xref: /linux/fs/nfs/file.c (revision d4821902e43453b85b31329441a9f6ac071228a8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/file.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  Changes Copyright (C) 1994 by Florian La Roche
8  *   - Do not copy data too often around in the kernel.
9  *   - In nfs_file_read the return value of kmalloc wasn't checked.
10  *   - Put in a better version of read look-ahead buffering. Original idea
11  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
12  *
13  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
14  *
15  *  Total rewrite of read side for new NFS buffer cache.. Linus.
16  *
17  *  nfs regular file handling functions
18  */
19 
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/stat.h>
26 #include <linux/nfs_fs.h>
27 #include <linux/nfs_mount.h>
28 #include <linux/mm.h>
29 #include <linux/pagemap.h>
30 #include <linux/gfp.h>
31 #include <linux/swap.h>
32 
33 #include <linux/uaccess.h>
34 
35 #include "delegation.h"
36 #include "internal.h"
37 #include "iostat.h"
38 #include "fscache.h"
39 #include "pnfs.h"
40 
41 #include "nfstrace.h"
42 
43 #define NFSDBG_FACILITY		NFSDBG_FILE
44 
45 static const struct vm_operations_struct nfs_file_vm_ops;
46 
47 int nfs_check_flags(int flags)
48 {
49 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
50 		return -EINVAL;
51 
52 	return 0;
53 }
54 EXPORT_SYMBOL_GPL(nfs_check_flags);
55 
56 /*
57  * Open file
58  */
59 static int
60 nfs_file_open(struct inode *inode, struct file *filp)
61 {
62 	int res;
63 
64 	dprintk("NFS: open file(%pD2)\n", filp);
65 
66 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
67 	res = nfs_check_flags(filp->f_flags);
68 	if (res)
69 		return res;
70 
71 	res = nfs_open(inode, filp);
72 	if (res == 0)
73 		filp->f_mode |= FMODE_CAN_ODIRECT;
74 	return res;
75 }
76 
77 int
78 nfs_file_release(struct inode *inode, struct file *filp)
79 {
80 	dprintk("NFS: release(%pD2)\n", filp);
81 
82 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
83 	nfs_file_clear_open_context(filp);
84 	nfs_fscache_release_file(inode, filp);
85 	return 0;
86 }
87 EXPORT_SYMBOL_GPL(nfs_file_release);
88 
89 /**
90  * nfs_revalidate_file_size - Revalidate the file size
91  * @inode: pointer to inode struct
92  * @filp: pointer to struct file
93  *
94  * Revalidates the file length. This is basically a wrapper around
95  * nfs_revalidate_inode() that takes into account the fact that we may
96  * have cached writes (in which case we don't care about the server's
97  * idea of what the file length is), or O_DIRECT (in which case we
98  * shouldn't trust the cache).
99  */
100 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101 {
102 	struct nfs_server *server = NFS_SERVER(inode);
103 
104 	if (filp->f_flags & O_DIRECT)
105 		goto force_reval;
106 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
107 		goto force_reval;
108 	return 0;
109 force_reval:
110 	return __nfs_revalidate_inode(server, inode);
111 }
112 
113 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
114 {
115 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
116 			filp, offset, whence);
117 
118 	/*
119 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
120 	 * the cached file length
121 	 */
122 	if (whence != SEEK_SET && whence != SEEK_CUR) {
123 		struct inode *inode = filp->f_mapping->host;
124 
125 		int retval = nfs_revalidate_file_size(inode, filp);
126 		if (retval < 0)
127 			return (loff_t)retval;
128 	}
129 
130 	return generic_file_llseek(filp, offset, whence);
131 }
132 EXPORT_SYMBOL_GPL(nfs_file_llseek);
133 
134 /*
135  * Flush all dirty pages, and check for write errors.
136  */
137 static int
138 nfs_file_flush(struct file *file, fl_owner_t id)
139 {
140 	struct inode	*inode = file_inode(file);
141 	errseq_t since;
142 
143 	dprintk("NFS: flush(%pD2)\n", file);
144 
145 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
146 	if ((file->f_mode & FMODE_WRITE) == 0)
147 		return 0;
148 
149 	/* Flush writes to the server and return any errors */
150 	since = filemap_sample_wb_err(file->f_mapping);
151 	nfs_wb_all(inode);
152 	return filemap_check_wb_err(file->f_mapping, since);
153 }
154 
155 ssize_t
156 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
157 {
158 	struct inode *inode = file_inode(iocb->ki_filp);
159 	ssize_t result;
160 
161 	if (iocb->ki_flags & IOCB_DIRECT)
162 		return nfs_file_direct_read(iocb, to, false);
163 
164 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
165 		iocb->ki_filp,
166 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
167 
168 	nfs_start_io_read(inode);
169 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
170 	if (!result) {
171 		result = generic_file_read_iter(iocb, to);
172 		if (result > 0)
173 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
174 	}
175 	nfs_end_io_read(inode);
176 	return result;
177 }
178 EXPORT_SYMBOL_GPL(nfs_file_read);
179 
180 int
181 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
182 {
183 	struct inode *inode = file_inode(file);
184 	int	status;
185 
186 	dprintk("NFS: mmap(%pD2)\n", file);
187 
188 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
189 	 *       so we call that before revalidating the mapping
190 	 */
191 	status = generic_file_mmap(file, vma);
192 	if (!status) {
193 		vma->vm_ops = &nfs_file_vm_ops;
194 		status = nfs_revalidate_mapping(inode, file->f_mapping);
195 	}
196 	return status;
197 }
198 EXPORT_SYMBOL_GPL(nfs_file_mmap);
199 
200 /*
201  * Flush any dirty pages for this process, and check for write errors.
202  * The return status from this call provides a reliable indication of
203  * whether any write errors occurred for this process.
204  */
205 static int
206 nfs_file_fsync_commit(struct file *file, int datasync)
207 {
208 	struct inode *inode = file_inode(file);
209 	int ret;
210 
211 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
212 
213 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
214 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
215 	if (ret < 0)
216 		return ret;
217 	return file_check_and_advance_wb_err(file);
218 }
219 
220 int
221 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
222 {
223 	struct nfs_open_context *ctx = nfs_file_open_context(file);
224 	struct inode *inode = file_inode(file);
225 	int ret;
226 
227 	trace_nfs_fsync_enter(inode);
228 
229 	for (;;) {
230 		ret = file_write_and_wait_range(file, start, end);
231 		if (ret != 0)
232 			break;
233 		ret = nfs_file_fsync_commit(file, datasync);
234 		if (ret != 0)
235 			break;
236 		ret = pnfs_sync_inode(inode, !!datasync);
237 		if (ret != 0)
238 			break;
239 		if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
240 			break;
241 		/*
242 		 * If nfs_file_fsync_commit detected a server reboot, then
243 		 * resend all dirty pages that might have been covered by
244 		 * the NFS_CONTEXT_RESEND_WRITES flag
245 		 */
246 		start = 0;
247 		end = LLONG_MAX;
248 	}
249 
250 	trace_nfs_fsync_exit(inode, ret);
251 	return ret;
252 }
253 EXPORT_SYMBOL_GPL(nfs_file_fsync);
254 
255 /*
256  * Decide whether a read/modify/write cycle may be more efficient
257  * then a modify/write/read cycle when writing to a page in the
258  * page cache.
259  *
260  * Some pNFS layout drivers can only read/write at a certain block
261  * granularity like all block devices and therefore we must perform
262  * read/modify/write whenever a page hasn't read yet and the data
263  * to be written there is not aligned to a block boundary and/or
264  * smaller than the block size.
265  *
266  * The modify/write/read cycle may occur if a page is read before
267  * being completely filled by the writer.  In this situation, the
268  * page must be completely written to stable storage on the server
269  * before it can be refilled by reading in the page from the server.
270  * This can lead to expensive, small, FILE_SYNC mode writes being
271  * done.
272  *
273  * It may be more efficient to read the page first if the file is
274  * open for reading in addition to writing, the page is not marked
275  * as Uptodate, it is not dirty or waiting to be committed,
276  * indicating that it was previously allocated and then modified,
277  * that there were valid bytes of data in that range of the file,
278  * and that the new data won't completely replace the old data in
279  * that range of the file.
280  */
281 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
282 {
283 	unsigned int pglen = nfs_page_length(page);
284 	unsigned int offset = pos & (PAGE_SIZE - 1);
285 	unsigned int end = offset + len;
286 
287 	return !pglen || (end >= pglen && !offset);
288 }
289 
290 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
291 			loff_t pos, unsigned int len)
292 {
293 	/*
294 	 * Up-to-date pages, those with ongoing or full-page write
295 	 * don't need read/modify/write
296 	 */
297 	if (PageUptodate(page) || PagePrivate(page) ||
298 	    nfs_full_page_write(page, pos, len))
299 		return false;
300 
301 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
302 		return true;
303 	/* Open for reading too? */
304 	if (file->f_mode & FMODE_READ)
305 		return true;
306 	return false;
307 }
308 
309 /*
310  * This does the "real" work of the write. We must allocate and lock the
311  * page to be sent back to the generic routine, which then copies the
312  * data from user space.
313  *
314  * If the writer ends up delaying the write, the writer needs to
315  * increment the page use counts until he is done with the page.
316  */
317 static int nfs_write_begin(struct file *file, struct address_space *mapping,
318 			loff_t pos, unsigned len,
319 			struct page **pagep, void **fsdata)
320 {
321 	int ret;
322 	pgoff_t index = pos >> PAGE_SHIFT;
323 	struct page *page;
324 	int once_thru = 0;
325 
326 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
327 		file, mapping->host->i_ino, len, (long long) pos);
328 
329 start:
330 	page = grab_cache_page_write_begin(mapping, index);
331 	if (!page)
332 		return -ENOMEM;
333 	*pagep = page;
334 
335 	ret = nfs_flush_incompatible(file, page);
336 	if (ret) {
337 		unlock_page(page);
338 		put_page(page);
339 	} else if (!once_thru &&
340 		   nfs_want_read_modify_write(file, page, pos, len)) {
341 		once_thru = 1;
342 		ret = nfs_read_folio(file, page_folio(page));
343 		put_page(page);
344 		if (!ret)
345 			goto start;
346 	}
347 	return ret;
348 }
349 
350 static int nfs_write_end(struct file *file, struct address_space *mapping,
351 			loff_t pos, unsigned len, unsigned copied,
352 			struct page *page, void *fsdata)
353 {
354 	unsigned offset = pos & (PAGE_SIZE - 1);
355 	struct nfs_open_context *ctx = nfs_file_open_context(file);
356 	int status;
357 
358 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
359 		file, mapping->host->i_ino, len, (long long) pos);
360 
361 	/*
362 	 * Zero any uninitialised parts of the page, and then mark the page
363 	 * as up to date if it turns out that we're extending the file.
364 	 */
365 	if (!PageUptodate(page)) {
366 		unsigned pglen = nfs_page_length(page);
367 		unsigned end = offset + copied;
368 
369 		if (pglen == 0) {
370 			zero_user_segments(page, 0, offset,
371 					end, PAGE_SIZE);
372 			SetPageUptodate(page);
373 		} else if (end >= pglen) {
374 			zero_user_segment(page, end, PAGE_SIZE);
375 			if (offset == 0)
376 				SetPageUptodate(page);
377 		} else
378 			zero_user_segment(page, pglen, PAGE_SIZE);
379 	}
380 
381 	status = nfs_updatepage(file, page, offset, copied);
382 
383 	unlock_page(page);
384 	put_page(page);
385 
386 	if (status < 0)
387 		return status;
388 	NFS_I(mapping->host)->write_io += copied;
389 
390 	if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
391 		status = nfs_wb_all(mapping->host);
392 		if (status < 0)
393 			return status;
394 	}
395 
396 	return copied;
397 }
398 
399 /*
400  * Partially or wholly invalidate a page
401  * - Release the private state associated with a page if undergoing complete
402  *   page invalidation
403  * - Called if either PG_private or PG_fscache is set on the page
404  * - Caller holds page lock
405  */
406 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
407 				size_t length)
408 {
409 	dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
410 		 folio->index, offset, length);
411 
412 	if (offset != 0 || length < folio_size(folio))
413 		return;
414 	/* Cancel any unstarted writes on this page */
415 	nfs_wb_folio_cancel(folio->mapping->host, folio);
416 	folio_wait_fscache(folio);
417 }
418 
419 /*
420  * Attempt to release the private state associated with a folio
421  * - Called if either private or fscache flags are set on the folio
422  * - Caller holds folio lock
423  * - Return true (may release folio) or false (may not)
424  */
425 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
426 {
427 	dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
428 
429 	/* If the private flag is set, then the folio is not freeable */
430 	if (folio_test_private(folio))
431 		return false;
432 	return nfs_fscache_release_folio(folio, gfp);
433 }
434 
435 static void nfs_check_dirty_writeback(struct folio *folio,
436 				bool *dirty, bool *writeback)
437 {
438 	struct nfs_inode *nfsi;
439 	struct address_space *mapping = folio->mapping;
440 
441 	/*
442 	 * Check if an unstable folio is currently being committed and
443 	 * if so, have the VM treat it as if the folio is under writeback
444 	 * so it will not block due to folios that will shortly be freeable.
445 	 */
446 	nfsi = NFS_I(mapping->host);
447 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
448 		*writeback = true;
449 		return;
450 	}
451 
452 	/*
453 	 * If the private flag is set, then the folio is not freeable
454 	 * and as the inode is not being committed, it's not going to
455 	 * be cleaned in the near future so treat it as dirty
456 	 */
457 	if (folio_test_private(folio))
458 		*dirty = true;
459 }
460 
461 /*
462  * Attempt to clear the private state associated with a page when an error
463  * occurs that requires the cached contents of an inode to be written back or
464  * destroyed
465  * - Called if either PG_private or fscache is set on the page
466  * - Caller holds page lock
467  * - Return 0 if successful, -error otherwise
468  */
469 static int nfs_launder_folio(struct folio *folio)
470 {
471 	struct inode *inode = folio->mapping->host;
472 
473 	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
474 		inode->i_ino, folio_pos(folio));
475 
476 	folio_wait_fscache(folio);
477 	return nfs_wb_page(inode, &folio->page);
478 }
479 
480 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
481 						sector_t *span)
482 {
483 	unsigned long blocks;
484 	long long isize;
485 	int ret;
486 	struct inode *inode = file_inode(file);
487 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
488 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
489 
490 	spin_lock(&inode->i_lock);
491 	blocks = inode->i_blocks;
492 	isize = inode->i_size;
493 	spin_unlock(&inode->i_lock);
494 	if (blocks*512 < isize) {
495 		pr_warn("swap activate: swapfile has holes\n");
496 		return -EINVAL;
497 	}
498 
499 	ret = rpc_clnt_swap_activate(clnt);
500 	if (ret)
501 		return ret;
502 	ret = add_swap_extent(sis, 0, sis->max, 0);
503 	if (ret < 0) {
504 		rpc_clnt_swap_deactivate(clnt);
505 		return ret;
506 	}
507 
508 	*span = sis->pages;
509 
510 	if (cl->rpc_ops->enable_swap)
511 		cl->rpc_ops->enable_swap(inode);
512 
513 	sis->flags |= SWP_FS_OPS;
514 	return ret;
515 }
516 
517 static void nfs_swap_deactivate(struct file *file)
518 {
519 	struct inode *inode = file_inode(file);
520 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
521 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
522 
523 	rpc_clnt_swap_deactivate(clnt);
524 	if (cl->rpc_ops->disable_swap)
525 		cl->rpc_ops->disable_swap(file_inode(file));
526 }
527 
528 const struct address_space_operations nfs_file_aops = {
529 	.read_folio = nfs_read_folio,
530 	.readahead = nfs_readahead,
531 	.dirty_folio = filemap_dirty_folio,
532 	.writepage = nfs_writepage,
533 	.writepages = nfs_writepages,
534 	.write_begin = nfs_write_begin,
535 	.write_end = nfs_write_end,
536 	.invalidate_folio = nfs_invalidate_folio,
537 	.release_folio = nfs_release_folio,
538 #ifdef CONFIG_MIGRATION
539 	.migratepage = nfs_migrate_page,
540 #endif
541 	.launder_folio = nfs_launder_folio,
542 	.is_dirty_writeback = nfs_check_dirty_writeback,
543 	.error_remove_page = generic_error_remove_page,
544 	.swap_activate = nfs_swap_activate,
545 	.swap_deactivate = nfs_swap_deactivate,
546 	.swap_rw = nfs_swap_rw,
547 };
548 
549 /*
550  * Notification that a PTE pointing to an NFS page is about to be made
551  * writable, implying that someone is about to modify the page through a
552  * shared-writable mapping
553  */
554 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
555 {
556 	struct page *page = vmf->page;
557 	struct file *filp = vmf->vma->vm_file;
558 	struct inode *inode = file_inode(filp);
559 	unsigned pagelen;
560 	vm_fault_t ret = VM_FAULT_NOPAGE;
561 	struct address_space *mapping;
562 
563 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
564 		filp, filp->f_mapping->host->i_ino,
565 		(long long)page_offset(page));
566 
567 	sb_start_pagefault(inode->i_sb);
568 
569 	/* make sure the cache has finished storing the page */
570 	if (PageFsCache(page) &&
571 	    wait_on_page_fscache_killable(vmf->page) < 0) {
572 		ret = VM_FAULT_RETRY;
573 		goto out;
574 	}
575 
576 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
577 			nfs_wait_bit_killable, TASK_KILLABLE);
578 
579 	lock_page(page);
580 	mapping = page_file_mapping(page);
581 	if (mapping != inode->i_mapping)
582 		goto out_unlock;
583 
584 	wait_on_page_writeback(page);
585 
586 	pagelen = nfs_page_length(page);
587 	if (pagelen == 0)
588 		goto out_unlock;
589 
590 	ret = VM_FAULT_LOCKED;
591 	if (nfs_flush_incompatible(filp, page) == 0 &&
592 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
593 		goto out;
594 
595 	ret = VM_FAULT_SIGBUS;
596 out_unlock:
597 	unlock_page(page);
598 out:
599 	sb_end_pagefault(inode->i_sb);
600 	return ret;
601 }
602 
603 static const struct vm_operations_struct nfs_file_vm_ops = {
604 	.fault = filemap_fault,
605 	.map_pages = filemap_map_pages,
606 	.page_mkwrite = nfs_vm_page_mkwrite,
607 };
608 
609 static int nfs_need_check_write(struct file *filp, struct inode *inode,
610 				int error)
611 {
612 	struct nfs_open_context *ctx;
613 
614 	ctx = nfs_file_open_context(filp);
615 	if (nfs_error_is_fatal_on_server(error) ||
616 	    nfs_ctx_key_to_expire(ctx, inode))
617 		return 1;
618 	return 0;
619 }
620 
621 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
622 {
623 	struct file *file = iocb->ki_filp;
624 	struct inode *inode = file_inode(file);
625 	unsigned int mntflags = NFS_SERVER(inode)->flags;
626 	ssize_t result, written;
627 	errseq_t since;
628 	int error;
629 
630 	result = nfs_key_timeout_notify(file, inode);
631 	if (result)
632 		return result;
633 
634 	if (iocb->ki_flags & IOCB_DIRECT)
635 		return nfs_file_direct_write(iocb, from, false);
636 
637 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
638 		file, iov_iter_count(from), (long long) iocb->ki_pos);
639 
640 	if (IS_SWAPFILE(inode))
641 		goto out_swapfile;
642 	/*
643 	 * O_APPEND implies that we must revalidate the file length.
644 	 */
645 	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
646 		result = nfs_revalidate_file_size(inode, file);
647 		if (result)
648 			goto out;
649 	}
650 
651 	nfs_clear_invalid_mapping(file->f_mapping);
652 
653 	since = filemap_sample_wb_err(file->f_mapping);
654 	nfs_start_io_write(inode);
655 	result = generic_write_checks(iocb, from);
656 	if (result > 0) {
657 		current->backing_dev_info = inode_to_bdi(inode);
658 		result = generic_perform_write(iocb, from);
659 		current->backing_dev_info = NULL;
660 	}
661 	nfs_end_io_write(inode);
662 	if (result <= 0)
663 		goto out;
664 
665 	written = result;
666 	iocb->ki_pos += written;
667 
668 	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
669 		result = filemap_fdatawrite_range(file->f_mapping,
670 						  iocb->ki_pos - written,
671 						  iocb->ki_pos - 1);
672 		if (result < 0)
673 			goto out;
674 	}
675 	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
676 		result = filemap_fdatawait_range(file->f_mapping,
677 						 iocb->ki_pos - written,
678 						 iocb->ki_pos - 1);
679 		if (result < 0)
680 			goto out;
681 	}
682 	result = generic_write_sync(iocb, written);
683 	if (result < 0)
684 		goto out;
685 
686 	/* Return error values */
687 	error = filemap_check_wb_err(file->f_mapping, since);
688 	if (nfs_need_check_write(file, inode, error)) {
689 		int err = nfs_wb_all(inode);
690 		if (err < 0)
691 			result = err;
692 	}
693 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
694 out:
695 	return result;
696 
697 out_swapfile:
698 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
699 	return -ETXTBSY;
700 }
701 EXPORT_SYMBOL_GPL(nfs_file_write);
702 
703 static int
704 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
705 {
706 	struct inode *inode = filp->f_mapping->host;
707 	int status = 0;
708 	unsigned int saved_type = fl->fl_type;
709 
710 	/* Try local locking first */
711 	posix_test_lock(filp, fl);
712 	if (fl->fl_type != F_UNLCK) {
713 		/* found a conflict */
714 		goto out;
715 	}
716 	fl->fl_type = saved_type;
717 
718 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
719 		goto out_noconflict;
720 
721 	if (is_local)
722 		goto out_noconflict;
723 
724 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
725 out:
726 	return status;
727 out_noconflict:
728 	fl->fl_type = F_UNLCK;
729 	goto out;
730 }
731 
732 static int
733 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
734 {
735 	struct inode *inode = filp->f_mapping->host;
736 	struct nfs_lock_context *l_ctx;
737 	int status;
738 
739 	/*
740 	 * Flush all pending writes before doing anything
741 	 * with locks..
742 	 */
743 	nfs_wb_all(inode);
744 
745 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
746 	if (!IS_ERR(l_ctx)) {
747 		status = nfs_iocounter_wait(l_ctx);
748 		nfs_put_lock_context(l_ctx);
749 		/*  NOTE: special case
750 		 * 	If we're signalled while cleaning up locks on process exit, we
751 		 * 	still need to complete the unlock.
752 		 */
753 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
754 			return status;
755 	}
756 
757 	/*
758 	 * Use local locking if mounted with "-onolock" or with appropriate
759 	 * "-olocal_lock="
760 	 */
761 	if (!is_local)
762 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
763 	else
764 		status = locks_lock_file_wait(filp, fl);
765 	return status;
766 }
767 
768 static int
769 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
770 {
771 	struct inode *inode = filp->f_mapping->host;
772 	int status;
773 
774 	/*
775 	 * Flush all pending writes before doing anything
776 	 * with locks..
777 	 */
778 	status = nfs_sync_mapping(filp->f_mapping);
779 	if (status != 0)
780 		goto out;
781 
782 	/*
783 	 * Use local locking if mounted with "-onolock" or with appropriate
784 	 * "-olocal_lock="
785 	 */
786 	if (!is_local)
787 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
788 	else
789 		status = locks_lock_file_wait(filp, fl);
790 	if (status < 0)
791 		goto out;
792 
793 	/*
794 	 * Invalidate cache to prevent missing any changes.  If
795 	 * the file is mapped, clear the page cache as well so
796 	 * those mappings will be loaded.
797 	 *
798 	 * This makes locking act as a cache coherency point.
799 	 */
800 	nfs_sync_mapping(filp->f_mapping);
801 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
802 		nfs_zap_caches(inode);
803 		if (mapping_mapped(filp->f_mapping))
804 			nfs_revalidate_mapping(inode, filp->f_mapping);
805 	}
806 out:
807 	return status;
808 }
809 
810 /*
811  * Lock a (portion of) a file
812  */
813 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
814 {
815 	struct inode *inode = filp->f_mapping->host;
816 	int ret = -ENOLCK;
817 	int is_local = 0;
818 
819 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
820 			filp, fl->fl_type, fl->fl_flags,
821 			(long long)fl->fl_start, (long long)fl->fl_end);
822 
823 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
824 
825 	if (fl->fl_flags & FL_RECLAIM)
826 		return -ENOGRACE;
827 
828 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
829 		is_local = 1;
830 
831 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
832 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
833 		if (ret < 0)
834 			goto out_err;
835 	}
836 
837 	if (IS_GETLK(cmd))
838 		ret = do_getlk(filp, cmd, fl, is_local);
839 	else if (fl->fl_type == F_UNLCK)
840 		ret = do_unlk(filp, cmd, fl, is_local);
841 	else
842 		ret = do_setlk(filp, cmd, fl, is_local);
843 out_err:
844 	return ret;
845 }
846 EXPORT_SYMBOL_GPL(nfs_lock);
847 
848 /*
849  * Lock a (portion of) a file
850  */
851 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
852 {
853 	struct inode *inode = filp->f_mapping->host;
854 	int is_local = 0;
855 
856 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
857 			filp, fl->fl_type, fl->fl_flags);
858 
859 	if (!(fl->fl_flags & FL_FLOCK))
860 		return -ENOLCK;
861 
862 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
863 		is_local = 1;
864 
865 	/* We're simulating flock() locks using posix locks on the server */
866 	if (fl->fl_type == F_UNLCK)
867 		return do_unlk(filp, cmd, fl, is_local);
868 	return do_setlk(filp, cmd, fl, is_local);
869 }
870 EXPORT_SYMBOL_GPL(nfs_flock);
871 
872 const struct file_operations nfs_file_operations = {
873 	.llseek		= nfs_file_llseek,
874 	.read_iter	= nfs_file_read,
875 	.write_iter	= nfs_file_write,
876 	.mmap		= nfs_file_mmap,
877 	.open		= nfs_file_open,
878 	.flush		= nfs_file_flush,
879 	.release	= nfs_file_release,
880 	.fsync		= nfs_file_fsync,
881 	.lock		= nfs_lock,
882 	.flock		= nfs_flock,
883 	.splice_read	= generic_file_splice_read,
884 	.splice_write	= iter_file_splice_write,
885 	.check_flags	= nfs_check_flags,
886 	.setlease	= simple_nosetlease,
887 };
888 EXPORT_SYMBOL_GPL(nfs_file_operations);
889