitted a request */
1162 			if (subreq == req)
1163 				break;
1164 			req->wb_pgbase += size;
1165 			req->wb_bytes -= size;
1166 			req->wb_offset += size;
1167 			subreq_size = req->wb_bytes;
1168 			subreq = req;
1169 			continue;
1170 		}
1171 		if (WARN_ON_ONCE(subreq != req)) {
1172 			nfs_page_group_unlock(req);
1173 			nfs_pageio_cleanup_request(desc, subreq);
1174 			subreq = req;
1175 			subreq_size = req->wb_bytes;
1176 			nfs_page_group_lock(req);
1177 		}
1178 		if (!size) {
1179 			/* Can't coalesce any more, so do I/O */
1180 			nfs_page_group_unlock(req);
1181 			desc->pg_moreio = 1;
1182 			nfs_pageio_doio(desc);
1183 			if (desc->pg_error < 0 || mirror->pg_recoalesce)
1184 				return 0;
1185 			/* retry add_request for this subreq */
1186 			nfs_page_group_lock(req);
1187 			continue;
1188 		}
1189 		subreq = nfs_create_subreq(req, req->wb_pgbase,
1190 				req->wb_offset, size);
1191 		if (IS_ERR(subreq))
1192 			goto err_ptr;
1193 		subreq_size = size;
1194 	}
1195 
1196 	nfs_page_group_unlock(req);
1197 	return 1;
1198 err_ptr:
1199 	desc->pg_error = PTR_ERR(subreq);
1200 	nfs_page_group_unlock(req);
1201 	return 0;
1202 }
1203 
1204 static int nfs_do_recoalesce(struct nfs_pageio_descriptor *desc)
1205 {
1206 	struct nfs_pgio_mirror *mirror = nfs_pgio_current_mirror(desc);
1207 	LIST_HEAD(head);
1208 
1209 	do {
1210 		list_splice_init(&mirror->pg_list, &head);
1211 		mirror->pg_bytes_written -= mirror->pg_count;
1212 		mirror->pg_count = 0;
1213 		mirror->pg_base = 0;
1214 		mirror->pg_recoalesce = 0;
1215 
1216 		while (!list_empty(&head)) {
1217 			struct nfs_page *req;
1218 
1219 			req = list_first_entry(&head, struct nfs_page, wb_list);
1220 			if (__nfs_pageio_add_request(desc, req))
1221 				continue;
1222 			if (desc->pg_error < 0) {
1223 				list_splice_tail(&head, &mirror->pg_list);
1224 				mirror->pg_recoalesce = 1;
1225 				return 0;
1226 			}
1227 			break;
1228 		}
1229 	} while (mirror->pg_recoalesce);
1230 	return 1;
1231 }
1232 
1233 static int nfs_pageio_add_request_mirror(struct nfs_pageio_descriptor *desc,
1234 		struct nfs_page *req)
1235 {
1236 	int ret;
1237 
1238 	do {
1239 		ret = __nfs_pageio_add_request(desc, req);
1240 		if (ret)
1241 			break;
1242 		if (desc->pg_error < 0)
1243 			break;
1244 		ret = nfs_do_recoalesce(desc);
1245 	} while (ret);
1246 
1247 	return ret;
1248 }
1249 
1250 static void nfs_pageio_error_cleanup(struct nfs_pageio_descriptor *desc)
1251 {
1252 	u32 midx;
1253 	struct nfs_pgio_mirror *mirror;
1254 
1255 	if (!desc->pg_error)
1256 		return;
1257 
1258 	for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1259 		mirror = &desc->pg_mirrors[midx];
1260 		desc->pg_completion_ops->error_cleanup(&mirror->pg_list,
1261 				desc->pg_error);
1262 	}
1263 }
1264 
1265 int nfs_pageio_add_request(struct nfs_pageio_descriptor *desc,
1266 			   struct nfs_page *req)
1267 {
1268 	u32 midx;
1269 	unsigned int pgbase, offset, bytes;
1270 	struct nfs_page *dupreq;
1271 
1272 	pgbase = req->wb_pgbase;
1273 	offset = req->wb_offset;
1274 	bytes = req->wb_bytes;
1275 
1276 	nfs_pageio_setup_mirroring(desc, req);
1277 	if (desc->pg_error < 0)
1278 		goto out_failed;
1279 
1280 	/* Create the mirror instances first, and fire them off */
1281 	for (midx = 1; midx < desc->pg_mirror_count; midx++) {
1282 		nfs_page_group_lock(req);
1283 
1284 		dupreq = nfs_create_subreq(req,
1285 				pgbase, offset, bytes);
1286 
1287 		nfs_page_group_unlock(req);
1288 		if (IS_ERR(dupreq)) {
1289 			desc->pg_error = PTR_ERR(dupreq);
1290 			goto out_failed;
1291 		}
1292 
1293 		desc->pg_mirror_idx = midx;
1294 		if (!nfs_pageio_add_request_mirror(desc, dupreq))
1295 			goto out_cleanup_subreq;
1296 	}
1297 
1298 	desc->pg_mirror_idx = 0;
1299 	if (!nfs_pageio_add_request_mirror(desc, req))
1300 		goto out_failed;
1301 
1302 	return 1;
1303 
1304 out_cleanup_subreq:
1305 	nfs_pageio_cleanup_request(desc, dupreq);
1306 out_failed:
1307 	nfs_pageio_error_cleanup(desc);
1308 	return 0;
1309 }
1310 
1311 /*
1312  * nfs_pageio_complete_mirror - Complete I/O on the current mirror of an
1313  *				nfs_pageio_descriptor
1314  * @desc: pointer to io descriptor
1315  * @mirror_idx: pointer to mirror index
1316  */
1317 static void nfs_pageio_complete_mirror(struct nfs_pageio_descriptor *desc,
1318 				       u32 mirror_idx)
1319 {
1320 	struct nfs_pgio_mirror *mirror = &desc->pg_mirrors[mirror_idx];
1321 	u32 restore_idx = desc->pg_mirror_idx;
1322 
1323 	desc->pg_mirror_idx = mirror_idx;
1324 	for (;;) {
1325 		nfs_pageio_doio(desc);
1326 		if (desc->pg_error < 0 || !mirror->pg_recoalesce)
1327 			break;
1328 		if (!nfs_do_recoalesce(desc))
1329 			break;
1330 	}
1331 	desc->pg_mirror_idx = restore_idx;
1332 }
1333 
1334 /*
1335  * nfs_pageio_resend - Transfer requests to new descriptor and resend
1336  * @hdr - the pgio header to move request from
1337  * @desc - the pageio descriptor to add requests to
1338  *
1339  * Try to move each request (nfs_page) from @hdr to @desc then attempt
1340  * to send them.
1341  *
1342  * Returns 0 on success and < 0 on error.
1343  */
1344 int nfs_pageio_resend(struct nfs_pageio_descriptor *desc,
1345 		      struct nfs_pgio_header *hdr)
1346 {
1347 	LIST_HEAD(pages);
1348 
1349 	desc->pg_io_completion = hdr->io_completion;
1350 	desc->pg_dreq = hdr->dreq;
1351 	list_splice_init(&hdr->pages, &pages);
1352 	while (!list_empty(&pages)) {
1353 		struct nfs_page *req = nfs_list_entry(pages.next);
1354 
1355 		if (!nfs_pageio_add_request(desc, req))
1356 			break;
1357 	}
1358 	nfs_pageio_complete(desc);
1359 	if (!list_empty(&pages)) {
1360 		int err = desc->pg_error < 0 ? desc->pg_error : -EIO;
1361 		hdr->completion_ops->error_cleanup(&pages, err);
1362 		nfs_set_pgio_error(hdr, err, hdr->io_start);
1363 		return err;
1364 	}
1365 	return 0;
1366 }
1367 EXPORT_SYMBOL_GPL(nfs_pageio_resend);
1368 
1369 /**
1370  * nfs_pageio_complete - Complete I/O then cleanup an nfs_pageio_descriptor
1371  * @desc: pointer to io descriptor
1372  */
1373 void nfs_pageio_complete(struct nfs_pageio_descriptor *desc)
1374 {
1375 	u32 midx;
1376 
1377 	for (midx = 0; midx < desc->pg_mirror_count; midx++)
1378 		nfs_pageio_complete_mirror(desc, midx);
1379 
1380 	if (desc->pg_error < 0)
1381 		nfs_pageio_error_cleanup(desc);
1382 	if (desc->pg_ops->pg_cleanup)
1383 		desc->pg_ops->pg_cleanup(desc);
1384 	nfs_pageio_cleanup_mirroring(desc);
1385 }
1386 
1387 /**
1388  * nfs_pageio_cond_complete - Conditional I/O completion
1389  * @desc: pointer to io descriptor
1390  * @index: page index
1391  *
1392  * It is important to ensure that processes don't try to take locks
1393  * on non-contiguous ranges of pages as that might deadlock. This
1394  * function should be called before attempting to wait on a locked
1395  * nfs_page. It will complete the I/O if the page index 'index'
1396  * is not contiguous with the existing list of pages in 'desc'.
1397  */
1398 void nfs_pageio_cond_complete(struct nfs_pageio_descriptor *desc, pgoff_t index)
1399 {
1400 	struct nfs_pgio_mirror *mirror;
1401 	struct nfs_page *prev;
1402 	u32 midx;
1403 
1404 	for (midx = 0; midx < desc->pg_mirror_count; midx++) {
1405 		mirror = &desc->pg_mirrors[midx];
1406 		if (!list_empty(&mirror->pg_list)) {
1407 			prev = nfs_list_entry(mirror->pg_list.prev);
1408 			if (index != prev->wb_index + 1) {
1409 				nfs_pageio_complete(desc);
1410 				break;
1411 			}
1412 		}
1413 	}
1414 }
1415 
1416 /*
1417  * nfs_pageio_stop_mirroring - stop using mirroring (set mirror count to 1)
1418  */
1419 void nfs_pageio_stop_mirroring(struct nfs_pageio_descriptor *pgio)
1420 {
1421 	nfs_pageio_complete(pgio);
1422 }
1423 
1424 int __init nfs_init_nfspagecache(void)
1425 {
1426 	nfs_page_cachep = kmem_cache_create("nfs_page",
1427 					    sizeof(struct nfs_page),
1428 					    0, SLAB_HWCACHE_ALIGN,
1429 					    NULL);
1430 	if (nfs_page_cachep == NULL)
1431 		return -ENOMEM;
1432 
1433 	return 0;
1434 }
1435 
1436 void nfs_destroy_nfspagecache(void)
1437 {
1438 	kmem_cache_destroy(nfs_page_cachep);
1439 }
1440 
1441 static const struct rpc_call_ops nfs_pgio_common_ops = {
1442 	.rpc_call_prepare = nfs_pgio_prepare,
1443 	.rpc_call_done = nfs_pgio_result,
1444 	.rpc_release = nfs_pgio_release,
1445 };
1446 
1447 const struct nfs_pageio_ops nfs_pgio_rw_ops = {
1448 	.pg_test = nfs_generic_pg_test,
1449 	.pg_doio = nfs_generic_pg_pgios,
1450 };
1451