xref: /linux/fs/nfs/file.c (revision b8321ed4a40c02054f930ca59d3570caa27bc86c)
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 	return res;
73 }
74 
75 int
76 nfs_file_release(struct inode *inode, struct file *filp)
77 {
78 	dprintk("NFS: release(%pD2)\n", filp);
79 
80 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
81 	nfs_file_clear_open_context(filp);
82 	nfs_fscache_release_file(inode, filp);
83 	return 0;
84 }
85 EXPORT_SYMBOL_GPL(nfs_file_release);
86 
87 /**
88  * nfs_revalidate_file_size - Revalidate the file size
89  * @inode: pointer to inode struct
90  * @filp: pointer to struct file
91  *
92  * Revalidates the file length. This is basically a wrapper around
93  * nfs_revalidate_inode() that takes into account the fact that we may
94  * have cached writes (in which case we don't care about the server's
95  * idea of what the file length is), or O_DIRECT (in which case we
96  * shouldn't trust the cache).
97  */
98 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
99 {
100 	struct nfs_server *server = NFS_SERVER(inode);
101 
102 	if (filp->f_flags & O_DIRECT)
103 		goto force_reval;
104 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
105 		goto force_reval;
106 	return 0;
107 force_reval:
108 	return __nfs_revalidate_inode(server, inode);
109 }
110 
111 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
112 {
113 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
114 			filp, offset, whence);
115 
116 	/*
117 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
118 	 * the cached file length
119 	 */
120 	if (whence != SEEK_SET && whence != SEEK_CUR) {
121 		struct inode *inode = filp->f_mapping->host;
122 
123 		int retval = nfs_revalidate_file_size(inode, filp);
124 		if (retval < 0)
125 			return (loff_t)retval;
126 	}
127 
128 	return generic_file_llseek(filp, offset, whence);
129 }
130 EXPORT_SYMBOL_GPL(nfs_file_llseek);
131 
132 /*
133  * Flush all dirty pages, and check for write errors.
134  */
135 static int
136 nfs_file_flush(struct file *file, fl_owner_t id)
137 {
138 	struct inode	*inode = file_inode(file);
139 	errseq_t since;
140 
141 	dprintk("NFS: flush(%pD2)\n", file);
142 
143 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
144 	if ((file->f_mode & FMODE_WRITE) == 0)
145 		return 0;
146 
147 	/* Flush writes to the server and return any errors */
148 	since = filemap_sample_wb_err(file->f_mapping);
149 	nfs_wb_all(inode);
150 	return filemap_check_wb_err(file->f_mapping, since);
151 }
152 
153 ssize_t
154 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
155 {
156 	struct inode *inode = file_inode(iocb->ki_filp);
157 	ssize_t result;
158 
159 	if (iocb->ki_flags & IOCB_DIRECT)
160 		return nfs_file_direct_read(iocb, to, false);
161 
162 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
163 		iocb->ki_filp,
164 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
165 
166 	nfs_start_io_read(inode);
167 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
168 	if (!result) {
169 		result = generic_file_read_iter(iocb, to);
170 		if (result > 0)
171 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
172 	}
173 	nfs_end_io_read(inode);
174 	return result;
175 }
176 EXPORT_SYMBOL_GPL(nfs_file_read);
177 
178 int
179 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
180 {
181 	struct inode *inode = file_inode(file);
182 	int	status;
183 
184 	dprintk("NFS: mmap(%pD2)\n", file);
185 
186 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
187 	 *       so we call that before revalidating the mapping
188 	 */
189 	status = generic_file_mmap(file, vma);
190 	if (!status) {
191 		vma->vm_ops = &nfs_file_vm_ops;
192 		status = nfs_revalidate_mapping(inode, file->f_mapping);
193 	}
194 	return status;
195 }
196 EXPORT_SYMBOL_GPL(nfs_file_mmap);
197 
198 /*
199  * Flush any dirty pages for this process, and check for write errors.
200  * The return status from this call provides a reliable indication of
201  * whether any write errors occurred for this process.
202  */
203 static int
204 nfs_file_fsync_commit(struct file *file, int datasync)
205 {
206 	struct inode *inode = file_inode(file);
207 	int ret;
208 
209 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
210 
211 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
212 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
213 	if (ret < 0)
214 		return ret;
215 	return file_check_and_advance_wb_err(file);
216 }
217 
218 int
219 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
220 {
221 	struct nfs_open_context *ctx = nfs_file_open_context(file);
222 	struct inode *inode = file_inode(file);
223 	int ret;
224 
225 	trace_nfs_fsync_enter(inode);
226 
227 	for (;;) {
228 		ret = file_write_and_wait_range(file, start, end);
229 		if (ret != 0)
230 			break;
231 		ret = nfs_file_fsync_commit(file, datasync);
232 		if (ret != 0)
233 			break;
234 		ret = pnfs_sync_inode(inode, !!datasync);
235 		if (ret != 0)
236 			break;
237 		if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
238 			break;
239 		/*
240 		 * If nfs_file_fsync_commit detected a server reboot, then
241 		 * resend all dirty pages that might have been covered by
242 		 * the NFS_CONTEXT_RESEND_WRITES flag
243 		 */
244 		start = 0;
245 		end = LLONG_MAX;
246 	}
247 
248 	trace_nfs_fsync_exit(inode, ret);
249 	return ret;
250 }
251 EXPORT_SYMBOL_GPL(nfs_file_fsync);
252 
253 /*
254  * Decide whether a read/modify/write cycle may be more efficient
255  * then a modify/write/read cycle when writing to a page in the
256  * page cache.
257  *
258  * Some pNFS layout drivers can only read/write at a certain block
259  * granularity like all block devices and therefore we must perform
260  * read/modify/write whenever a page hasn't read yet and the data
261  * to be written there is not aligned to a block boundary and/or
262  * smaller than the block size.
263  *
264  * The modify/write/read cycle may occur if a page is read before
265  * being completely filled by the writer.  In this situation, the
266  * page must be completely written to stable storage on the server
267  * before it can be refilled by reading in the page from the server.
268  * This can lead to expensive, small, FILE_SYNC mode writes being
269  * done.
270  *
271  * It may be more efficient to read the page first if the file is
272  * open for reading in addition to writing, the page is not marked
273  * as Uptodate, it is not dirty or waiting to be committed,
274  * indicating that it was previously allocated and then modified,
275  * that there were valid bytes of data in that range of the file,
276  * and that the new data won't completely replace the old data in
277  * that range of the file.
278  */
279 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
280 {
281 	unsigned int pglen = nfs_page_length(page);
282 	unsigned int offset = pos & (PAGE_SIZE - 1);
283 	unsigned int end = offset + len;
284 
285 	return !pglen || (end >= pglen && !offset);
286 }
287 
288 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
289 			loff_t pos, unsigned int len)
290 {
291 	/*
292 	 * Up-to-date pages, those with ongoing or full-page write
293 	 * don't need read/modify/write
294 	 */
295 	if (PageUptodate(page) || PagePrivate(page) ||
296 	    nfs_full_page_write(page, pos, len))
297 		return false;
298 
299 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
300 		return true;
301 	/* Open for reading too? */
302 	if (file->f_mode & FMODE_READ)
303 		return true;
304 	return false;
305 }
306 
307 /*
308  * This does the "real" work of the write. We must allocate and lock the
309  * page to be sent back to the generic routine, which then copies the
310  * data from user space.
311  *
312  * If the writer ends up delaying the write, the writer needs to
313  * increment the page use counts until he is done with the page.
314  */
315 static int nfs_write_begin(struct file *file, struct address_space *mapping,
316 			loff_t pos, unsigned len, unsigned flags,
317 			struct page **pagep, void **fsdata)
318 {
319 	int ret;
320 	pgoff_t index = pos >> PAGE_SHIFT;
321 	struct page *page;
322 	int once_thru = 0;
323 
324 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
325 		file, mapping->host->i_ino, len, (long long) pos);
326 
327 start:
328 	page = grab_cache_page_write_begin(mapping, index, flags);
329 	if (!page)
330 		return -ENOMEM;
331 	*pagep = page;
332 
333 	ret = nfs_flush_incompatible(file, page);
334 	if (ret) {
335 		unlock_page(page);
336 		put_page(page);
337 	} else if (!once_thru &&
338 		   nfs_want_read_modify_write(file, page, pos, len)) {
339 		once_thru = 1;
340 		ret = nfs_readpage(file, page);
341 		put_page(page);
342 		if (!ret)
343 			goto start;
344 	}
345 	return ret;
346 }
347 
348 static int nfs_write_end(struct file *file, struct address_space *mapping,
349 			loff_t pos, unsigned len, unsigned copied,
350 			struct page *page, void *fsdata)
351 {
352 	unsigned offset = pos & (PAGE_SIZE - 1);
353 	struct nfs_open_context *ctx = nfs_file_open_context(file);
354 	int status;
355 
356 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
357 		file, mapping->host->i_ino, len, (long long) pos);
358 
359 	/*
360 	 * Zero any uninitialised parts of the page, and then mark the page
361 	 * as up to date if it turns out that we're extending the file.
362 	 */
363 	if (!PageUptodate(page)) {
364 		unsigned pglen = nfs_page_length(page);
365 		unsigned end = offset + copied;
366 
367 		if (pglen == 0) {
368 			zero_user_segments(page, 0, offset,
369 					end, PAGE_SIZE);
370 			SetPageUptodate(page);
371 		} else if (end >= pglen) {
372 			zero_user_segment(page, end, PAGE_SIZE);
373 			if (offset == 0)
374 				SetPageUptodate(page);
375 		} else
376 			zero_user_segment(page, pglen, PAGE_SIZE);
377 	}
378 
379 	status = nfs_updatepage(file, page, offset, copied);
380 
381 	unlock_page(page);
382 	put_page(page);
383 
384 	if (status < 0)
385 		return status;
386 	NFS_I(mapping->host)->write_io += copied;
387 
388 	if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
389 		status = nfs_wb_all(mapping->host);
390 		if (status < 0)
391 			return status;
392 	}
393 
394 	return copied;
395 }
396 
397 /*
398  * Partially or wholly invalidate a page
399  * - Release the private state associated with a page if undergoing complete
400  *   page invalidation
401  * - Called if either PG_private or PG_fscache is set on the page
402  * - Caller holds page lock
403  */
404 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
405 				size_t length)
406 {
407 	dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
408 		 folio->index, offset, length);
409 
410 	if (offset != 0 || length < folio_size(folio))
411 		return;
412 	/* Cancel any unstarted writes on this page */
413 	nfs_wb_folio_cancel(folio->mapping->host, folio);
414 	folio_wait_fscache(folio);
415 }
416 
417 /*
418  * Attempt to release the private state associated with a page
419  * - Called if either PG_private or PG_fscache is set on the page
420  * - Caller holds page lock
421  * - Return true (may release page) or false (may not)
422  */
423 static int nfs_release_page(struct page *page, gfp_t gfp)
424 {
425 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
426 
427 	/* If PagePrivate() is set, then the page is not freeable */
428 	if (PagePrivate(page))
429 		return 0;
430 	return nfs_fscache_release_page(page, gfp);
431 }
432 
433 static void nfs_check_dirty_writeback(struct page *page,
434 				bool *dirty, bool *writeback)
435 {
436 	struct nfs_inode *nfsi;
437 	struct address_space *mapping = page_file_mapping(page);
438 
439 	if (!mapping || PageSwapCache(page))
440 		return;
441 
442 	/*
443 	 * Check if an unstable page is currently being committed and
444 	 * if so, have the VM treat it as if the page is under writeback
445 	 * so it will not block due to pages that will shortly be freeable.
446 	 */
447 	nfsi = NFS_I(mapping->host);
448 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
449 		*writeback = true;
450 		return;
451 	}
452 
453 	/*
454 	 * If PagePrivate() is set, then the page is not freeable and as the
455 	 * inode is not being committed, it's not going to be cleaned in the
456 	 * near future so treat it as dirty
457 	 */
458 	if (PagePrivate(page))
459 		*dirty = true;
460 }
461 
462 /*
463  * Attempt to clear the private state associated with a page when an error
464  * occurs that requires the cached contents of an inode to be written back or
465  * destroyed
466  * - Called if either PG_private or fscache is set on the page
467  * - Caller holds page lock
468  * - Return 0 if successful, -error otherwise
469  */
470 static int nfs_launder_folio(struct folio *folio)
471 {
472 	struct inode *inode = folio->mapping->host;
473 
474 	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
475 		inode->i_ino, folio_pos(folio));
476 
477 	folio_wait_fscache(folio);
478 	return nfs_wb_page(inode, &folio->page);
479 }
480 
481 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
482 						sector_t *span)
483 {
484 	unsigned long blocks;
485 	long long isize;
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 	*span = sis->pages;
500 
501 
502 	if (cl->rpc_ops->enable_swap)
503 		cl->rpc_ops->enable_swap(inode);
504 
505 	return rpc_clnt_swap_activate(clnt);
506 }
507 
508 static void nfs_swap_deactivate(struct file *file)
509 {
510 	struct inode *inode = file_inode(file);
511 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
512 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
513 
514 	rpc_clnt_swap_deactivate(clnt);
515 	if (cl->rpc_ops->disable_swap)
516 		cl->rpc_ops->disable_swap(file_inode(file));
517 }
518 
519 const struct address_space_operations nfs_file_aops = {
520 	.readpage = nfs_readpage,
521 	.readahead = nfs_readahead,
522 	.dirty_folio = filemap_dirty_folio,
523 	.writepage = nfs_writepage,
524 	.writepages = nfs_writepages,
525 	.write_begin = nfs_write_begin,
526 	.write_end = nfs_write_end,
527 	.invalidate_folio = nfs_invalidate_folio,
528 	.releasepage = nfs_release_page,
529 	.direct_IO = nfs_direct_IO,
530 #ifdef CONFIG_MIGRATION
531 	.migratepage = nfs_migrate_page,
532 #endif
533 	.launder_folio = nfs_launder_folio,
534 	.is_dirty_writeback = nfs_check_dirty_writeback,
535 	.error_remove_page = generic_error_remove_page,
536 	.swap_activate = nfs_swap_activate,
537 	.swap_deactivate = nfs_swap_deactivate,
538 };
539 
540 /*
541  * Notification that a PTE pointing to an NFS page is about to be made
542  * writable, implying that someone is about to modify the page through a
543  * shared-writable mapping
544  */
545 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
546 {
547 	struct page *page = vmf->page;
548 	struct file *filp = vmf->vma->vm_file;
549 	struct inode *inode = file_inode(filp);
550 	unsigned pagelen;
551 	vm_fault_t ret = VM_FAULT_NOPAGE;
552 	struct address_space *mapping;
553 
554 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
555 		filp, filp->f_mapping->host->i_ino,
556 		(long long)page_offset(page));
557 
558 	sb_start_pagefault(inode->i_sb);
559 
560 	/* make sure the cache has finished storing the page */
561 	if (PageFsCache(page) &&
562 	    wait_on_page_fscache_killable(vmf->page) < 0) {
563 		ret = VM_FAULT_RETRY;
564 		goto out;
565 	}
566 
567 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
568 			nfs_wait_bit_killable, TASK_KILLABLE);
569 
570 	lock_page(page);
571 	mapping = page_file_mapping(page);
572 	if (mapping != inode->i_mapping)
573 		goto out_unlock;
574 
575 	wait_on_page_writeback(page);
576 
577 	pagelen = nfs_page_length(page);
578 	if (pagelen == 0)
579 		goto out_unlock;
580 
581 	ret = VM_FAULT_LOCKED;
582 	if (nfs_flush_incompatible(filp, page) == 0 &&
583 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
584 		goto out;
585 
586 	ret = VM_FAULT_SIGBUS;
587 out_unlock:
588 	unlock_page(page);
589 out:
590 	sb_end_pagefault(inode->i_sb);
591 	return ret;
592 }
593 
594 static const struct vm_operations_struct nfs_file_vm_ops = {
595 	.fault = filemap_fault,
596 	.map_pages = filemap_map_pages,
597 	.page_mkwrite = nfs_vm_page_mkwrite,
598 };
599 
600 static int nfs_need_check_write(struct file *filp, struct inode *inode,
601 				int error)
602 {
603 	struct nfs_open_context *ctx;
604 
605 	ctx = nfs_file_open_context(filp);
606 	if (nfs_error_is_fatal_on_server(error) ||
607 	    nfs_ctx_key_to_expire(ctx, inode))
608 		return 1;
609 	return 0;
610 }
611 
612 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
613 {
614 	struct file *file = iocb->ki_filp;
615 	struct inode *inode = file_inode(file);
616 	unsigned int mntflags = NFS_SERVER(inode)->flags;
617 	ssize_t result, written;
618 	errseq_t since;
619 	int error;
620 
621 	result = nfs_key_timeout_notify(file, inode);
622 	if (result)
623 		return result;
624 
625 	if (iocb->ki_flags & IOCB_DIRECT)
626 		return nfs_file_direct_write(iocb, from, false);
627 
628 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
629 		file, iov_iter_count(from), (long long) iocb->ki_pos);
630 
631 	if (IS_SWAPFILE(inode))
632 		goto out_swapfile;
633 	/*
634 	 * O_APPEND implies that we must revalidate the file length.
635 	 */
636 	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
637 		result = nfs_revalidate_file_size(inode, file);
638 		if (result)
639 			goto out;
640 	}
641 
642 	nfs_clear_invalid_mapping(file->f_mapping);
643 
644 	since = filemap_sample_wb_err(file->f_mapping);
645 	nfs_start_io_write(inode);
646 	result = generic_write_checks(iocb, from);
647 	if (result > 0) {
648 		current->backing_dev_info = inode_to_bdi(inode);
649 		result = generic_perform_write(file, from, iocb->ki_pos);
650 		current->backing_dev_info = NULL;
651 	}
652 	nfs_end_io_write(inode);
653 	if (result <= 0)
654 		goto out;
655 
656 	written = result;
657 	iocb->ki_pos += written;
658 
659 	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
660 		result = filemap_fdatawrite_range(file->f_mapping,
661 						  iocb->ki_pos - written,
662 						  iocb->ki_pos - 1);
663 		if (result < 0)
664 			goto out;
665 	}
666 	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
667 		result = filemap_fdatawait_range(file->f_mapping,
668 						 iocb->ki_pos - written,
669 						 iocb->ki_pos - 1);
670 		if (result < 0)
671 			goto out;
672 	}
673 	result = generic_write_sync(iocb, written);
674 	if (result < 0)
675 		goto out;
676 
677 	/* Return error values */
678 	error = filemap_check_wb_err(file->f_mapping, since);
679 	if (nfs_need_check_write(file, inode, error)) {
680 		int err = nfs_wb_all(inode);
681 		if (err < 0)
682 			result = err;
683 	}
684 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
685 out:
686 	return result;
687 
688 out_swapfile:
689 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 	return -ETXTBSY;
691 }
692 EXPORT_SYMBOL_GPL(nfs_file_write);
693 
694 static int
695 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
696 {
697 	struct inode *inode = filp->f_mapping->host;
698 	int status = 0;
699 	unsigned int saved_type = fl->fl_type;
700 
701 	/* Try local locking first */
702 	posix_test_lock(filp, fl);
703 	if (fl->fl_type != F_UNLCK) {
704 		/* found a conflict */
705 		goto out;
706 	}
707 	fl->fl_type = saved_type;
708 
709 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
710 		goto out_noconflict;
711 
712 	if (is_local)
713 		goto out_noconflict;
714 
715 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
716 out:
717 	return status;
718 out_noconflict:
719 	fl->fl_type = F_UNLCK;
720 	goto out;
721 }
722 
723 static int
724 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
725 {
726 	struct inode *inode = filp->f_mapping->host;
727 	struct nfs_lock_context *l_ctx;
728 	int status;
729 
730 	/*
731 	 * Flush all pending writes before doing anything
732 	 * with locks..
733 	 */
734 	nfs_wb_all(inode);
735 
736 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
737 	if (!IS_ERR(l_ctx)) {
738 		status = nfs_iocounter_wait(l_ctx);
739 		nfs_put_lock_context(l_ctx);
740 		/*  NOTE: special case
741 		 * 	If we're signalled while cleaning up locks on process exit, we
742 		 * 	still need to complete the unlock.
743 		 */
744 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
745 			return status;
746 	}
747 
748 	/*
749 	 * Use local locking if mounted with "-onolock" or with appropriate
750 	 * "-olocal_lock="
751 	 */
752 	if (!is_local)
753 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
754 	else
755 		status = locks_lock_file_wait(filp, fl);
756 	return status;
757 }
758 
759 static int
760 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
761 {
762 	struct inode *inode = filp->f_mapping->host;
763 	int status;
764 
765 	/*
766 	 * Flush all pending writes before doing anything
767 	 * with locks..
768 	 */
769 	status = nfs_sync_mapping(filp->f_mapping);
770 	if (status != 0)
771 		goto out;
772 
773 	/*
774 	 * Use local locking if mounted with "-onolock" or with appropriate
775 	 * "-olocal_lock="
776 	 */
777 	if (!is_local)
778 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
779 	else
780 		status = locks_lock_file_wait(filp, fl);
781 	if (status < 0)
782 		goto out;
783 
784 	/*
785 	 * Invalidate cache to prevent missing any changes.  If
786 	 * the file is mapped, clear the page cache as well so
787 	 * those mappings will be loaded.
788 	 *
789 	 * This makes locking act as a cache coherency point.
790 	 */
791 	nfs_sync_mapping(filp->f_mapping);
792 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
793 		nfs_zap_caches(inode);
794 		if (mapping_mapped(filp->f_mapping))
795 			nfs_revalidate_mapping(inode, filp->f_mapping);
796 	}
797 out:
798 	return status;
799 }
800 
801 /*
802  * Lock a (portion of) a file
803  */
804 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
805 {
806 	struct inode *inode = filp->f_mapping->host;
807 	int ret = -ENOLCK;
808 	int is_local = 0;
809 
810 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
811 			filp, fl->fl_type, fl->fl_flags,
812 			(long long)fl->fl_start, (long long)fl->fl_end);
813 
814 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
815 
816 	if (fl->fl_flags & FL_RECLAIM)
817 		return -ENOGRACE;
818 
819 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
820 		is_local = 1;
821 
822 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
823 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
824 		if (ret < 0)
825 			goto out_err;
826 	}
827 
828 	if (IS_GETLK(cmd))
829 		ret = do_getlk(filp, cmd, fl, is_local);
830 	else if (fl->fl_type == F_UNLCK)
831 		ret = do_unlk(filp, cmd, fl, is_local);
832 	else
833 		ret = do_setlk(filp, cmd, fl, is_local);
834 out_err:
835 	return ret;
836 }
837 EXPORT_SYMBOL_GPL(nfs_lock);
838 
839 /*
840  * Lock a (portion of) a file
841  */
842 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
843 {
844 	struct inode *inode = filp->f_mapping->host;
845 	int is_local = 0;
846 
847 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
848 			filp, fl->fl_type, fl->fl_flags);
849 
850 	if (!(fl->fl_flags & FL_FLOCK))
851 		return -ENOLCK;
852 
853 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
854 		is_local = 1;
855 
856 	/* We're simulating flock() locks using posix locks on the server */
857 	if (fl->fl_type == F_UNLCK)
858 		return do_unlk(filp, cmd, fl, is_local);
859 	return do_setlk(filp, cmd, fl, is_local);
860 }
861 EXPORT_SYMBOL_GPL(nfs_flock);
862 
863 const struct file_operations nfs_file_operations = {
864 	.llseek		= nfs_file_llseek,
865 	.read_iter	= nfs_file_read,
866 	.write_iter	= nfs_file_write,
867 	.mmap		= nfs_file_mmap,
868 	.open		= nfs_file_open,
869 	.flush		= nfs_file_flush,
870 	.release	= nfs_file_release,
871 	.fsync		= nfs_file_fsync,
872 	.lock		= nfs_lock,
873 	.flock		= nfs_flock,
874 	.splice_read	= generic_file_splice_read,
875 	.splice_write	= iter_file_splice_write,
876 	.check_flags	= nfs_check_flags,
877 	.setlease	= simple_nosetlease,
878 };
879 EXPORT_SYMBOL_GPL(nfs_file_operations);
880