xref: /openbmc/linux/fs/nfs/file.c (revision ae213c44)
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 /* Hack for future NFS swap support */
48 #ifndef IS_SWAPFILE
49 # define IS_SWAPFILE(inode)	(0)
50 #endif
51 
52 int nfs_check_flags(int flags)
53 {
54 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
55 		return -EINVAL;
56 
57 	return 0;
58 }
59 EXPORT_SYMBOL_GPL(nfs_check_flags);
60 
61 /*
62  * Open file
63  */
64 static int
65 nfs_file_open(struct inode *inode, struct file *filp)
66 {
67 	int res;
68 
69 	dprintk("NFS: open file(%pD2)\n", filp);
70 
71 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
72 	res = nfs_check_flags(filp->f_flags);
73 	if (res)
74 		return res;
75 
76 	res = nfs_open(inode, filp);
77 	return res;
78 }
79 
80 int
81 nfs_file_release(struct inode *inode, struct file *filp)
82 {
83 	dprintk("NFS: release(%pD2)\n", filp);
84 
85 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
86 	nfs_file_clear_open_context(filp);
87 	return 0;
88 }
89 EXPORT_SYMBOL_GPL(nfs_file_release);
90 
91 /**
92  * nfs_revalidate_size - Revalidate the file size
93  * @inode: pointer to inode struct
94  * @filp: pointer to struct file
95  *
96  * Revalidates the file length. This is basically a wrapper around
97  * nfs_revalidate_inode() that takes into account the fact that we may
98  * have cached writes (in which case we don't care about the server's
99  * idea of what the file length is), or O_DIRECT (in which case we
100  * shouldn't trust the cache).
101  */
102 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103 {
104 	struct nfs_server *server = NFS_SERVER(inode);
105 
106 	if (filp->f_flags & O_DIRECT)
107 		goto force_reval;
108 	if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
109 		goto force_reval;
110 	return 0;
111 force_reval:
112 	return __nfs_revalidate_inode(server, inode);
113 }
114 
115 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
116 {
117 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
118 			filp, offset, whence);
119 
120 	/*
121 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
122 	 * the cached file length
123 	 */
124 	if (whence != SEEK_SET && whence != SEEK_CUR) {
125 		struct inode *inode = filp->f_mapping->host;
126 
127 		int retval = nfs_revalidate_file_size(inode, filp);
128 		if (retval < 0)
129 			return (loff_t)retval;
130 	}
131 
132 	return generic_file_llseek(filp, offset, whence);
133 }
134 EXPORT_SYMBOL_GPL(nfs_file_llseek);
135 
136 /*
137  * Flush all dirty pages, and check for write errors.
138  */
139 static int
140 nfs_file_flush(struct file *file, fl_owner_t id)
141 {
142 	struct inode	*inode = file_inode(file);
143 
144 	dprintk("NFS: flush(%pD2)\n", file);
145 
146 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
147 	if ((file->f_mode & FMODE_WRITE) == 0)
148 		return 0;
149 
150 	/* Flush writes to the server and return any errors */
151 	return nfs_wb_all(inode);
152 }
153 
154 ssize_t
155 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
156 {
157 	struct inode *inode = file_inode(iocb->ki_filp);
158 	ssize_t result;
159 
160 	if (iocb->ki_flags & IOCB_DIRECT)
161 		return nfs_file_direct_read(iocb, to);
162 
163 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
164 		iocb->ki_filp,
165 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
166 
167 	nfs_start_io_read(inode);
168 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
169 	if (!result) {
170 		result = generic_file_read_iter(iocb, to);
171 		if (result > 0)
172 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
173 	}
174 	nfs_end_io_read(inode);
175 	return result;
176 }
177 EXPORT_SYMBOL_GPL(nfs_file_read);
178 
179 int
180 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
181 {
182 	struct inode *inode = file_inode(file);
183 	int	status;
184 
185 	dprintk("NFS: mmap(%pD2)\n", file);
186 
187 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
188 	 *       so we call that before revalidating the mapping
189 	 */
190 	status = generic_file_mmap(file, vma);
191 	if (!status) {
192 		vma->vm_ops = &nfs_file_vm_ops;
193 		status = nfs_revalidate_mapping(inode, file->f_mapping);
194 	}
195 	return status;
196 }
197 EXPORT_SYMBOL_GPL(nfs_file_mmap);
198 
199 /*
200  * Flush any dirty pages for this process, and check for write errors.
201  * The return status from this call provides a reliable indication of
202  * whether any write errors occurred for this process.
203  */
204 static int
205 nfs_file_fsync_commit(struct file *file, int datasync)
206 {
207 	struct nfs_open_context *ctx = nfs_file_open_context(file);
208 	struct inode *inode = file_inode(file);
209 	int do_resend, status;
210 	int ret = 0;
211 
212 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
213 
214 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
215 	do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
216 	status = nfs_commit_inode(inode, FLUSH_SYNC);
217 	if (status == 0)
218 		status = file_check_and_advance_wb_err(file);
219 	if (status < 0) {
220 		ret = status;
221 		goto out;
222 	}
223 	do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
224 	if (do_resend)
225 		ret = -EAGAIN;
226 out:
227 	return ret;
228 }
229 
230 int
231 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
232 {
233 	int ret;
234 	struct inode *inode = file_inode(file);
235 
236 	trace_nfs_fsync_enter(inode);
237 
238 	do {
239 		ret = file_write_and_wait_range(file, start, end);
240 		if (ret != 0)
241 			break;
242 		ret = nfs_file_fsync_commit(file, datasync);
243 		if (!ret)
244 			ret = pnfs_sync_inode(inode, !!datasync);
245 		/*
246 		 * If nfs_file_fsync_commit detected a server reboot, then
247 		 * resend all dirty pages that might have been covered by
248 		 * the NFS_CONTEXT_RESEND_WRITES flag
249 		 */
250 		start = 0;
251 		end = LLONG_MAX;
252 	} while (ret == -EAGAIN);
253 
254 	trace_nfs_fsync_exit(inode, ret);
255 	return ret;
256 }
257 EXPORT_SYMBOL_GPL(nfs_file_fsync);
258 
259 /*
260  * Decide whether a read/modify/write cycle may be more efficient
261  * then a modify/write/read cycle when writing to a page in the
262  * page cache.
263  *
264  * Some pNFS layout drivers can only read/write at a certain block
265  * granularity like all block devices and therefore we must perform
266  * read/modify/write whenever a page hasn't read yet and the data
267  * to be written there is not aligned to a block boundary and/or
268  * smaller than the block size.
269  *
270  * The modify/write/read cycle may occur if a page is read before
271  * being completely filled by the writer.  In this situation, the
272  * page must be completely written to stable storage on the server
273  * before it can be refilled by reading in the page from the server.
274  * This can lead to expensive, small, FILE_SYNC mode writes being
275  * done.
276  *
277  * It may be more efficient to read the page first if the file is
278  * open for reading in addition to writing, the page is not marked
279  * as Uptodate, it is not dirty or waiting to be committed,
280  * indicating that it was previously allocated and then modified,
281  * that there were valid bytes of data in that range of the file,
282  * and that the new data won't completely replace the old data in
283  * that range of the file.
284  */
285 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
286 {
287 	unsigned int pglen = nfs_page_length(page);
288 	unsigned int offset = pos & (PAGE_SIZE - 1);
289 	unsigned int end = offset + len;
290 
291 	return !pglen || (end >= pglen && !offset);
292 }
293 
294 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
295 			loff_t pos, unsigned int len)
296 {
297 	/*
298 	 * Up-to-date pages, those with ongoing or full-page write
299 	 * don't need read/modify/write
300 	 */
301 	if (PageUptodate(page) || PagePrivate(page) ||
302 	    nfs_full_page_write(page, pos, len))
303 		return false;
304 
305 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
306 		return true;
307 	/* Open for reading too? */
308 	if (file->f_mode & FMODE_READ)
309 		return true;
310 	return false;
311 }
312 
313 /*
314  * This does the "real" work of the write. We must allocate and lock the
315  * page to be sent back to the generic routine, which then copies the
316  * data from user space.
317  *
318  * If the writer ends up delaying the write, the writer needs to
319  * increment the page use counts until he is done with the page.
320  */
321 static int nfs_write_begin(struct file *file, struct address_space *mapping,
322 			loff_t pos, unsigned len, unsigned flags,
323 			struct page **pagep, void **fsdata)
324 {
325 	int ret;
326 	pgoff_t index = pos >> PAGE_SHIFT;
327 	struct page *page;
328 	int once_thru = 0;
329 
330 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
331 		file, mapping->host->i_ino, len, (long long) pos);
332 
333 start:
334 	page = grab_cache_page_write_begin(mapping, index, flags);
335 	if (!page)
336 		return -ENOMEM;
337 	*pagep = page;
338 
339 	ret = nfs_flush_incompatible(file, page);
340 	if (ret) {
341 		unlock_page(page);
342 		put_page(page);
343 	} else if (!once_thru &&
344 		   nfs_want_read_modify_write(file, page, pos, len)) {
345 		once_thru = 1;
346 		ret = nfs_readpage(file, page);
347 		put_page(page);
348 		if (!ret)
349 			goto start;
350 	}
351 	return ret;
352 }
353 
354 static int nfs_write_end(struct file *file, struct address_space *mapping,
355 			loff_t pos, unsigned len, unsigned copied,
356 			struct page *page, void *fsdata)
357 {
358 	unsigned offset = pos & (PAGE_SIZE - 1);
359 	struct nfs_open_context *ctx = nfs_file_open_context(file);
360 	int status;
361 
362 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
363 		file, mapping->host->i_ino, len, (long long) pos);
364 
365 	/*
366 	 * Zero any uninitialised parts of the page, and then mark the page
367 	 * as up to date if it turns out that we're extending the file.
368 	 */
369 	if (!PageUptodate(page)) {
370 		unsigned pglen = nfs_page_length(page);
371 		unsigned end = offset + copied;
372 
373 		if (pglen == 0) {
374 			zero_user_segments(page, 0, offset,
375 					end, PAGE_SIZE);
376 			SetPageUptodate(page);
377 		} else if (end >= pglen) {
378 			zero_user_segment(page, end, PAGE_SIZE);
379 			if (offset == 0)
380 				SetPageUptodate(page);
381 		} else
382 			zero_user_segment(page, pglen, PAGE_SIZE);
383 	}
384 
385 	status = nfs_updatepage(file, page, offset, copied);
386 
387 	unlock_page(page);
388 	put_page(page);
389 
390 	if (status < 0)
391 		return status;
392 	NFS_I(mapping->host)->write_io += copied;
393 
394 	if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
395 		status = nfs_wb_all(mapping->host);
396 		if (status < 0)
397 			return status;
398 	}
399 
400 	return copied;
401 }
402 
403 /*
404  * Partially or wholly invalidate a page
405  * - Release the private state associated with a page if undergoing complete
406  *   page invalidation
407  * - Called if either PG_private or PG_fscache is set on the page
408  * - Caller holds page lock
409  */
410 static void nfs_invalidate_page(struct page *page, unsigned int offset,
411 				unsigned int length)
412 {
413 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
414 		 page, offset, length);
415 
416 	if (offset != 0 || length < PAGE_SIZE)
417 		return;
418 	/* Cancel any unstarted writes on this page */
419 	nfs_wb_page_cancel(page_file_mapping(page)->host, page);
420 
421 	nfs_fscache_invalidate_page(page, page->mapping->host);
422 }
423 
424 /*
425  * Attempt to release the private state associated with a page
426  * - Called if either PG_private or PG_fscache is set on the page
427  * - Caller holds page lock
428  * - Return true (may release page) or false (may not)
429  */
430 static int nfs_release_page(struct page *page, gfp_t gfp)
431 {
432 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
433 
434 	/* If PagePrivate() is set, then the page is not freeable */
435 	if (PagePrivate(page))
436 		return 0;
437 	return nfs_fscache_release_page(page, gfp);
438 }
439 
440 static void nfs_check_dirty_writeback(struct page *page,
441 				bool *dirty, bool *writeback)
442 {
443 	struct nfs_inode *nfsi;
444 	struct address_space *mapping = page_file_mapping(page);
445 
446 	if (!mapping || PageSwapCache(page))
447 		return;
448 
449 	/*
450 	 * Check if an unstable page is currently being committed and
451 	 * if so, have the VM treat it as if the page is under writeback
452 	 * so it will not block due to pages that will shortly be freeable.
453 	 */
454 	nfsi = NFS_I(mapping->host);
455 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
456 		*writeback = true;
457 		return;
458 	}
459 
460 	/*
461 	 * If PagePrivate() is set, then the page is not freeable and as the
462 	 * inode is not being committed, it's not going to be cleaned in the
463 	 * near future so treat it as dirty
464 	 */
465 	if (PagePrivate(page))
466 		*dirty = true;
467 }
468 
469 /*
470  * Attempt to clear the private state associated with a page when an error
471  * occurs that requires the cached contents of an inode to be written back or
472  * destroyed
473  * - Called if either PG_private or fscache is set on the page
474  * - Caller holds page lock
475  * - Return 0 if successful, -error otherwise
476  */
477 static int nfs_launder_page(struct page *page)
478 {
479 	struct inode *inode = page_file_mapping(page)->host;
480 	struct nfs_inode *nfsi = NFS_I(inode);
481 
482 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
483 		inode->i_ino, (long long)page_offset(page));
484 
485 	nfs_fscache_wait_on_page_write(nfsi, page);
486 	return nfs_wb_page(inode, page);
487 }
488 
489 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
490 						sector_t *span)
491 {
492 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
493 
494 	*span = sis->pages;
495 
496 	return rpc_clnt_swap_activate(clnt);
497 }
498 
499 static void nfs_swap_deactivate(struct file *file)
500 {
501 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
502 
503 	rpc_clnt_swap_deactivate(clnt);
504 }
505 
506 const struct address_space_operations nfs_file_aops = {
507 	.readpage = nfs_readpage,
508 	.readpages = nfs_readpages,
509 	.set_page_dirty = __set_page_dirty_nobuffers,
510 	.writepage = nfs_writepage,
511 	.writepages = nfs_writepages,
512 	.write_begin = nfs_write_begin,
513 	.write_end = nfs_write_end,
514 	.invalidatepage = nfs_invalidate_page,
515 	.releasepage = nfs_release_page,
516 	.direct_IO = nfs_direct_IO,
517 #ifdef CONFIG_MIGRATION
518 	.migratepage = nfs_migrate_page,
519 #endif
520 	.launder_page = nfs_launder_page,
521 	.is_dirty_writeback = nfs_check_dirty_writeback,
522 	.error_remove_page = generic_error_remove_page,
523 	.swap_activate = nfs_swap_activate,
524 	.swap_deactivate = nfs_swap_deactivate,
525 };
526 
527 /*
528  * Notification that a PTE pointing to an NFS page is about to be made
529  * writable, implying that someone is about to modify the page through a
530  * shared-writable mapping
531  */
532 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
533 {
534 	struct page *page = vmf->page;
535 	struct file *filp = vmf->vma->vm_file;
536 	struct inode *inode = file_inode(filp);
537 	unsigned pagelen;
538 	vm_fault_t ret = VM_FAULT_NOPAGE;
539 	struct address_space *mapping;
540 
541 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
542 		filp, filp->f_mapping->host->i_ino,
543 		(long long)page_offset(page));
544 
545 	sb_start_pagefault(inode->i_sb);
546 
547 	/* make sure the cache has finished storing the page */
548 	nfs_fscache_wait_on_page_write(NFS_I(inode), page);
549 
550 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
551 			nfs_wait_bit_killable, TASK_KILLABLE);
552 
553 	lock_page(page);
554 	mapping = page_file_mapping(page);
555 	if (mapping != inode->i_mapping)
556 		goto out_unlock;
557 
558 	wait_on_page_writeback(page);
559 
560 	pagelen = nfs_page_length(page);
561 	if (pagelen == 0)
562 		goto out_unlock;
563 
564 	ret = VM_FAULT_LOCKED;
565 	if (nfs_flush_incompatible(filp, page) == 0 &&
566 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
567 		goto out;
568 
569 	ret = VM_FAULT_SIGBUS;
570 out_unlock:
571 	unlock_page(page);
572 out:
573 	sb_end_pagefault(inode->i_sb);
574 	return ret;
575 }
576 
577 static const struct vm_operations_struct nfs_file_vm_ops = {
578 	.fault = filemap_fault,
579 	.map_pages = filemap_map_pages,
580 	.page_mkwrite = nfs_vm_page_mkwrite,
581 };
582 
583 static int nfs_need_check_write(struct file *filp, struct inode *inode)
584 {
585 	struct nfs_open_context *ctx;
586 
587 	ctx = nfs_file_open_context(filp);
588 	if (nfs_ctx_key_to_expire(ctx, inode))
589 		return 1;
590 	return 0;
591 }
592 
593 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
594 {
595 	struct file *file = iocb->ki_filp;
596 	struct inode *inode = file_inode(file);
597 	unsigned long written = 0;
598 	ssize_t result;
599 
600 	result = nfs_key_timeout_notify(file, inode);
601 	if (result)
602 		return result;
603 
604 	if (iocb->ki_flags & IOCB_DIRECT)
605 		return nfs_file_direct_write(iocb, from);
606 
607 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
608 		file, iov_iter_count(from), (long long) iocb->ki_pos);
609 
610 	if (IS_SWAPFILE(inode))
611 		goto out_swapfile;
612 	/*
613 	 * O_APPEND implies that we must revalidate the file length.
614 	 */
615 	if (iocb->ki_flags & IOCB_APPEND) {
616 		result = nfs_revalidate_file_size(inode, file);
617 		if (result)
618 			goto out;
619 	}
620 	if (iocb->ki_pos > i_size_read(inode))
621 		nfs_revalidate_mapping(inode, file->f_mapping);
622 
623 	nfs_start_io_write(inode);
624 	result = generic_write_checks(iocb, from);
625 	if (result > 0) {
626 		current->backing_dev_info = inode_to_bdi(inode);
627 		result = generic_perform_write(file, from, iocb->ki_pos);
628 		current->backing_dev_info = NULL;
629 	}
630 	nfs_end_io_write(inode);
631 	if (result <= 0)
632 		goto out;
633 
634 	written = result;
635 	iocb->ki_pos += written;
636 	result = generic_write_sync(iocb, written);
637 	if (result < 0)
638 		goto out;
639 
640 	/* Return error values */
641 	if (nfs_need_check_write(file, inode)) {
642 		int err = nfs_wb_all(inode);
643 		if (err < 0)
644 			result = err;
645 	}
646 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
647 out:
648 	return result;
649 
650 out_swapfile:
651 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
652 	return -EBUSY;
653 }
654 EXPORT_SYMBOL_GPL(nfs_file_write);
655 
656 static int
657 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
658 {
659 	struct inode *inode = filp->f_mapping->host;
660 	int status = 0;
661 	unsigned int saved_type = fl->fl_type;
662 
663 	/* Try local locking first */
664 	posix_test_lock(filp, fl);
665 	if (fl->fl_type != F_UNLCK) {
666 		/* found a conflict */
667 		goto out;
668 	}
669 	fl->fl_type = saved_type;
670 
671 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
672 		goto out_noconflict;
673 
674 	if (is_local)
675 		goto out_noconflict;
676 
677 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
678 out:
679 	return status;
680 out_noconflict:
681 	fl->fl_type = F_UNLCK;
682 	goto out;
683 }
684 
685 static int
686 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
687 {
688 	struct inode *inode = filp->f_mapping->host;
689 	struct nfs_lock_context *l_ctx;
690 	int status;
691 
692 	/*
693 	 * Flush all pending writes before doing anything
694 	 * with locks..
695 	 */
696 	nfs_wb_all(inode);
697 
698 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
699 	if (!IS_ERR(l_ctx)) {
700 		status = nfs_iocounter_wait(l_ctx);
701 		nfs_put_lock_context(l_ctx);
702 		/*  NOTE: special case
703 		 * 	If we're signalled while cleaning up locks on process exit, we
704 		 * 	still need to complete the unlock.
705 		 */
706 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
707 			return status;
708 	}
709 
710 	/*
711 	 * Use local locking if mounted with "-onolock" or with appropriate
712 	 * "-olocal_lock="
713 	 */
714 	if (!is_local)
715 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
716 	else
717 		status = locks_lock_file_wait(filp, fl);
718 	return status;
719 }
720 
721 static int
722 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
723 {
724 	struct inode *inode = filp->f_mapping->host;
725 	int status;
726 
727 	/*
728 	 * Flush all pending writes before doing anything
729 	 * with locks..
730 	 */
731 	status = nfs_sync_mapping(filp->f_mapping);
732 	if (status != 0)
733 		goto out;
734 
735 	/*
736 	 * Use local locking if mounted with "-onolock" or with appropriate
737 	 * "-olocal_lock="
738 	 */
739 	if (!is_local)
740 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
741 	else
742 		status = locks_lock_file_wait(filp, fl);
743 	if (status < 0)
744 		goto out;
745 
746 	/*
747 	 * Invalidate cache to prevent missing any changes.  If
748 	 * the file is mapped, clear the page cache as well so
749 	 * those mappings will be loaded.
750 	 *
751 	 * This makes locking act as a cache coherency point.
752 	 */
753 	nfs_sync_mapping(filp->f_mapping);
754 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
755 		nfs_zap_caches(inode);
756 		if (mapping_mapped(filp->f_mapping))
757 			nfs_revalidate_mapping(inode, filp->f_mapping);
758 	}
759 out:
760 	return status;
761 }
762 
763 /*
764  * Lock a (portion of) a file
765  */
766 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
767 {
768 	struct inode *inode = filp->f_mapping->host;
769 	int ret = -ENOLCK;
770 	int is_local = 0;
771 
772 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
773 			filp, fl->fl_type, fl->fl_flags,
774 			(long long)fl->fl_start, (long long)fl->fl_end);
775 
776 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
777 
778 	/* No mandatory locks over NFS */
779 	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
780 		goto out_err;
781 
782 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
783 		is_local = 1;
784 
785 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
786 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
787 		if (ret < 0)
788 			goto out_err;
789 	}
790 
791 	if (IS_GETLK(cmd))
792 		ret = do_getlk(filp, cmd, fl, is_local);
793 	else if (fl->fl_type == F_UNLCK)
794 		ret = do_unlk(filp, cmd, fl, is_local);
795 	else
796 		ret = do_setlk(filp, cmd, fl, is_local);
797 out_err:
798 	return ret;
799 }
800 EXPORT_SYMBOL_GPL(nfs_lock);
801 
802 /*
803  * Lock a (portion of) a file
804  */
805 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
806 {
807 	struct inode *inode = filp->f_mapping->host;
808 	int is_local = 0;
809 
810 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
811 			filp, fl->fl_type, fl->fl_flags);
812 
813 	if (!(fl->fl_flags & FL_FLOCK))
814 		return -ENOLCK;
815 
816 	/*
817 	 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
818 	 * any standard. In principle we might be able to support LOCK_MAND
819 	 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
820 	 * NFS code is not set up for it.
821 	 */
822 	if (fl->fl_type & LOCK_MAND)
823 		return -EINVAL;
824 
825 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
826 		is_local = 1;
827 
828 	/* We're simulating flock() locks using posix locks on the server */
829 	if (fl->fl_type == F_UNLCK)
830 		return do_unlk(filp, cmd, fl, is_local);
831 	return do_setlk(filp, cmd, fl, is_local);
832 }
833 EXPORT_SYMBOL_GPL(nfs_flock);
834 
835 const struct file_operations nfs_file_operations = {
836 	.llseek		= nfs_file_llseek,
837 	.read_iter	= nfs_file_read,
838 	.write_iter	= nfs_file_write,
839 	.mmap		= nfs_file_mmap,
840 	.open		= nfs_file_open,
841 	.flush		= nfs_file_flush,
842 	.release	= nfs_file_release,
843 	.fsync		= nfs_file_fsync,
844 	.lock		= nfs_lock,
845 	.flock		= nfs_flock,
846 	.splice_read	= generic_file_splice_read,
847 	.splice_write	= iter_file_splice_write,
848 	.check_flags	= nfs_check_flags,
849 	.setlease	= simple_nosetlease,
850 };
851 EXPORT_SYMBOL_GPL(nfs_file_operations);
852