1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/nfs/write.c
4  *
5  * Write file data over NFS.
6  *
7  * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
8  */
9 
10 #include <linux/types.h>
11 #include <linux/slab.h>
12 #include <linux/mm.h>
13 #include <linux/pagemap.h>
14 #include <linux/file.h>
15 #include <linux/writeback.h>
16 #include <linux/swap.h>
17 #include <linux/migrate.h>
18 
19 #include <linux/sunrpc/clnt.h>
20 #include <linux/nfs_fs.h>
21 #include <linux/nfs_mount.h>
22 #include <linux/nfs_page.h>
23 #include <linux/backing-dev.h>
24 #include <linux/export.h>
25 #include <linux/freezer.h>
26 #include <linux/wait.h>
27 #include <linux/iversion.h>
28 
29 #include <linux/uaccess.h>
30 #include <linux/sched/mm.h>
31 
32 #include "delegation.h"
33 #include "internal.h"
34 #include "iostat.h"
35 #include "nfs4_fs.h"
36 #include "fscache.h"
37 #include "pnfs.h"
38 
39 #include "nfstrace.h"
40 
41 #define NFSDBG_FACILITY		NFSDBG_PAGECACHE
42 
43 #define MIN_POOL_WRITE		(32)
44 #define MIN_POOL_COMMIT		(4)
45 
46 struct nfs_io_completion {
47 	void (*complete)(void *data);
48 	void *data;
49 	struct kref refcount;
50 };
51 
52 /*
53  * Local function declarations
54  */
55 static void nfs_redirty_request(struct nfs_page *req);
56 static const struct rpc_call_ops nfs_commit_ops;
57 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
58 static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
59 static const struct nfs_rw_ops nfs_rw_write_ops;
60 static void nfs_inode_remove_request(struct nfs_page *req);
61 static void nfs_clear_request_commit(struct nfs_page *req);
62 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
63 				      struct inode *inode);
64 static struct nfs_page *
65 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
66 						struct page *page);
67 
68 static struct kmem_cache *nfs_wdata_cachep;
69 static mempool_t *nfs_wdata_mempool;
70 static struct kmem_cache *nfs_cdata_cachep;
71 static mempool_t *nfs_commit_mempool;
72 
nfs_commitdata_alloc(void)73 struct nfs_commit_data *nfs_commitdata_alloc(void)
74 {
75 	struct nfs_commit_data *p;
76 
77 	p = kmem_cache_zalloc(nfs_cdata_cachep, nfs_io_gfp_mask());
78 	if (!p) {
79 		p = mempool_alloc(nfs_commit_mempool, GFP_NOWAIT);
80 		if (!p)
81 			return NULL;
82 		memset(p, 0, sizeof(*p));
83 	}
84 	INIT_LIST_HEAD(&p->pages);
85 	return p;
86 }
87 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
88 
nfs_commit_free(struct nfs_commit_data * p)89 void nfs_commit_free(struct nfs_commit_data *p)
90 {
91 	mempool_free(p, nfs_commit_mempool);
92 }
93 EXPORT_SYMBOL_GPL(nfs_commit_free);
94 
nfs_writehdr_alloc(void)95 static struct nfs_pgio_header *nfs_writehdr_alloc(void)
96 {
97 	struct nfs_pgio_header *p;
98 
99 	p = kmem_cache_zalloc(nfs_wdata_cachep, nfs_io_gfp_mask());
100 	if (!p) {
101 		p = mempool_alloc(nfs_wdata_mempool, GFP_NOWAIT);
102 		if (!p)
103 			return NULL;
104 		memset(p, 0, sizeof(*p));
105 	}
106 	p->rw_mode = FMODE_WRITE;
107 	return p;
108 }
109 
nfs_writehdr_free(struct nfs_pgio_header * hdr)110 static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
111 {
112 	mempool_free(hdr, nfs_wdata_mempool);
113 }
114 
nfs_io_completion_alloc(gfp_t gfp_flags)115 static struct nfs_io_completion *nfs_io_completion_alloc(gfp_t gfp_flags)
116 {
117 	return kmalloc(sizeof(struct nfs_io_completion), gfp_flags);
118 }
119 
nfs_io_completion_init(struct nfs_io_completion * ioc,void (* complete)(void *),void * data)120 static void nfs_io_completion_init(struct nfs_io_completion *ioc,
121 		void (*complete)(void *), void *data)
122 {
123 	ioc->complete = complete;
124 	ioc->data = data;
125 	kref_init(&ioc->refcount);
126 }
127 
nfs_io_completion_release(struct kref * kref)128 static void nfs_io_completion_release(struct kref *kref)
129 {
130 	struct nfs_io_completion *ioc = container_of(kref,
131 			struct nfs_io_completion, refcount);
132 	ioc->complete(ioc->data);
133 	kfree(ioc);
134 }
135 
nfs_io_completion_get(struct nfs_io_completion * ioc)136 static void nfs_io_completion_get(struct nfs_io_completion *ioc)
137 {
138 	if (ioc != NULL)
139 		kref_get(&ioc->refcount);
140 }
141 
nfs_io_completion_put(struct nfs_io_completion * ioc)142 static void nfs_io_completion_put(struct nfs_io_completion *ioc)
143 {
144 	if (ioc != NULL)
145 		kref_put(&ioc->refcount, nfs_io_completion_release);
146 }
147 
148 static void
nfs_page_set_inode_ref(struct nfs_page * req,struct inode * inode)149 nfs_page_set_inode_ref(struct nfs_page *req, struct inode *inode)
150 {
151 	if (!test_and_set_bit(PG_INODE_REF, &req->wb_flags)) {
152 		kref_get(&req->wb_kref);
153 		atomic_long_inc(&NFS_I(inode)->nrequests);
154 	}
155 }
156 
157 static int
nfs_cancel_remove_inode(struct nfs_page * req,struct inode * inode)158 nfs_cancel_remove_inode(struct nfs_page *req, struct inode *inode)
159 {
160 	int ret;
161 
162 	if (!test_bit(PG_REMOVE, &req->wb_flags))
163 		return 0;
164 	ret = nfs_page_group_lock(req);
165 	if (ret)
166 		return ret;
167 	if (test_and_clear_bit(PG_REMOVE, &req->wb_flags))
168 		nfs_page_set_inode_ref(req, inode);
169 	nfs_page_group_unlock(req);
170 	return 0;
171 }
172 
173 static struct nfs_page *
nfs_page_private_request(struct page * page)174 nfs_page_private_request(struct page *page)
175 {
176 	if (!PagePrivate(page))
177 		return NULL;
178 	return (struct nfs_page *)page_private(page);
179 }
180 
181 /*
182  * nfs_page_find_head_request_locked - find head request associated with @page
183  *
184  * must be called while holding the inode lock.
185  *
186  * returns matching head request with reference held, or NULL if not found.
187  */
188 static struct nfs_page *
nfs_page_find_private_request(struct page * page)189 nfs_page_find_private_request(struct page *page)
190 {
191 	struct address_space *mapping = page_file_mapping(page);
192 	struct nfs_page *req;
193 
194 	if (!PagePrivate(page))
195 		return NULL;
196 	spin_lock(&mapping->private_lock);
197 	req = nfs_page_private_request(page);
198 	if (req) {
199 		WARN_ON_ONCE(req->wb_head != req);
200 		kref_get(&req->wb_kref);
201 	}
202 	spin_unlock(&mapping->private_lock);
203 	return req;
204 }
205 
206 static struct nfs_page *
nfs_page_find_swap_request(struct page * page)207 nfs_page_find_swap_request(struct page *page)
208 {
209 	struct inode *inode = page_file_mapping(page)->host;
210 	struct nfs_inode *nfsi = NFS_I(inode);
211 	struct nfs_page *req = NULL;
212 	if (!PageSwapCache(page))
213 		return NULL;
214 	mutex_lock(&nfsi->commit_mutex);
215 	if (PageSwapCache(page)) {
216 		req = nfs_page_search_commits_for_head_request_locked(nfsi,
217 			page);
218 		if (req) {
219 			WARN_ON_ONCE(req->wb_head != req);
220 			kref_get(&req->wb_kref);
221 		}
222 	}
223 	mutex_unlock(&nfsi->commit_mutex);
224 	return req;
225 }
226 
227 /*
228  * nfs_page_find_head_request - find head request associated with @page
229  *
230  * returns matching head request with reference held, or NULL if not found.
231  */
nfs_page_find_head_request(struct page * page)232 static struct nfs_page *nfs_page_find_head_request(struct page *page)
233 {
234 	struct nfs_page *req;
235 
236 	req = nfs_page_find_private_request(page);
237 	if (!req)
238 		req = nfs_page_find_swap_request(page);
239 	return req;
240 }
241 
nfs_find_and_lock_page_request(struct page * page)242 static struct nfs_page *nfs_find_and_lock_page_request(struct page *page)
243 {
244 	struct inode *inode = page_file_mapping(page)->host;
245 	struct nfs_page *req, *head;
246 	int ret;
247 
248 	for (;;) {
249 		req = nfs_page_find_head_request(page);
250 		if (!req)
251 			return req;
252 		head = nfs_page_group_lock_head(req);
253 		if (head != req)
254 			nfs_release_request(req);
255 		if (IS_ERR(head))
256 			return head;
257 		ret = nfs_cancel_remove_inode(head, inode);
258 		if (ret < 0) {
259 			nfs_unlock_and_release_request(head);
260 			return ERR_PTR(ret);
261 		}
262 		/* Ensure that nobody removed the request before we locked it */
263 		if (head == nfs_page_private_request(page))
264 			break;
265 		if (PageSwapCache(page))
266 			break;
267 		nfs_unlock_and_release_request(head);
268 	}
269 	return head;
270 }
271 
272 /* Adjust the file length if we're writing beyond the end */
nfs_grow_file(struct page * page,unsigned int offset,unsigned int count)273 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
274 {
275 	struct inode *inode = page_file_mapping(page)->host;
276 	loff_t end, i_size;
277 	pgoff_t end_index;
278 
279 	spin_lock(&inode->i_lock);
280 	i_size = i_size_read(inode);
281 	end_index = (i_size - 1) >> PAGE_SHIFT;
282 	if (i_size > 0 && page_index(page) < end_index)
283 		goto out;
284 	end = page_file_offset(page) + ((loff_t)offset+count);
285 	if (i_size >= end)
286 		goto out;
287 	trace_nfs_size_grow(inode, end);
288 	i_size_write(inode, end);
289 	NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
290 	nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
291 out:
292 	spin_unlock(&inode->i_lock);
293 	nfs_fscache_invalidate(inode, 0);
294 }
295 
296 /* A writeback failed: mark the page as bad, and invalidate the page cache */
nfs_set_pageerror(struct address_space * mapping)297 static void nfs_set_pageerror(struct address_space *mapping)
298 {
299 	struct inode *inode = mapping->host;
300 
301 	nfs_zap_mapping(mapping->host, mapping);
302 	/* Force file size revalidation */
303 	spin_lock(&inode->i_lock);
304 	nfs_set_cache_invalid(inode, NFS_INO_REVAL_FORCED |
305 					     NFS_INO_INVALID_CHANGE |
306 					     NFS_INO_INVALID_SIZE);
307 	spin_unlock(&inode->i_lock);
308 }
309 
nfs_mapping_set_error(struct page * page,int error)310 static void nfs_mapping_set_error(struct page *page, int error)
311 {
312 	struct address_space *mapping = page_file_mapping(page);
313 
314 	SetPageError(page);
315 	filemap_set_wb_err(mapping, error);
316 	if (mapping->host)
317 		errseq_set(&mapping->host->i_sb->s_wb_err,
318 			   error == -ENOSPC ? -ENOSPC : -EIO);
319 	nfs_set_pageerror(mapping);
320 }
321 
322 /*
323  * nfs_page_group_search_locked
324  * @head - head request of page group
325  * @page_offset - offset into page
326  *
327  * Search page group with head @head to find a request that contains the
328  * page offset @page_offset.
329  *
330  * Returns a pointer to the first matching nfs request, or NULL if no
331  * match is found.
332  *
333  * Must be called with the page group lock held
334  */
335 static struct nfs_page *
nfs_page_group_search_locked(struct nfs_page * head,unsigned int page_offset)336 nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
337 {
338 	struct nfs_page *req;
339 
340 	req = head;
341 	do {
342 		if (page_offset >= req->wb_pgbase &&
343 		    page_offset < (req->wb_pgbase + req->wb_bytes))
344 			return req;
345 
346 		req = req->wb_this_page;
347 	} while (req != head);
348 
349 	return NULL;
350 }
351 
352 /*
353  * nfs_page_group_covers_page
354  * @head - head request of page group
355  *
356  * Return true if the page group with head @head covers the whole page,
357  * returns false otherwise
358  */
nfs_page_group_covers_page(struct nfs_page * req)359 static bool nfs_page_group_covers_page(struct nfs_page *req)
360 {
361 	struct nfs_page *tmp;
362 	unsigned int pos = 0;
363 	unsigned int len = nfs_page_length(req->wb_page);
364 
365 	nfs_page_group_lock(req);
366 
367 	for (;;) {
368 		tmp = nfs_page_group_search_locked(req->wb_head, pos);
369 		if (!tmp)
370 			break;
371 		pos = tmp->wb_pgbase + tmp->wb_bytes;
372 	}
373 
374 	nfs_page_group_unlock(req);
375 	return pos >= len;
376 }
377 
378 /* We can set the PG_uptodate flag if we see that a write request
379  * covers the full page.
380  */
nfs_mark_uptodate(struct nfs_page * req)381 static void nfs_mark_uptodate(struct nfs_page *req)
382 {
383 	if (PageUptodate(req->wb_page))
384 		return;
385 	if (!nfs_page_group_covers_page(req))
386 		return;
387 	SetPageUptodate(req->wb_page);
388 }
389 
wb_priority(struct writeback_control * wbc)390 static int wb_priority(struct writeback_control *wbc)
391 {
392 	int ret = 0;
393 
394 	if (wbc->sync_mode == WB_SYNC_ALL)
395 		ret = FLUSH_COND_STABLE;
396 	return ret;
397 }
398 
399 /*
400  * NFS congestion control
401  */
402 
403 int nfs_congestion_kb;
404 
405 #define NFS_CONGESTION_ON_THRESH 	(nfs_congestion_kb >> (PAGE_SHIFT-10))
406 #define NFS_CONGESTION_OFF_THRESH	\
407 	(NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
408 
nfs_set_page_writeback(struct page * page)409 static void nfs_set_page_writeback(struct page *page)
410 {
411 	struct inode *inode = page_file_mapping(page)->host;
412 	struct nfs_server *nfss = NFS_SERVER(inode);
413 	int ret = test_set_page_writeback(page);
414 
415 	WARN_ON_ONCE(ret != 0);
416 
417 	if (atomic_long_inc_return(&nfss->writeback) >
418 			NFS_CONGESTION_ON_THRESH)
419 		nfss->write_congested = 1;
420 }
421 
nfs_end_page_writeback(struct nfs_page * req)422 static void nfs_end_page_writeback(struct nfs_page *req)
423 {
424 	struct inode *inode = page_file_mapping(req->wb_page)->host;
425 	struct nfs_server *nfss = NFS_SERVER(inode);
426 	bool is_done;
427 
428 	is_done = nfs_page_group_sync_on_bit(req, PG_WB_END);
429 	nfs_unlock_request(req);
430 	if (!is_done)
431 		return;
432 
433 	end_page_writeback(req->wb_page);
434 	if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
435 		nfss->write_congested = 0;
436 }
437 
438 /*
439  * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
440  *
441  * @destroy_list - request list (using wb_this_page) terminated by @old_head
442  * @old_head - the old head of the list
443  *
444  * All subrequests must be locked and removed from all lists, so at this point
445  * they are only "active" in this function, and possibly in nfs_wait_on_request
446  * with a reference held by some other context.
447  */
448 static void
nfs_destroy_unlinked_subrequests(struct nfs_page * destroy_list,struct nfs_page * old_head,struct inode * inode)449 nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
450 				 struct nfs_page *old_head,
451 				 struct inode *inode)
452 {
453 	while (destroy_list) {
454 		struct nfs_page *subreq = destroy_list;
455 
456 		destroy_list = (subreq->wb_this_page == old_head) ?
457 				   NULL : subreq->wb_this_page;
458 
459 		/* Note: lock subreq in order to change subreq->wb_head */
460 		nfs_page_set_headlock(subreq);
461 		WARN_ON_ONCE(old_head != subreq->wb_head);
462 
463 		/* make sure old group is not used */
464 		subreq->wb_this_page = subreq;
465 		subreq->wb_head = subreq;
466 
467 		clear_bit(PG_REMOVE, &subreq->wb_flags);
468 
469 		/* Note: races with nfs_page_group_destroy() */
470 		if (!kref_read(&subreq->wb_kref)) {
471 			/* Check if we raced with nfs_page_group_destroy() */
472 			if (test_and_clear_bit(PG_TEARDOWN, &subreq->wb_flags)) {
473 				nfs_page_clear_headlock(subreq);
474 				nfs_free_request(subreq);
475 			} else
476 				nfs_page_clear_headlock(subreq);
477 			continue;
478 		}
479 		nfs_page_clear_headlock(subreq);
480 
481 		nfs_release_request(old_head);
482 
483 		if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags)) {
484 			nfs_release_request(subreq);
485 			atomic_long_dec(&NFS_I(inode)->nrequests);
486 		}
487 
488 		/* subreq is now totally disconnected from page group or any
489 		 * write / commit lists. last chance to wake any waiters */
490 		nfs_unlock_and_release_request(subreq);
491 	}
492 }
493 
494 /*
495  * nfs_join_page_group - destroy subrequests of the head req
496  * @head: the page used to lookup the "page group" of nfs_page structures
497  * @inode: Inode to which the request belongs.
498  *
499  * This function joins all sub requests to the head request by first
500  * locking all requests in the group, cancelling any pending operations
501  * and finally updating the head request to cover the whole range covered by
502  * the (former) group.  All subrequests are removed from any write or commit
503  * lists, unlinked from the group and destroyed.
504  */
505 void
nfs_join_page_group(struct nfs_page * head,struct inode * inode)506 nfs_join_page_group(struct nfs_page *head, struct inode *inode)
507 {
508 	struct nfs_page *subreq;
509 	struct nfs_page *destroy_list = NULL;
510 	unsigned int pgbase, off, bytes;
511 
512 	pgbase = head->wb_pgbase;
513 	bytes = head->wb_bytes;
514 	off = head->wb_offset;
515 	for (subreq = head->wb_this_page; subreq != head;
516 			subreq = subreq->wb_this_page) {
517 		/* Subrequests should always form a contiguous range */
518 		if (pgbase > subreq->wb_pgbase) {
519 			off -= pgbase - subreq->wb_pgbase;
520 			bytes += pgbase - subreq->wb_pgbase;
521 			pgbase = subreq->wb_pgbase;
522 		}
523 		bytes = max(subreq->wb_pgbase + subreq->wb_bytes
524 				- pgbase, bytes);
525 	}
526 
527 	/* Set the head request's range to cover the former page group */
528 	head->wb_pgbase = pgbase;
529 	head->wb_bytes = bytes;
530 	head->wb_offset = off;
531 
532 	/* Now that all requests are locked, make sure they aren't on any list.
533 	 * Commit list removal accounting is done after locks are dropped */
534 	subreq = head;
535 	do {
536 		nfs_clear_request_commit(subreq);
537 		subreq = subreq->wb_this_page;
538 	} while (subreq != head);
539 
540 	/* unlink subrequests from head, destroy them later */
541 	if (head->wb_this_page != head) {
542 		/* destroy list will be terminated by head */
543 		destroy_list = head->wb_this_page;
544 		head->wb_this_page = head;
545 	}
546 
547 	nfs_destroy_unlinked_subrequests(destroy_list, head, inode);
548 }
549 
550 /*
551  * nfs_lock_and_join_requests - join all subreqs to the head req
552  * @page: the page used to lookup the "page group" of nfs_page structures
553  *
554  * This function joins all sub requests to the head request by first
555  * locking all requests in the group, cancelling any pending operations
556  * and finally updating the head request to cover the whole range covered by
557  * the (former) group.  All subrequests are removed from any write or commit
558  * lists, unlinked from the group and destroyed.
559  *
560  * Returns a locked, referenced pointer to the head request - which after
561  * this call is guaranteed to be the only request associated with the page.
562  * Returns NULL if no requests are found for @page, or a ERR_PTR if an
563  * error was encountered.
564  */
565 static struct nfs_page *
nfs_lock_and_join_requests(struct page * page)566 nfs_lock_and_join_requests(struct page *page)
567 {
568 	struct inode *inode = page_file_mapping(page)->host;
569 	struct nfs_page *head;
570 	int ret;
571 
572 	/*
573 	 * A reference is taken only on the head request which acts as a
574 	 * reference to the whole page group - the group will not be destroyed
575 	 * until the head reference is released.
576 	 */
577 	head = nfs_find_and_lock_page_request(page);
578 	if (IS_ERR_OR_NULL(head))
579 		return head;
580 
581 	/* lock each request in the page group */
582 	ret = nfs_page_group_lock_subrequests(head);
583 	if (ret < 0) {
584 		nfs_unlock_and_release_request(head);
585 		return ERR_PTR(ret);
586 	}
587 
588 	nfs_join_page_group(head, inode);
589 
590 	return head;
591 }
592 
nfs_write_error(struct nfs_page * req,int error)593 static void nfs_write_error(struct nfs_page *req, int error)
594 {
595 	trace_nfs_write_error(page_file_mapping(req->wb_page)->host, req,
596 			      error);
597 	nfs_mapping_set_error(req->wb_page, error);
598 	nfs_inode_remove_request(req);
599 	nfs_end_page_writeback(req);
600 	nfs_release_request(req);
601 }
602 
603 /*
604  * Find an associated nfs write request, and prepare to flush it out
605  * May return an error if the user signalled nfs_wait_on_request().
606  */
nfs_page_async_flush(struct page * page,struct writeback_control * wbc,struct nfs_pageio_descriptor * pgio)607 static int nfs_page_async_flush(struct page *page,
608 				struct writeback_control *wbc,
609 				struct nfs_pageio_descriptor *pgio)
610 {
611 	struct nfs_page *req;
612 	int ret = 0;
613 
614 	req = nfs_lock_and_join_requests(page);
615 	if (!req)
616 		goto out;
617 	ret = PTR_ERR(req);
618 	if (IS_ERR(req))
619 		goto out;
620 
621 	nfs_set_page_writeback(page);
622 	WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
623 
624 	/* If there is a fatal error that covers this write, just exit */
625 	ret = pgio->pg_error;
626 	if (nfs_error_is_fatal_on_server(ret))
627 		goto out_launder;
628 
629 	ret = 0;
630 	if (!nfs_pageio_add_request(pgio, req)) {
631 		ret = pgio->pg_error;
632 		/*
633 		 * Remove the problematic req upon fatal errors on the server
634 		 */
635 		if (nfs_error_is_fatal_on_server(ret))
636 			goto out_launder;
637 		if (wbc->sync_mode == WB_SYNC_NONE)
638 			ret = AOP_WRITEPAGE_ACTIVATE;
639 		redirty_page_for_writepage(wbc, page);
640 		nfs_redirty_request(req);
641 		pgio->pg_error = 0;
642 	} else
643 		nfs_add_stats(page_file_mapping(page)->host,
644 				NFSIOS_WRITEPAGES, 1);
645 out:
646 	return ret;
647 out_launder:
648 	nfs_write_error(req, ret);
649 	return 0;
650 }
651 
nfs_do_writepage(struct page * page,struct writeback_control * wbc,struct nfs_pageio_descriptor * pgio)652 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc,
653 			    struct nfs_pageio_descriptor *pgio)
654 {
655 	nfs_pageio_cond_complete(pgio, page_index(page));
656 	return nfs_page_async_flush(page, wbc, pgio);
657 }
658 
659 /*
660  * Write an mmapped page to the server.
661  */
nfs_writepage_locked(struct page * page,struct writeback_control * wbc)662 static int nfs_writepage_locked(struct page *page,
663 				struct writeback_control *wbc)
664 {
665 	struct nfs_pageio_descriptor pgio;
666 	struct inode *inode = page_file_mapping(page)->host;
667 	int err;
668 
669 	if (wbc->sync_mode == WB_SYNC_NONE &&
670 	    NFS_SERVER(inode)->write_congested)
671 		return AOP_WRITEPAGE_ACTIVATE;
672 
673 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
674 	nfs_pageio_init_write(&pgio, inode, 0,
675 				false, &nfs_async_write_completion_ops);
676 	err = nfs_do_writepage(page, wbc, &pgio);
677 	pgio.pg_error = 0;
678 	nfs_pageio_complete(&pgio);
679 	return err;
680 }
681 
nfs_writepage(struct page * page,struct writeback_control * wbc)682 int nfs_writepage(struct page *page, struct writeback_control *wbc)
683 {
684 	int ret;
685 
686 	ret = nfs_writepage_locked(page, wbc);
687 	if (ret != AOP_WRITEPAGE_ACTIVATE)
688 		unlock_page(page);
689 	return ret;
690 }
691 
nfs_writepages_callback(struct page * page,struct writeback_control * wbc,void * data)692 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
693 {
694 	int ret;
695 
696 	ret = nfs_do_writepage(page, wbc, data);
697 	if (ret != AOP_WRITEPAGE_ACTIVATE)
698 		unlock_page(page);
699 	return ret;
700 }
701 
nfs_io_completion_commit(void * inode)702 static void nfs_io_completion_commit(void *inode)
703 {
704 	nfs_commit_inode(inode, 0);
705 }
706 
nfs_writepages(struct address_space * mapping,struct writeback_control * wbc)707 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
708 {
709 	struct inode *inode = mapping->host;
710 	struct nfs_pageio_descriptor pgio;
711 	struct nfs_io_completion *ioc = NULL;
712 	unsigned int mntflags = NFS_SERVER(inode)->flags;
713 	int priority = 0;
714 	int err;
715 
716 	if (wbc->sync_mode == WB_SYNC_NONE &&
717 	    NFS_SERVER(inode)->write_congested)
718 		return 0;
719 
720 	nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
721 
722 	if (!(mntflags & NFS_MOUNT_WRITE_EAGER) || wbc->for_kupdate ||
723 	    wbc->for_background || wbc->for_sync || wbc->for_reclaim) {
724 		ioc = nfs_io_completion_alloc(GFP_KERNEL);
725 		if (ioc)
726 			nfs_io_completion_init(ioc, nfs_io_completion_commit,
727 					       inode);
728 		priority = wb_priority(wbc);
729 	}
730 
731 	do {
732 		nfs_pageio_init_write(&pgio, inode, priority, false,
733 				      &nfs_async_write_completion_ops);
734 		pgio.pg_io_completion = ioc;
735 		err = write_cache_pages(mapping, wbc, nfs_writepages_callback,
736 					&pgio);
737 		pgio.pg_error = 0;
738 		nfs_pageio_complete(&pgio);
739 	} while (err < 0 && !nfs_error_is_fatal(err));
740 	nfs_io_completion_put(ioc);
741 
742 	if (err < 0)
743 		goto out_err;
744 	return 0;
745 out_err:
746 	return err;
747 }
748 
749 /*
750  * Insert a write request into an inode
751  */
nfs_inode_add_request(struct inode * inode,struct nfs_page * req)752 static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
753 {
754 	struct address_space *mapping = page_file_mapping(req->wb_page);
755 	struct nfs_inode *nfsi = NFS_I(inode);
756 
757 	WARN_ON_ONCE(req->wb_this_page != req);
758 
759 	/* Lock the request! */
760 	nfs_lock_request(req);
761 
762 	/*
763 	 * Swap-space should not get truncated. Hence no need to plug the race
764 	 * with invalidate/truncate.
765 	 */
766 	spin_lock(&mapping->private_lock);
767 	if (likely(!PageSwapCache(req->wb_page))) {
768 		set_bit(PG_MAPPED, &req->wb_flags);
769 		SetPagePrivate(req->wb_page);
770 		set_page_private(req->wb_page, (unsigned long)req);
771 	}
772 	spin_unlock(&mapping->private_lock);
773 	atomic_long_inc(&nfsi->nrequests);
774 	/* this a head request for a page group - mark it as having an
775 	 * extra reference so sub groups can follow suit.
776 	 * This flag also informs pgio layer when to bump nrequests when
777 	 * adding subrequests. */
778 	WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
779 	kref_get(&req->wb_kref);
780 }
781 
782 /*
783  * Remove a write request from an inode
784  */
nfs_inode_remove_request(struct nfs_page * req)785 static void nfs_inode_remove_request(struct nfs_page *req)
786 {
787 	struct address_space *mapping = page_file_mapping(req->wb_page);
788 	struct inode *inode = mapping->host;
789 	struct nfs_inode *nfsi = NFS_I(inode);
790 	struct nfs_page *head;
791 
792 	if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
793 		head = req->wb_head;
794 
795 		spin_lock(&mapping->private_lock);
796 		if (likely(head->wb_page && !PageSwapCache(head->wb_page))) {
797 			set_page_private(head->wb_page, 0);
798 			ClearPagePrivate(head->wb_page);
799 			clear_bit(PG_MAPPED, &head->wb_flags);
800 		}
801 		spin_unlock(&mapping->private_lock);
802 	}
803 
804 	if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags)) {
805 		nfs_release_request(req);
806 		atomic_long_dec(&nfsi->nrequests);
807 	}
808 }
809 
810 static void
nfs_mark_request_dirty(struct nfs_page * req)811 nfs_mark_request_dirty(struct nfs_page *req)
812 {
813 	if (req->wb_page)
814 		__set_page_dirty_nobuffers(req->wb_page);
815 }
816 
817 /*
818  * nfs_page_search_commits_for_head_request_locked
819  *
820  * Search through commit lists on @inode for the head request for @page.
821  * Must be called while holding the inode (which is cinfo) lock.
822  *
823  * Returns the head request if found, or NULL if not found.
824  */
825 static struct nfs_page *
nfs_page_search_commits_for_head_request_locked(struct nfs_inode * nfsi,struct page * page)826 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
827 						struct page *page)
828 {
829 	struct nfs_page *freq, *t;
830 	struct nfs_commit_info cinfo;
831 	struct inode *inode = &nfsi->vfs_inode;
832 
833 	nfs_init_cinfo_from_inode(&cinfo, inode);
834 
835 	/* search through pnfs commit lists */
836 	freq = pnfs_search_commit_reqs(inode, &cinfo, page);
837 	if (freq)
838 		return freq->wb_head;
839 
840 	/* Linearly search the commit list for the correct request */
841 	list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
842 		if (freq->wb_page == page)
843 			return freq->wb_head;
844 	}
845 
846 	return NULL;
847 }
848 
849 /**
850  * nfs_request_add_commit_list_locked - add request to a commit list
851  * @req: pointer to a struct nfs_page
852  * @dst: commit list head
853  * @cinfo: holds list lock and accounting info
854  *
855  * This sets the PG_CLEAN bit, updates the cinfo count of
856  * number of outstanding requests requiring a commit as well as
857  * the MM page stats.
858  *
859  * The caller must hold NFS_I(cinfo->inode)->commit_mutex, and the
860  * nfs_page lock.
861  */
862 void
nfs_request_add_commit_list_locked(struct nfs_page * req,struct list_head * dst,struct nfs_commit_info * cinfo)863 nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
864 			    struct nfs_commit_info *cinfo)
865 {
866 	set_bit(PG_CLEAN, &req->wb_flags);
867 	nfs_list_add_request(req, dst);
868 	atomic_long_inc(&cinfo->mds->ncommit);
869 }
870 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
871 
872 /**
873  * nfs_request_add_commit_list - add request to a commit list
874  * @req: pointer to a struct nfs_page
875  * @cinfo: holds list lock and accounting info
876  *
877  * This sets the PG_CLEAN bit, updates the cinfo count of
878  * number of outstanding requests requiring a commit as well as
879  * the MM page stats.
880  *
881  * The caller must _not_ hold the cinfo->lock, but must be
882  * holding the nfs_page lock.
883  */
884 void
nfs_request_add_commit_list(struct nfs_page * req,struct nfs_commit_info * cinfo)885 nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo)
886 {
887 	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
888 	nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo);
889 	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
890 	if (req->wb_page)
891 		nfs_mark_page_unstable(req->wb_page, cinfo);
892 }
893 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
894 
895 /**
896  * nfs_request_remove_commit_list - Remove request from a commit list
897  * @req: pointer to a nfs_page
898  * @cinfo: holds list lock and accounting info
899  *
900  * This clears the PG_CLEAN bit, and updates the cinfo's count of
901  * number of outstanding requests requiring a commit
902  * It does not update the MM page stats.
903  *
904  * The caller _must_ hold the cinfo->lock and the nfs_page lock.
905  */
906 void
nfs_request_remove_commit_list(struct nfs_page * req,struct nfs_commit_info * cinfo)907 nfs_request_remove_commit_list(struct nfs_page *req,
908 			       struct nfs_commit_info *cinfo)
909 {
910 	if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
911 		return;
912 	nfs_list_remove_request(req);
913 	atomic_long_dec(&cinfo->mds->ncommit);
914 }
915 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
916 
nfs_init_cinfo_from_inode(struct nfs_commit_info * cinfo,struct inode * inode)917 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
918 				      struct inode *inode)
919 {
920 	cinfo->inode = inode;
921 	cinfo->mds = &NFS_I(inode)->commit_info;
922 	cinfo->ds = pnfs_get_ds_info(inode);
923 	cinfo->dreq = NULL;
924 	cinfo->completion_ops = &nfs_commit_completion_ops;
925 }
926 
nfs_init_cinfo(struct nfs_commit_info * cinfo,struct inode * inode,struct nfs_direct_req * dreq)927 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
928 		    struct inode *inode,
929 		    struct nfs_direct_req *dreq)
930 {
931 	if (dreq)
932 		nfs_init_cinfo_from_dreq(cinfo, dreq);
933 	else
934 		nfs_init_cinfo_from_inode(cinfo, inode);
935 }
936 EXPORT_SYMBOL_GPL(nfs_init_cinfo);
937 
938 /*
939  * Add a request to the inode's commit list.
940  */
941 void
nfs_mark_request_commit(struct nfs_page * req,struct pnfs_layout_segment * lseg,struct nfs_commit_info * cinfo,u32 ds_commit_idx)942 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
943 			struct nfs_commit_info *cinfo, u32 ds_commit_idx)
944 {
945 	if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
946 		return;
947 	nfs_request_add_commit_list(req, cinfo);
948 }
949 
950 static void
nfs_clear_page_commit(struct page * page)951 nfs_clear_page_commit(struct page *page)
952 {
953 	dec_node_page_state(page, NR_WRITEBACK);
954 	dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
955 		    WB_WRITEBACK);
956 }
957 
958 /* Called holding the request lock on @req */
959 static void
nfs_clear_request_commit(struct nfs_page * req)960 nfs_clear_request_commit(struct nfs_page *req)
961 {
962 	if (test_bit(PG_CLEAN, &req->wb_flags)) {
963 		struct nfs_open_context *ctx = nfs_req_openctx(req);
964 		struct inode *inode = d_inode(ctx->dentry);
965 		struct nfs_commit_info cinfo;
966 
967 		nfs_init_cinfo_from_inode(&cinfo, inode);
968 		mutex_lock(&NFS_I(inode)->commit_mutex);
969 		if (!pnfs_clear_request_commit(req, &cinfo)) {
970 			nfs_request_remove_commit_list(req, &cinfo);
971 		}
972 		mutex_unlock(&NFS_I(inode)->commit_mutex);
973 		nfs_clear_page_commit(req->wb_page);
974 	}
975 }
976 
nfs_write_need_commit(struct nfs_pgio_header * hdr)977 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
978 {
979 	if (hdr->verf.committed == NFS_DATA_SYNC)
980 		return hdr->lseg == NULL;
981 	return hdr->verf.committed != NFS_FILE_SYNC;
982 }
983 
nfs_async_write_init(struct nfs_pgio_header * hdr)984 static void nfs_async_write_init(struct nfs_pgio_header *hdr)
985 {
986 	nfs_io_completion_get(hdr->io_completion);
987 }
988 
nfs_write_completion(struct nfs_pgio_header * hdr)989 static void nfs_write_completion(struct nfs_pgio_header *hdr)
990 {
991 	struct nfs_commit_info cinfo;
992 	unsigned long bytes = 0;
993 
994 	if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
995 		goto out;
996 	nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
997 	while (!list_empty(&hdr->pages)) {
998 		struct nfs_page *req = nfs_list_entry(hdr->pages.next);
999 
1000 		bytes += req->wb_bytes;
1001 		nfs_list_remove_request(req);
1002 		if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
1003 		    (hdr->good_bytes < bytes)) {
1004 			trace_nfs_comp_error(hdr->inode, req, hdr->error);
1005 			nfs_mapping_set_error(req->wb_page, hdr->error);
1006 			goto remove_req;
1007 		}
1008 		if (nfs_write_need_commit(hdr)) {
1009 			/* Reset wb_nio, since the write was successful. */
1010 			req->wb_nio = 0;
1011 			memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
1012 			nfs_mark_request_commit(req, hdr->lseg, &cinfo,
1013 				hdr->pgio_mirror_idx);
1014 			goto next;
1015 		}
1016 remove_req:
1017 		nfs_inode_remove_request(req);
1018 next:
1019 		nfs_end_page_writeback(req);
1020 		nfs_release_request(req);
1021 	}
1022 out:
1023 	nfs_io_completion_put(hdr->io_completion);
1024 	hdr->release(hdr);
1025 }
1026 
1027 unsigned long
nfs_reqs_to_commit(struct nfs_commit_info * cinfo)1028 nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
1029 {
1030 	return atomic_long_read(&cinfo->mds->ncommit);
1031 }
1032 
1033 /* NFS_I(cinfo->inode)->commit_mutex held by caller */
1034 int
nfs_scan_commit_list(struct list_head * src,struct list_head * dst,struct nfs_commit_info * cinfo,int max)1035 nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
1036 		     struct nfs_commit_info *cinfo, int max)
1037 {
1038 	struct nfs_page *req, *tmp;
1039 	int ret = 0;
1040 
1041 	list_for_each_entry_safe(req, tmp, src, wb_list) {
1042 		kref_get(&req->wb_kref);
1043 		if (!nfs_lock_request(req)) {
1044 			nfs_release_request(req);
1045 			continue;
1046 		}
1047 		nfs_request_remove_commit_list(req, cinfo);
1048 		clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1049 		nfs_list_add_request(req, dst);
1050 		ret++;
1051 		if ((ret == max) && !cinfo->dreq)
1052 			break;
1053 		cond_resched();
1054 	}
1055 	return ret;
1056 }
1057 EXPORT_SYMBOL_GPL(nfs_scan_commit_list);
1058 
1059 /*
1060  * nfs_scan_commit - Scan an inode for commit requests
1061  * @inode: NFS inode to scan
1062  * @dst: mds destination list
1063  * @cinfo: mds and ds lists of reqs ready to commit
1064  *
1065  * Moves requests from the inode's 'commit' request list.
1066  * The requests are *not* checked to ensure that they form a contiguous set.
1067  */
1068 int
nfs_scan_commit(struct inode * inode,struct list_head * dst,struct nfs_commit_info * cinfo)1069 nfs_scan_commit(struct inode *inode, struct list_head *dst,
1070 		struct nfs_commit_info *cinfo)
1071 {
1072 	int ret = 0;
1073 
1074 	if (!atomic_long_read(&cinfo->mds->ncommit))
1075 		return 0;
1076 	mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1077 	if (atomic_long_read(&cinfo->mds->ncommit) > 0) {
1078 		const int max = INT_MAX;
1079 
1080 		ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1081 					   cinfo, max);
1082 		ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
1083 	}
1084 	mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1085 	return ret;
1086 }
1087 
1088 /*
1089  * Search for an existing write request, and attempt to update
1090  * it to reflect a new dirty region on a given page.
1091  *
1092  * If the attempt fails, then the existing request is flushed out
1093  * to disk.
1094  */
nfs_try_to_update_request(struct inode * inode,struct page * page,unsigned int offset,unsigned int bytes)1095 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1096 		struct page *page,
1097 		unsigned int offset,
1098 		unsigned int bytes)
1099 {
1100 	struct nfs_page *req;
1101 	unsigned int rqend;
1102 	unsigned int end;
1103 	int error;
1104 
1105 	end = offset + bytes;
1106 
1107 	req = nfs_lock_and_join_requests(page);
1108 	if (IS_ERR_OR_NULL(req))
1109 		return req;
1110 
1111 	rqend = req->wb_offset + req->wb_bytes;
1112 	/*
1113 	 * Tell the caller to flush out the request if
1114 	 * the offsets are non-contiguous.
1115 	 * Note: nfs_flush_incompatible() will already
1116 	 * have flushed out requests having wrong owners.
1117 	 */
1118 	if (offset > rqend || end < req->wb_offset)
1119 		goto out_flushme;
1120 
1121 	/* Okay, the request matches. Update the region */
1122 	if (offset < req->wb_offset) {
1123 		req->wb_offset = offset;
1124 		req->wb_pgbase = offset;
1125 	}
1126 	if (end > rqend)
1127 		req->wb_bytes = end - req->wb_offset;
1128 	else
1129 		req->wb_bytes = rqend - req->wb_offset;
1130 	req->wb_nio = 0;
1131 	return req;
1132 out_flushme:
1133 	/*
1134 	 * Note: we mark the request dirty here because
1135 	 * nfs_lock_and_join_requests() cannot preserve
1136 	 * commit flags, so we have to replay the write.
1137 	 */
1138 	nfs_mark_request_dirty(req);
1139 	nfs_unlock_and_release_request(req);
1140 	error = nfs_wb_page(inode, page);
1141 	return (error < 0) ? ERR_PTR(error) : NULL;
1142 }
1143 
1144 /*
1145  * Try to update an existing write request, or create one if there is none.
1146  *
1147  * Note: Should always be called with the Page Lock held to prevent races
1148  * if we have to add a new request. Also assumes that the caller has
1149  * already called nfs_flush_incompatible() if necessary.
1150  */
nfs_setup_write_request(struct nfs_open_context * ctx,struct page * page,unsigned int offset,unsigned int bytes)1151 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1152 		struct page *page, unsigned int offset, unsigned int bytes)
1153 {
1154 	struct inode *inode = page_file_mapping(page)->host;
1155 	struct nfs_page	*req;
1156 
1157 	req = nfs_try_to_update_request(inode, page, offset, bytes);
1158 	if (req != NULL)
1159 		goto out;
1160 	req = nfs_create_request(ctx, page, offset, bytes);
1161 	if (IS_ERR(req))
1162 		goto out;
1163 	nfs_inode_add_request(inode, req);
1164 out:
1165 	return req;
1166 }
1167 
nfs_writepage_setup(struct nfs_open_context * ctx,struct page * page,unsigned int offset,unsigned int count)1168 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1169 		unsigned int offset, unsigned int count)
1170 {
1171 	struct nfs_page	*req;
1172 
1173 	req = nfs_setup_write_request(ctx, page, offset, count);
1174 	if (IS_ERR(req))
1175 		return PTR_ERR(req);
1176 	/* Update file length */
1177 	nfs_grow_file(page, offset, count);
1178 	nfs_mark_uptodate(req);
1179 	nfs_mark_request_dirty(req);
1180 	nfs_unlock_and_release_request(req);
1181 	return 0;
1182 }
1183 
nfs_flush_incompatible(struct file * file,struct page * page)1184 int nfs_flush_incompatible(struct file *file, struct page *page)
1185 {
1186 	struct nfs_open_context *ctx = nfs_file_open_context(file);
1187 	struct nfs_lock_context *l_ctx;
1188 	struct file_lock_context *flctx = file_inode(file)->i_flctx;
1189 	struct nfs_page	*req;
1190 	int do_flush, status;
1191 	/*
1192 	 * Look for a request corresponding to this page. If there
1193 	 * is one, and it belongs to another file, we flush it out
1194 	 * before we try to copy anything into the page. Do this
1195 	 * due to the lack of an ACCESS-type call in NFSv2.
1196 	 * Also do the same if we find a request from an existing
1197 	 * dropped page.
1198 	 */
1199 	do {
1200 		req = nfs_page_find_head_request(page);
1201 		if (req == NULL)
1202 			return 0;
1203 		l_ctx = req->wb_lock_context;
1204 		do_flush = req->wb_page != page ||
1205 			!nfs_match_open_context(nfs_req_openctx(req), ctx);
1206 		if (l_ctx && flctx &&
1207 		    !(list_empty_careful(&flctx->flc_posix) &&
1208 		      list_empty_careful(&flctx->flc_flock))) {
1209 			do_flush |= l_ctx->lockowner != current->files;
1210 		}
1211 		nfs_release_request(req);
1212 		if (!do_flush)
1213 			return 0;
1214 		status = nfs_wb_page(page_file_mapping(page)->host, page);
1215 	} while (status == 0);
1216 	return status;
1217 }
1218 
1219 /*
1220  * Avoid buffered writes when a open context credential's key would
1221  * expire soon.
1222  *
1223  * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1224  *
1225  * Return 0 and set a credential flag which triggers the inode to flush
1226  * and performs  NFS_FILE_SYNC writes if the key will expired within
1227  * RPC_KEY_EXPIRE_TIMEO.
1228  */
1229 int
nfs_key_timeout_notify(struct file * filp,struct inode * inode)1230 nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1231 {
1232 	struct nfs_open_context *ctx = nfs_file_open_context(filp);
1233 
1234 	if (nfs_ctx_key_to_expire(ctx, inode) &&
1235 	    !rcu_access_pointer(ctx->ll_cred))
1236 		/* Already expired! */
1237 		return -EACCES;
1238 	return 0;
1239 }
1240 
1241 /*
1242  * Test if the open context credential key is marked to expire soon.
1243  */
nfs_ctx_key_to_expire(struct nfs_open_context * ctx,struct inode * inode)1244 bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx, struct inode *inode)
1245 {
1246 	struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1247 	struct rpc_cred *cred, *new, *old = NULL;
1248 	struct auth_cred acred = {
1249 		.cred = ctx->cred,
1250 	};
1251 	bool ret = false;
1252 
1253 	rcu_read_lock();
1254 	cred = rcu_dereference(ctx->ll_cred);
1255 	if (cred && !(cred->cr_ops->crkey_timeout &&
1256 		      cred->cr_ops->crkey_timeout(cred)))
1257 		goto out;
1258 	rcu_read_unlock();
1259 
1260 	new = auth->au_ops->lookup_cred(auth, &acred, 0);
1261 	if (new == cred) {
1262 		put_rpccred(new);
1263 		return true;
1264 	}
1265 	if (IS_ERR_OR_NULL(new)) {
1266 		new = NULL;
1267 		ret = true;
1268 	} else if (new->cr_ops->crkey_timeout &&
1269 		   new->cr_ops->crkey_timeout(new))
1270 		ret = true;
1271 
1272 	rcu_read_lock();
1273 	old = rcu_dereference_protected(xchg(&ctx->ll_cred,
1274 					     RCU_INITIALIZER(new)), 1);
1275 out:
1276 	rcu_read_unlock();
1277 	put_rpccred(old);
1278 	return ret;
1279 }
1280 
1281 /*
1282  * If the page cache is marked as unsafe or invalid, then we can't rely on
1283  * the PageUptodate() flag. In this case, we will need to turn off
1284  * write optimisations that depend on the page contents being correct.
1285  */
nfs_write_pageuptodate(struct page * page,struct inode * inode,unsigned int pagelen)1286 static bool nfs_write_pageuptodate(struct page *page, struct inode *inode,
1287 				   unsigned int pagelen)
1288 {
1289 	struct nfs_inode *nfsi = NFS_I(inode);
1290 
1291 	if (nfs_have_delegated_attributes(inode))
1292 		goto out;
1293 	if (nfsi->cache_validity &
1294 	    (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE))
1295 		return false;
1296 	smp_rmb();
1297 	if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags) && pagelen != 0)
1298 		return false;
1299 out:
1300 	if (nfsi->cache_validity & NFS_INO_INVALID_DATA && pagelen != 0)
1301 		return false;
1302 	return PageUptodate(page) != 0;
1303 }
1304 
1305 static bool
is_whole_file_wrlock(struct file_lock * fl)1306 is_whole_file_wrlock(struct file_lock *fl)
1307 {
1308 	return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1309 			fl->fl_type == F_WRLCK;
1310 }
1311 
1312 /* If we know the page is up to date, and we're not using byte range locks (or
1313  * if we have the whole file locked for writing), it may be more efficient to
1314  * extend the write to cover the entire page in order to avoid fragmentation
1315  * inefficiencies.
1316  *
1317  * If the file is opened for synchronous writes then we can just skip the rest
1318  * of the checks.
1319  */
nfs_can_extend_write(struct file * file,struct page * page,struct inode * inode,unsigned int pagelen)1320 static int nfs_can_extend_write(struct file *file, struct page *page,
1321 				struct inode *inode, unsigned int pagelen)
1322 {
1323 	int ret;
1324 	struct file_lock_context *flctx = inode->i_flctx;
1325 	struct file_lock *fl;
1326 
1327 	if (file->f_flags & O_DSYNC)
1328 		return 0;
1329 	if (!nfs_write_pageuptodate(page, inode, pagelen))
1330 		return 0;
1331 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1332 		return 1;
1333 	if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1334 		       list_empty_careful(&flctx->flc_posix)))
1335 		return 1;
1336 
1337 	/* Check to see if there are whole file write locks */
1338 	ret = 0;
1339 	spin_lock(&flctx->flc_lock);
1340 	if (!list_empty(&flctx->flc_posix)) {
1341 		fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1342 					fl_list);
1343 		if (is_whole_file_wrlock(fl))
1344 			ret = 1;
1345 	} else if (!list_empty(&flctx->flc_flock)) {
1346 		fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1347 					fl_list);
1348 		if (fl->fl_type == F_WRLCK)
1349 			ret = 1;
1350 	}
1351 	spin_unlock(&flctx->flc_lock);
1352 	return ret;
1353 }
1354 
1355 /*
1356  * Update and possibly write a cached page of an NFS file.
1357  *
1358  * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1359  * things with a page scheduled for an RPC call (e.g. invalidate it).
1360  */
nfs_updatepage(struct file * file,struct page * page,unsigned int offset,unsigned int count)1361 int nfs_updatepage(struct file *file, struct page *page,
1362 		unsigned int offset, unsigned int count)
1363 {
1364 	struct nfs_open_context *ctx = nfs_file_open_context(file);
1365 	struct address_space *mapping = page_file_mapping(page);
1366 	struct inode	*inode = mapping->host;
1367 	unsigned int	pagelen = nfs_page_length(page);
1368 	int		status = 0;
1369 
1370 	nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1371 
1372 	dprintk("NFS:       nfs_updatepage(%pD2 %d@%lld)\n",
1373 		file, count, (long long)(page_file_offset(page) + offset));
1374 
1375 	if (!count)
1376 		goto out;
1377 
1378 	if (nfs_can_extend_write(file, page, inode, pagelen)) {
1379 		count = max(count + offset, pagelen);
1380 		offset = 0;
1381 	}
1382 
1383 	status = nfs_writepage_setup(ctx, page, offset, count);
1384 	if (status < 0)
1385 		nfs_set_pageerror(mapping);
1386 out:
1387 	dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
1388 			status, (long long)i_size_read(inode));
1389 	return status;
1390 }
1391 
flush_task_priority(int how)1392 static int flush_task_priority(int how)
1393 {
1394 	switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1395 		case FLUSH_HIGHPRI:
1396 			return RPC_PRIORITY_HIGH;
1397 		case FLUSH_LOWPRI:
1398 			return RPC_PRIORITY_LOW;
1399 	}
1400 	return RPC_PRIORITY_NORMAL;
1401 }
1402 
nfs_initiate_write(struct nfs_pgio_header * hdr,struct rpc_message * msg,const struct nfs_rpc_ops * rpc_ops,struct rpc_task_setup * task_setup_data,int how)1403 static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1404 			       struct rpc_message *msg,
1405 			       const struct nfs_rpc_ops *rpc_ops,
1406 			       struct rpc_task_setup *task_setup_data, int how)
1407 {
1408 	int priority = flush_task_priority(how);
1409 
1410 	if (IS_SWAPFILE(hdr->inode))
1411 		task_setup_data->flags |= RPC_TASK_SWAPPER;
1412 	task_setup_data->priority = priority;
1413 	rpc_ops->write_setup(hdr, msg, &task_setup_data->rpc_client);
1414 	trace_nfs_initiate_write(hdr);
1415 }
1416 
1417 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1418  * call this on each, which will prepare them to be retried on next
1419  * writeback using standard nfs.
1420  */
nfs_redirty_request(struct nfs_page * req)1421 static void nfs_redirty_request(struct nfs_page *req)
1422 {
1423 	struct nfs_inode *nfsi = NFS_I(page_file_mapping(req->wb_page)->host);
1424 
1425 	/* Bump the transmission count */
1426 	req->wb_nio++;
1427 	nfs_mark_request_dirty(req);
1428 	atomic_long_inc(&nfsi->redirtied_pages);
1429 	nfs_end_page_writeback(req);
1430 	nfs_release_request(req);
1431 }
1432 
nfs_async_write_error(struct list_head * head,int error)1433 static void nfs_async_write_error(struct list_head *head, int error)
1434 {
1435 	struct nfs_page	*req;
1436 
1437 	while (!list_empty(head)) {
1438 		req = nfs_list_entry(head->next);
1439 		nfs_list_remove_request(req);
1440 		if (nfs_error_is_fatal_on_server(error))
1441 			nfs_write_error(req, error);
1442 		else
1443 			nfs_redirty_request(req);
1444 	}
1445 }
1446 
nfs_async_write_reschedule_io(struct nfs_pgio_header * hdr)1447 static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
1448 {
1449 	nfs_async_write_error(&hdr->pages, 0);
1450 }
1451 
1452 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1453 	.init_hdr = nfs_async_write_init,
1454 	.error_cleanup = nfs_async_write_error,
1455 	.completion = nfs_write_completion,
1456 	.reschedule_io = nfs_async_write_reschedule_io,
1457 };
1458 
nfs_pageio_init_write(struct nfs_pageio_descriptor * pgio,struct inode * inode,int ioflags,bool force_mds,const struct nfs_pgio_completion_ops * compl_ops)1459 void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1460 			       struct inode *inode, int ioflags, bool force_mds,
1461 			       const struct nfs_pgio_completion_ops *compl_ops)
1462 {
1463 	struct nfs_server *server = NFS_SERVER(inode);
1464 	const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1465 
1466 #ifdef CONFIG_NFS_V4_1
1467 	if (server->pnfs_curr_ld && !force_mds)
1468 		pg_ops = server->pnfs_curr_ld->pg_write_ops;
1469 #endif
1470 	nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1471 			server->wsize, ioflags);
1472 }
1473 EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1474 
nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor * pgio)1475 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1476 {
1477 	struct nfs_pgio_mirror *mirror;
1478 
1479 	if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1480 		pgio->pg_ops->pg_cleanup(pgio);
1481 
1482 	pgio->pg_ops = &nfs_pgio_rw_ops;
1483 
1484 	nfs_pageio_stop_mirroring(pgio);
1485 
1486 	mirror = &pgio->pg_mirrors[0];
1487 	mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1488 }
1489 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1490 
1491 
nfs_commit_prepare(struct rpc_task * task,void * calldata)1492 void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1493 {
1494 	struct nfs_commit_data *data = calldata;
1495 
1496 	NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1497 }
1498 
nfs_writeback_check_extend(struct nfs_pgio_header * hdr,struct nfs_fattr * fattr)1499 static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1500 		struct nfs_fattr *fattr)
1501 {
1502 	struct nfs_pgio_args *argp = &hdr->args;
1503 	struct nfs_pgio_res *resp = &hdr->res;
1504 	u64 size = argp->offset + resp->count;
1505 
1506 	if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1507 		fattr->size = size;
1508 	if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1509 		fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
1510 		return;
1511 	}
1512 	if (size != fattr->size)
1513 		return;
1514 	/* Set attribute barrier */
1515 	nfs_fattr_set_barrier(fattr);
1516 	/* ...and update size */
1517 	fattr->valid |= NFS_ATTR_FATTR_SIZE;
1518 }
1519 
nfs_writeback_update_inode(struct nfs_pgio_header * hdr)1520 void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1521 {
1522 	struct nfs_fattr *fattr = &hdr->fattr;
1523 	struct inode *inode = hdr->inode;
1524 
1525 	spin_lock(&inode->i_lock);
1526 	nfs_writeback_check_extend(hdr, fattr);
1527 	nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1528 	spin_unlock(&inode->i_lock);
1529 }
1530 EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1531 
1532 /*
1533  * This function is called when the WRITE call is complete.
1534  */
nfs_writeback_done(struct rpc_task * task,struct nfs_pgio_header * hdr,struct inode * inode)1535 static int nfs_writeback_done(struct rpc_task *task,
1536 			      struct nfs_pgio_header *hdr,
1537 			      struct inode *inode)
1538 {
1539 	int status;
1540 
1541 	/*
1542 	 * ->write_done will attempt to use post-op attributes to detect
1543 	 * conflicting writes by other clients.  A strict interpretation
1544 	 * of close-to-open would allow us to continue caching even if
1545 	 * another writer had changed the file, but some applications
1546 	 * depend on tighter cache coherency when writing.
1547 	 */
1548 	status = NFS_PROTO(inode)->write_done(task, hdr);
1549 	if (status != 0)
1550 		return status;
1551 
1552 	nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1553 	trace_nfs_writeback_done(task, hdr);
1554 
1555 	if (task->tk_status >= 0) {
1556 		enum nfs3_stable_how committed = hdr->res.verf->committed;
1557 
1558 		if (committed == NFS_UNSTABLE) {
1559 			/*
1560 			 * We have some uncommitted data on the server at
1561 			 * this point, so ensure that we keep track of that
1562 			 * fact irrespective of what later writes do.
1563 			 */
1564 			set_bit(NFS_IOHDR_UNSTABLE_WRITES, &hdr->flags);
1565 		}
1566 
1567 		if (committed < hdr->args.stable) {
1568 			/* We tried a write call, but the server did not
1569 			 * commit data to stable storage even though we
1570 			 * requested it.
1571 			 * Note: There is a known bug in Tru64 < 5.0 in which
1572 			 *	 the server reports NFS_DATA_SYNC, but performs
1573 			 *	 NFS_FILE_SYNC. We therefore implement this checking
1574 			 *	 as a dprintk() in order to avoid filling syslog.
1575 			 */
1576 			static unsigned long    complain;
1577 
1578 			/* Note this will print the MDS for a DS write */
1579 			if (time_before(complain, jiffies)) {
1580 				dprintk("NFS:       faulty NFS server %s:"
1581 					" (committed = %d) != (stable = %d)\n",
1582 					NFS_SERVER(inode)->nfs_client->cl_hostname,
1583 					committed, hdr->args.stable);
1584 				complain = jiffies + 300 * HZ;
1585 			}
1586 		}
1587 	}
1588 
1589 	/* Deal with the suid/sgid bit corner case */
1590 	if (nfs_should_remove_suid(inode)) {
1591 		spin_lock(&inode->i_lock);
1592 		nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
1593 		spin_unlock(&inode->i_lock);
1594 	}
1595 	return 0;
1596 }
1597 
1598 /*
1599  * This function is called when the WRITE call is complete.
1600  */
nfs_writeback_result(struct rpc_task * task,struct nfs_pgio_header * hdr)1601 static void nfs_writeback_result(struct rpc_task *task,
1602 				 struct nfs_pgio_header *hdr)
1603 {
1604 	struct nfs_pgio_args	*argp = &hdr->args;
1605 	struct nfs_pgio_res	*resp = &hdr->res;
1606 
1607 	if (resp->count < argp->count) {
1608 		static unsigned long    complain;
1609 
1610 		/* This a short write! */
1611 		nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1612 
1613 		/* Has the server at least made some progress? */
1614 		if (resp->count == 0) {
1615 			if (time_before(complain, jiffies)) {
1616 				printk(KERN_WARNING
1617 				       "NFS: Server wrote zero bytes, expected %u.\n",
1618 				       argp->count);
1619 				complain = jiffies + 300 * HZ;
1620 			}
1621 			nfs_set_pgio_error(hdr, -EIO, argp->offset);
1622 			task->tk_status = -EIO;
1623 			return;
1624 		}
1625 
1626 		/* For non rpc-based layout drivers, retry-through-MDS */
1627 		if (!task->tk_ops) {
1628 			hdr->pnfs_error = -EAGAIN;
1629 			return;
1630 		}
1631 
1632 		/* Was this an NFSv2 write or an NFSv3 stable write? */
1633 		if (resp->verf->committed != NFS_UNSTABLE) {
1634 			/* Resend from where the server left off */
1635 			hdr->mds_offset += resp->count;
1636 			argp->offset += resp->count;
1637 			argp->pgbase += resp->count;
1638 			argp->count -= resp->count;
1639 		} else {
1640 			/* Resend as a stable write in order to avoid
1641 			 * headaches in the case of a server crash.
1642 			 */
1643 			argp->stable = NFS_FILE_SYNC;
1644 		}
1645 		resp->count = 0;
1646 		resp->verf->committed = 0;
1647 		rpc_restart_call_prepare(task);
1648 	}
1649 }
1650 
wait_on_commit(struct nfs_mds_commit_info * cinfo)1651 static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
1652 {
1653 	return wait_var_event_killable(&cinfo->rpcs_out,
1654 				       !atomic_read(&cinfo->rpcs_out));
1655 }
1656 
nfs_commit_begin(struct nfs_mds_commit_info * cinfo)1657 static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
1658 {
1659 	atomic_inc(&cinfo->rpcs_out);
1660 }
1661 
nfs_commit_end(struct nfs_mds_commit_info * cinfo)1662 bool nfs_commit_end(struct nfs_mds_commit_info *cinfo)
1663 {
1664 	if (atomic_dec_and_test(&cinfo->rpcs_out)) {
1665 		wake_up_var(&cinfo->rpcs_out);
1666 		return true;
1667 	}
1668 	return false;
1669 }
1670 
nfs_commitdata_release(struct nfs_commit_data * data)1671 void nfs_commitdata_release(struct nfs_commit_data *data)
1672 {
1673 	put_nfs_open_context(data->context);
1674 	nfs_commit_free(data);
1675 }
1676 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1677 
nfs_initiate_commit(struct rpc_clnt * clnt,struct nfs_commit_data * data,const struct nfs_rpc_ops * nfs_ops,const struct rpc_call_ops * call_ops,int how,int flags)1678 int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1679 			const struct nfs_rpc_ops *nfs_ops,
1680 			const struct rpc_call_ops *call_ops,
1681 			int how, int flags)
1682 {
1683 	struct rpc_task *task;
1684 	int priority = flush_task_priority(how);
1685 	struct rpc_message msg = {
1686 		.rpc_argp = &data->args,
1687 		.rpc_resp = &data->res,
1688 		.rpc_cred = data->cred,
1689 	};
1690 	struct rpc_task_setup task_setup_data = {
1691 		.task = &data->task,
1692 		.rpc_client = clnt,
1693 		.rpc_message = &msg,
1694 		.callback_ops = call_ops,
1695 		.callback_data = data,
1696 		.workqueue = nfsiod_workqueue,
1697 		.flags = RPC_TASK_ASYNC | flags,
1698 		.priority = priority,
1699 	};
1700 
1701 	if (nfs_server_capable(data->inode, NFS_CAP_MOVEABLE))
1702 		task_setup_data.flags |= RPC_TASK_MOVEABLE;
1703 
1704 	/* Set up the initial task struct.  */
1705 	nfs_ops->commit_setup(data, &msg, &task_setup_data.rpc_client);
1706 	trace_nfs_initiate_commit(data);
1707 
1708 	dprintk("NFS: initiated commit call\n");
1709 
1710 	task = rpc_run_task(&task_setup_data);
1711 	if (IS_ERR(task))
1712 		return PTR_ERR(task);
1713 	if (how & FLUSH_SYNC)
1714 		rpc_wait_for_completion_task(task);
1715 	rpc_put_task(task);
1716 	return 0;
1717 }
1718 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1719 
nfs_get_lwb(struct list_head * head)1720 static loff_t nfs_get_lwb(struct list_head *head)
1721 {
1722 	loff_t lwb = 0;
1723 	struct nfs_page *req;
1724 
1725 	list_for_each_entry(req, head, wb_list)
1726 		if (lwb < (req_offset(req) + req->wb_bytes))
1727 			lwb = req_offset(req) + req->wb_bytes;
1728 
1729 	return lwb;
1730 }
1731 
1732 /*
1733  * Set up the argument/result storage required for the RPC call.
1734  */
nfs_init_commit(struct nfs_commit_data * data,struct list_head * head,struct pnfs_layout_segment * lseg,struct nfs_commit_info * cinfo)1735 void nfs_init_commit(struct nfs_commit_data *data,
1736 		     struct list_head *head,
1737 		     struct pnfs_layout_segment *lseg,
1738 		     struct nfs_commit_info *cinfo)
1739 {
1740 	struct nfs_page *first;
1741 	struct nfs_open_context *ctx;
1742 	struct inode *inode;
1743 
1744 	/* Set up the RPC argument and reply structs
1745 	 * NB: take care not to mess about with data->commit et al. */
1746 
1747 	if (head)
1748 		list_splice_init(head, &data->pages);
1749 
1750 	first = nfs_list_entry(data->pages.next);
1751 	ctx = nfs_req_openctx(first);
1752 	inode = d_inode(ctx->dentry);
1753 
1754 	data->inode	  = inode;
1755 	data->cred	  = ctx->cred;
1756 	data->lseg	  = lseg; /* reference transferred */
1757 	/* only set lwb for pnfs commit */
1758 	if (lseg)
1759 		data->lwb = nfs_get_lwb(&data->pages);
1760 	data->mds_ops     = &nfs_commit_ops;
1761 	data->completion_ops = cinfo->completion_ops;
1762 	data->dreq	  = cinfo->dreq;
1763 
1764 	data->args.fh     = NFS_FH(data->inode);
1765 	/* Note: we always request a commit of the entire inode */
1766 	data->args.offset = 0;
1767 	data->args.count  = 0;
1768 	data->context     = get_nfs_open_context(ctx);
1769 	data->res.fattr   = &data->fattr;
1770 	data->res.verf    = &data->verf;
1771 	nfs_fattr_init(&data->fattr);
1772 	nfs_commit_begin(cinfo->mds);
1773 }
1774 EXPORT_SYMBOL_GPL(nfs_init_commit);
1775 
nfs_retry_commit(struct list_head * page_list,struct pnfs_layout_segment * lseg,struct nfs_commit_info * cinfo,u32 ds_commit_idx)1776 void nfs_retry_commit(struct list_head *page_list,
1777 		      struct pnfs_layout_segment *lseg,
1778 		      struct nfs_commit_info *cinfo,
1779 		      u32 ds_commit_idx)
1780 {
1781 	struct nfs_page *req;
1782 
1783 	while (!list_empty(page_list)) {
1784 		req = nfs_list_entry(page_list->next);
1785 		nfs_list_remove_request(req);
1786 		nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
1787 		if (!cinfo->dreq)
1788 			nfs_clear_page_commit(req->wb_page);
1789 		nfs_unlock_and_release_request(req);
1790 	}
1791 }
1792 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1793 
1794 static void
nfs_commit_resched_write(struct nfs_commit_info * cinfo,struct nfs_page * req)1795 nfs_commit_resched_write(struct nfs_commit_info *cinfo,
1796 		struct nfs_page *req)
1797 {
1798 	__set_page_dirty_nobuffers(req->wb_page);
1799 }
1800 
1801 /*
1802  * Commit dirty pages
1803  */
1804 static int
nfs_commit_list(struct inode * inode,struct list_head * head,int how,struct nfs_commit_info * cinfo)1805 nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1806 		struct nfs_commit_info *cinfo)
1807 {
1808 	struct nfs_commit_data	*data;
1809 	unsigned short task_flags = 0;
1810 
1811 	/* another commit raced with us */
1812 	if (list_empty(head))
1813 		return 0;
1814 
1815 	data = nfs_commitdata_alloc();
1816 	if (!data) {
1817 		nfs_retry_commit(head, NULL, cinfo, -1);
1818 		return -ENOMEM;
1819 	}
1820 
1821 	/* Set up the argument struct */
1822 	nfs_init_commit(data, head, NULL, cinfo);
1823 	if (NFS_SERVER(inode)->nfs_client->cl_minorversion)
1824 		task_flags = RPC_TASK_MOVEABLE;
1825 	return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1826 				   data->mds_ops, how,
1827 				   RPC_TASK_CRED_NOREF | task_flags);
1828 }
1829 
1830 /*
1831  * COMMIT call returned
1832  */
nfs_commit_done(struct rpc_task * task,void * calldata)1833 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1834 {
1835 	struct nfs_commit_data	*data = calldata;
1836 
1837 	/* Call the NFS version-specific code */
1838 	NFS_PROTO(data->inode)->commit_done(task, data);
1839 	trace_nfs_commit_done(task, data);
1840 }
1841 
nfs_commit_release_pages(struct nfs_commit_data * data)1842 static void nfs_commit_release_pages(struct nfs_commit_data *data)
1843 {
1844 	const struct nfs_writeverf *verf = data->res.verf;
1845 	struct nfs_page	*req;
1846 	int status = data->task.tk_status;
1847 	struct nfs_commit_info cinfo;
1848 	struct nfs_server *nfss;
1849 
1850 	while (!list_empty(&data->pages)) {
1851 		req = nfs_list_entry(data->pages.next);
1852 		nfs_list_remove_request(req);
1853 		if (req->wb_page)
1854 			nfs_clear_page_commit(req->wb_page);
1855 
1856 		dprintk("NFS:       commit (%s/%llu %d@%lld)",
1857 			nfs_req_openctx(req)->dentry->d_sb->s_id,
1858 			(unsigned long long)NFS_FILEID(d_inode(nfs_req_openctx(req)->dentry)),
1859 			req->wb_bytes,
1860 			(long long)req_offset(req));
1861 		if (status < 0) {
1862 			if (req->wb_page) {
1863 				trace_nfs_commit_error(data->inode, req,
1864 						       status);
1865 				nfs_mapping_set_error(req->wb_page, status);
1866 				nfs_inode_remove_request(req);
1867 			}
1868 			dprintk_cont(", error = %d\n", status);
1869 			goto next;
1870 		}
1871 
1872 		/* Okay, COMMIT succeeded, apparently. Check the verifier
1873 		 * returned by the server against all stored verfs. */
1874 		if (nfs_write_match_verf(verf, req)) {
1875 			/* We have a match */
1876 			if (req->wb_page)
1877 				nfs_inode_remove_request(req);
1878 			dprintk_cont(" OK\n");
1879 			goto next;
1880 		}
1881 		/* We have a mismatch. Write the page again */
1882 		dprintk_cont(" mismatch\n");
1883 		nfs_mark_request_dirty(req);
1884 		atomic_long_inc(&NFS_I(data->inode)->redirtied_pages);
1885 	next:
1886 		nfs_unlock_and_release_request(req);
1887 		/* Latency breaker */
1888 		cond_resched();
1889 	}
1890 	nfss = NFS_SERVER(data->inode);
1891 	if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1892 		nfss->write_congested = 0;
1893 
1894 	nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1895 	nfs_commit_end(cinfo.mds);
1896 }
1897 
nfs_commit_release(void * calldata)1898 static void nfs_commit_release(void *calldata)
1899 {
1900 	struct nfs_commit_data *data = calldata;
1901 
1902 	data->completion_ops->completion(data);
1903 	nfs_commitdata_release(calldata);
1904 }
1905 
1906 static const struct rpc_call_ops nfs_commit_ops = {
1907 	.rpc_call_prepare = nfs_commit_prepare,
1908 	.rpc_call_done = nfs_commit_done,
1909 	.rpc_release = nfs_commit_release,
1910 };
1911 
1912 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1913 	.completion = nfs_commit_release_pages,
1914 	.resched_write = nfs_commit_resched_write,
1915 };
1916 
nfs_generic_commit_list(struct inode * inode,struct list_head * head,int how,struct nfs_commit_info * cinfo)1917 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1918 			    int how, struct nfs_commit_info *cinfo)
1919 {
1920 	int status;
1921 
1922 	status = pnfs_commit_list(inode, head, how, cinfo);
1923 	if (status == PNFS_NOT_ATTEMPTED)
1924 		status = nfs_commit_list(inode, head, how, cinfo);
1925 	return status;
1926 }
1927 
__nfs_commit_inode(struct inode * inode,int how,struct writeback_control * wbc)1928 static int __nfs_commit_inode(struct inode *inode, int how,
1929 		struct writeback_control *wbc)
1930 {
1931 	LIST_HEAD(head);
1932 	struct nfs_commit_info cinfo;
1933 	int may_wait = how & FLUSH_SYNC;
1934 	int ret, nscan;
1935 
1936 	how &= ~FLUSH_SYNC;
1937 	nfs_init_cinfo_from_inode(&cinfo, inode);
1938 	nfs_commit_begin(cinfo.mds);
1939 	for (;;) {
1940 		ret = nscan = nfs_scan_commit(inode, &head, &cinfo);
1941 		if (ret <= 0)
1942 			break;
1943 		ret = nfs_generic_commit_list(inode, &head, how, &cinfo);
1944 		if (ret < 0)
1945 			break;
1946 		ret = 0;
1947 		if (wbc && wbc->sync_mode == WB_SYNC_NONE) {
1948 			if (nscan < wbc->nr_to_write)
1949 				wbc->nr_to_write -= nscan;
1950 			else
1951 				wbc->nr_to_write = 0;
1952 		}
1953 		if (nscan < INT_MAX)
1954 			break;
1955 		cond_resched();
1956 	}
1957 	nfs_commit_end(cinfo.mds);
1958 	if (ret || !may_wait)
1959 		return ret;
1960 	return wait_on_commit(cinfo.mds);
1961 }
1962 
nfs_commit_inode(struct inode * inode,int how)1963 int nfs_commit_inode(struct inode *inode, int how)
1964 {
1965 	return __nfs_commit_inode(inode, how, NULL);
1966 }
1967 EXPORT_SYMBOL_GPL(nfs_commit_inode);
1968 
nfs_write_inode(struct inode * inode,struct writeback_control * wbc)1969 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1970 {
1971 	struct nfs_inode *nfsi = NFS_I(inode);
1972 	int flags = FLUSH_SYNC;
1973 	int ret = 0;
1974 
1975 	if (wbc->sync_mode == WB_SYNC_NONE) {
1976 		/* no commits means nothing needs to be done */
1977 		if (!atomic_long_read(&nfsi->commit_info.ncommit))
1978 			goto check_requests_outstanding;
1979 
1980 		/* Don't commit yet if this is a non-blocking flush and there
1981 		 * are a lot of outstanding writes for this mapping.
1982 		 */
1983 		if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1984 			goto out_mark_dirty;
1985 
1986 		/* don't wait for the COMMIT response */
1987 		flags = 0;
1988 	}
1989 
1990 	ret = __nfs_commit_inode(inode, flags, wbc);
1991 	if (!ret) {
1992 		if (flags & FLUSH_SYNC)
1993 			return 0;
1994 	} else if (atomic_long_read(&nfsi->commit_info.ncommit))
1995 		goto out_mark_dirty;
1996 
1997 check_requests_outstanding:
1998 	if (!atomic_read(&nfsi->commit_info.rpcs_out))
1999 		return ret;
2000 out_mark_dirty:
2001 	__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
2002 	return ret;
2003 }
2004 EXPORT_SYMBOL_GPL(nfs_write_inode);
2005 
2006 /*
2007  * Wrapper for filemap_write_and_wait_range()
2008  *
2009  * Needed for pNFS in order to ensure data becomes visible to the
2010  * client.
2011  */
nfs_filemap_write_and_wait_range(struct address_space * mapping,loff_t lstart,loff_t lend)2012 int nfs_filemap_write_and_wait_range(struct address_space *mapping,
2013 		loff_t lstart, loff_t lend)
2014 {
2015 	int ret;
2016 
2017 	ret = filemap_write_and_wait_range(mapping, lstart, lend);
2018 	if (ret == 0)
2019 		ret = pnfs_sync_inode(mapping->host, true);
2020 	return ret;
2021 }
2022 EXPORT_SYMBOL_GPL(nfs_filemap_write_and_wait_range);
2023 
2024 /*
2025  * flush the inode to disk.
2026  */
nfs_wb_all(struct inode * inode)2027 int nfs_wb_all(struct inode *inode)
2028 {
2029 	int ret;
2030 
2031 	trace_nfs_writeback_inode_enter(inode);
2032 
2033 	ret = filemap_write_and_wait(inode->i_mapping);
2034 	if (ret)
2035 		goto out;
2036 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
2037 	if (ret < 0)
2038 		goto out;
2039 	pnfs_sync_inode(inode, true);
2040 	ret = 0;
2041 
2042 out:
2043 	trace_nfs_writeback_inode_exit(inode, ret);
2044 	return ret;
2045 }
2046 EXPORT_SYMBOL_GPL(nfs_wb_all);
2047 
nfs_wb_folio_cancel(struct inode * inode,struct folio * folio)2048 int nfs_wb_folio_cancel(struct inode *inode, struct folio *folio)
2049 {
2050 	struct nfs_page *req;
2051 	int ret = 0;
2052 
2053 	folio_wait_writeback(folio);
2054 
2055 	/* blocking call to cancel all requests and join to a single (head)
2056 	 * request */
2057 	req = nfs_lock_and_join_requests(&folio->page);
2058 
2059 	if (IS_ERR(req)) {
2060 		ret = PTR_ERR(req);
2061 	} else if (req) {
2062 		/* all requests from this folio have been cancelled by
2063 		 * nfs_lock_and_join_requests, so just remove the head
2064 		 * request from the inode / page_private pointer and
2065 		 * release it */
2066 		nfs_inode_remove_request(req);
2067 		nfs_unlock_and_release_request(req);
2068 	}
2069 
2070 	return ret;
2071 }
2072 
2073 /*
2074  * Write back all requests on one page - we do this before reading it.
2075  */
nfs_wb_page(struct inode * inode,struct page * page)2076 int nfs_wb_page(struct inode *inode, struct page *page)
2077 {
2078 	loff_t range_start = page_file_offset(page);
2079 	loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
2080 	struct writeback_control wbc = {
2081 		.sync_mode = WB_SYNC_ALL,
2082 		.nr_to_write = 0,
2083 		.range_start = range_start,
2084 		.range_end = range_end,
2085 	};
2086 	int ret;
2087 
2088 	trace_nfs_writeback_page_enter(inode);
2089 
2090 	for (;;) {
2091 		wait_on_page_writeback(page);
2092 		if (clear_page_dirty_for_io(page)) {
2093 			ret = nfs_writepage_locked(page, &wbc);
2094 			if (ret < 0)
2095 				goto out_error;
2096 			continue;
2097 		}
2098 		ret = 0;
2099 		if (!PagePrivate(page))
2100 			break;
2101 		ret = nfs_commit_inode(inode, FLUSH_SYNC);
2102 		if (ret < 0)
2103 			goto out_error;
2104 	}
2105 out_error:
2106 	trace_nfs_writeback_page_exit(inode, ret);
2107 	return ret;
2108 }
2109 
2110 #ifdef CONFIG_MIGRATION
nfs_migrate_folio(struct address_space * mapping,struct folio * dst,struct folio * src,enum migrate_mode mode)2111 int nfs_migrate_folio(struct address_space *mapping, struct folio *dst,
2112 		struct folio *src, enum migrate_mode mode)
2113 {
2114 	/*
2115 	 * If the private flag is set, the folio is currently associated with
2116 	 * an in-progress read or write request. Don't try to migrate it.
2117 	 *
2118 	 * FIXME: we could do this in principle, but we'll need a way to ensure
2119 	 *        that we can safely release the inode reference while holding
2120 	 *        the folio lock.
2121 	 */
2122 	if (folio_test_private(src))
2123 		return -EBUSY;
2124 
2125 	if (folio_test_fscache(src)) {
2126 		if (mode == MIGRATE_ASYNC)
2127 			return -EBUSY;
2128 		folio_wait_fscache(src);
2129 	}
2130 
2131 	return migrate_folio(mapping, dst, src, mode);
2132 }
2133 #endif
2134 
nfs_init_writepagecache(void)2135 int __init nfs_init_writepagecache(void)
2136 {
2137 	nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
2138 					     sizeof(struct nfs_pgio_header),
2139 					     0, SLAB_HWCACHE_ALIGN,
2140 					     NULL);
2141 	if (nfs_wdata_cachep == NULL)
2142 		return -ENOMEM;
2143 
2144 	nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2145 						     nfs_wdata_cachep);
2146 	if (nfs_wdata_mempool == NULL)
2147 		goto out_destroy_write_cache;
2148 
2149 	nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2150 					     sizeof(struct nfs_commit_data),
2151 					     0, SLAB_HWCACHE_ALIGN,
2152 					     NULL);
2153 	if (nfs_cdata_cachep == NULL)
2154 		goto out_destroy_write_mempool;
2155 
2156 	nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
2157 						      nfs_cdata_cachep);
2158 	if (nfs_commit_mempool == NULL)
2159 		goto out_destroy_commit_cache;
2160 
2161 	/*
2162 	 * NFS congestion size, scale with available memory.
2163 	 *
2164 	 *  64MB:    8192k
2165 	 * 128MB:   11585k
2166 	 * 256MB:   16384k
2167 	 * 512MB:   23170k
2168 	 *   1GB:   32768k
2169 	 *   2GB:   46340k
2170 	 *   4GB:   65536k
2171 	 *   8GB:   92681k
2172 	 *  16GB:  131072k
2173 	 *
2174 	 * This allows larger machines to have larger/more transfers.
2175 	 * Limit the default to 256M
2176 	 */
2177 	nfs_congestion_kb = (16*int_sqrt(totalram_pages())) << (PAGE_SHIFT-10);
2178 	if (nfs_congestion_kb > 256*1024)
2179 		nfs_congestion_kb = 256*1024;
2180 
2181 	return 0;
2182 
2183 out_destroy_commit_cache:
2184 	kmem_cache_destroy(nfs_cdata_cachep);
2185 out_destroy_write_mempool:
2186 	mempool_destroy(nfs_wdata_mempool);
2187 out_destroy_write_cache:
2188 	kmem_cache_destroy(nfs_wdata_cachep);
2189 	return -ENOMEM;
2190 }
2191 
nfs_destroy_writepagecache(void)2192 void nfs_destroy_writepagecache(void)
2193 {
2194 	mempool_destroy(nfs_commit_mempool);
2195 	kmem_cache_destroy(nfs_cdata_cachep);
2196 	mempool_destroy(nfs_wdata_mempool);
2197 	kmem_cache_destroy(nfs_wdata_cachep);
2198 }
2199 
2200 static const struct nfs_rw_ops nfs_rw_write_ops = {
2201 	.rw_alloc_header	= nfs_writehdr_alloc,
2202 	.rw_free_header		= nfs_writehdr_free,
2203 	.rw_done		= nfs_writeback_done,
2204 	.rw_result		= nfs_writeback_result,
2205 	.rw_initiate		= nfs_initiate_write,
2206 };
2207