xref: /linux/fs/nfs/file.c (revision 1c4b5ecb7ea190fa3e9f9d6891e6c90b60e04f24)
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 	nfs_fscache_release_file(inode, filp);
88 	return 0;
89 }
90 EXPORT_SYMBOL_GPL(nfs_file_release);
91 
92 /**
93  * nfs_revalidate_file_size - Revalidate the file size
94  * @inode: pointer to inode struct
95  * @filp: pointer to struct file
96  *
97  * Revalidates the file length. This is basically a wrapper around
98  * nfs_revalidate_inode() that takes into account the fact that we may
99  * have cached writes (in which case we don't care about the server's
100  * idea of what the file length is), or O_DIRECT (in which case we
101  * shouldn't trust the cache).
102  */
103 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
104 {
105 	struct nfs_server *server = NFS_SERVER(inode);
106 
107 	if (filp->f_flags & O_DIRECT)
108 		goto force_reval;
109 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
110 		goto force_reval;
111 	return 0;
112 force_reval:
113 	return __nfs_revalidate_inode(server, inode);
114 }
115 
116 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
117 {
118 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
119 			filp, offset, whence);
120 
121 	/*
122 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
123 	 * the cached file length
124 	 */
125 	if (whence != SEEK_SET && whence != SEEK_CUR) {
126 		struct inode *inode = filp->f_mapping->host;
127 
128 		int retval = nfs_revalidate_file_size(inode, filp);
129 		if (retval < 0)
130 			return (loff_t)retval;
131 	}
132 
133 	return generic_file_llseek(filp, offset, whence);
134 }
135 EXPORT_SYMBOL_GPL(nfs_file_llseek);
136 
137 /*
138  * Flush all dirty pages, and check for write errors.
139  */
140 static int
141 nfs_file_flush(struct file *file, fl_owner_t id)
142 {
143 	struct inode	*inode = file_inode(file);
144 	errseq_t since;
145 
146 	dprintk("NFS: flush(%pD2)\n", file);
147 
148 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
149 	if ((file->f_mode & FMODE_WRITE) == 0)
150 		return 0;
151 
152 	/* Flush writes to the server and return any errors */
153 	since = filemap_sample_wb_err(file->f_mapping);
154 	nfs_wb_all(inode);
155 	return filemap_check_wb_err(file->f_mapping, since);
156 }
157 
158 ssize_t
159 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
160 {
161 	struct inode *inode = file_inode(iocb->ki_filp);
162 	ssize_t result;
163 
164 	if (iocb->ki_flags & IOCB_DIRECT)
165 		return nfs_file_direct_read(iocb, to);
166 
167 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
168 		iocb->ki_filp,
169 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
170 
171 	nfs_start_io_read(inode);
172 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
173 	if (!result) {
174 		result = generic_file_read_iter(iocb, to);
175 		if (result > 0)
176 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
177 	}
178 	nfs_end_io_read(inode);
179 	return result;
180 }
181 EXPORT_SYMBOL_GPL(nfs_file_read);
182 
183 int
184 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
185 {
186 	struct inode *inode = file_inode(file);
187 	int	status;
188 
189 	dprintk("NFS: mmap(%pD2)\n", file);
190 
191 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
192 	 *       so we call that before revalidating the mapping
193 	 */
194 	status = generic_file_mmap(file, vma);
195 	if (!status) {
196 		vma->vm_ops = &nfs_file_vm_ops;
197 		status = nfs_revalidate_mapping(inode, file->f_mapping);
198 	}
199 	return status;
200 }
201 EXPORT_SYMBOL_GPL(nfs_file_mmap);
202 
203 /*
204  * Flush any dirty pages for this process, and check for write errors.
205  * The return status from this call provides a reliable indication of
206  * whether any write errors occurred for this process.
207  */
208 static int
209 nfs_file_fsync_commit(struct file *file, int datasync)
210 {
211 	struct inode *inode = file_inode(file);
212 	int ret;
213 
214 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
215 
216 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
217 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
218 	if (ret < 0)
219 		return ret;
220 	return file_check_and_advance_wb_err(file);
221 }
222 
223 int
224 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
225 {
226 	struct nfs_open_context *ctx = nfs_file_open_context(file);
227 	struct inode *inode = file_inode(file);
228 	int ret;
229 
230 	trace_nfs_fsync_enter(inode);
231 
232 	for (;;) {
233 		ret = file_write_and_wait_range(file, start, end);
234 		if (ret != 0)
235 			break;
236 		ret = nfs_file_fsync_commit(file, datasync);
237 		if (ret != 0)
238 			break;
239 		ret = pnfs_sync_inode(inode, !!datasync);
240 		if (ret != 0)
241 			break;
242 		if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
243 			break;
244 		/*
245 		 * If nfs_file_fsync_commit detected a server reboot, then
246 		 * resend all dirty pages that might have been covered by
247 		 * the NFS_CONTEXT_RESEND_WRITES flag
248 		 */
249 		start = 0;
250 		end = LLONG_MAX;
251 	}
252 
253 	trace_nfs_fsync_exit(inode, ret);
254 	return ret;
255 }
256 EXPORT_SYMBOL_GPL(nfs_file_fsync);
257 
258 /*
259  * Decide whether a read/modify/write cycle may be more efficient
260  * then a modify/write/read cycle when writing to a page in the
261  * page cache.
262  *
263  * Some pNFS layout drivers can only read/write at a certain block
264  * granularity like all block devices and therefore we must perform
265  * read/modify/write whenever a page hasn't read yet and the data
266  * to be written there is not aligned to a block boundary and/or
267  * smaller than the block size.
268  *
269  * The modify/write/read cycle may occur if a page is read before
270  * being completely filled by the writer.  In this situation, the
271  * page must be completely written to stable storage on the server
272  * before it can be refilled by reading in the page from the server.
273  * This can lead to expensive, small, FILE_SYNC mode writes being
274  * done.
275  *
276  * It may be more efficient to read the page first if the file is
277  * open for reading in addition to writing, the page is not marked
278  * as Uptodate, it is not dirty or waiting to be committed,
279  * indicating that it was previously allocated and then modified,
280  * that there were valid bytes of data in that range of the file,
281  * and that the new data won't completely replace the old data in
282  * that range of the file.
283  */
284 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
285 {
286 	unsigned int pglen = nfs_page_length(page);
287 	unsigned int offset = pos & (PAGE_SIZE - 1);
288 	unsigned int end = offset + len;
289 
290 	return !pglen || (end >= pglen && !offset);
291 }
292 
293 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
294 			loff_t pos, unsigned int len)
295 {
296 	/*
297 	 * Up-to-date pages, those with ongoing or full-page write
298 	 * don't need read/modify/write
299 	 */
300 	if (PageUptodate(page) || PagePrivate(page) ||
301 	    nfs_full_page_write(page, pos, len))
302 		return false;
303 
304 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
305 		return true;
306 	/* Open for reading too? */
307 	if (file->f_mode & FMODE_READ)
308 		return true;
309 	return false;
310 }
311 
312 /*
313  * This does the "real" work of the write. We must allocate and lock the
314  * page to be sent back to the generic routine, which then copies the
315  * data from user space.
316  *
317  * If the writer ends up delaying the write, the writer needs to
318  * increment the page use counts until he is done with the page.
319  */
320 static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 			loff_t pos, unsigned len, unsigned flags,
322 			struct page **pagep, void **fsdata)
323 {
324 	int ret;
325 	pgoff_t index = pos >> PAGE_SHIFT;
326 	struct page *page;
327 	int once_thru = 0;
328 
329 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
330 		file, mapping->host->i_ino, len, (long long) pos);
331 
332 start:
333 	page = grab_cache_page_write_begin(mapping, index, flags);
334 	if (!page)
335 		return -ENOMEM;
336 	*pagep = page;
337 
338 	ret = nfs_flush_incompatible(file, page);
339 	if (ret) {
340 		unlock_page(page);
341 		put_page(page);
342 	} else if (!once_thru &&
343 		   nfs_want_read_modify_write(file, page, pos, len)) {
344 		once_thru = 1;
345 		ret = nfs_readpage(file, page);
346 		put_page(page);
347 		if (!ret)
348 			goto start;
349 	}
350 	return ret;
351 }
352 
353 static int nfs_write_end(struct file *file, struct address_space *mapping,
354 			loff_t pos, unsigned len, unsigned copied,
355 			struct page *page, void *fsdata)
356 {
357 	unsigned offset = pos & (PAGE_SIZE - 1);
358 	struct nfs_open_context *ctx = nfs_file_open_context(file);
359 	int status;
360 
361 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
362 		file, mapping->host->i_ino, len, (long long) pos);
363 
364 	/*
365 	 * Zero any uninitialised parts of the page, and then mark the page
366 	 * as up to date if it turns out that we're extending the file.
367 	 */
368 	if (!PageUptodate(page)) {
369 		unsigned pglen = nfs_page_length(page);
370 		unsigned end = offset + copied;
371 
372 		if (pglen == 0) {
373 			zero_user_segments(page, 0, offset,
374 					end, PAGE_SIZE);
375 			SetPageUptodate(page);
376 		} else if (end >= pglen) {
377 			zero_user_segment(page, end, PAGE_SIZE);
378 			if (offset == 0)
379 				SetPageUptodate(page);
380 		} else
381 			zero_user_segment(page, pglen, PAGE_SIZE);
382 	}
383 
384 	status = nfs_updatepage(file, page, offset, copied);
385 
386 	unlock_page(page);
387 	put_page(page);
388 
389 	if (status < 0)
390 		return status;
391 	NFS_I(mapping->host)->write_io += copied;
392 
393 	if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
394 		status = nfs_wb_all(mapping->host);
395 		if (status < 0)
396 			return status;
397 	}
398 
399 	return copied;
400 }
401 
402 /*
403  * Partially or wholly invalidate a page
404  * - Release the private state associated with a page if undergoing complete
405  *   page invalidation
406  * - Called if either PG_private or PG_fscache is set on the page
407  * - Caller holds page lock
408  */
409 static void nfs_invalidate_page(struct page *page, unsigned int offset,
410 				unsigned int length)
411 {
412 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
413 		 page, offset, length);
414 
415 	if (offset != 0 || length < PAGE_SIZE)
416 		return;
417 	/* Cancel any unstarted writes on this page */
418 	nfs_wb_page_cancel(page_file_mapping(page)->host, page);
419 	wait_on_page_fscache(page);
420 }
421 
422 /*
423  * Attempt to release the private state associated with a page
424  * - Called if either PG_private or PG_fscache is set on the page
425  * - Caller holds page lock
426  * - Return true (may release page) or false (may not)
427  */
428 static int nfs_release_page(struct page *page, gfp_t gfp)
429 {
430 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
431 
432 	/* If PagePrivate() is set, then the page is not freeable */
433 	if (PagePrivate(page))
434 		return 0;
435 	return nfs_fscache_release_page(page, gfp);
436 }
437 
438 static void nfs_check_dirty_writeback(struct page *page,
439 				bool *dirty, bool *writeback)
440 {
441 	struct nfs_inode *nfsi;
442 	struct address_space *mapping = page_file_mapping(page);
443 
444 	if (!mapping || PageSwapCache(page))
445 		return;
446 
447 	/*
448 	 * Check if an unstable page is currently being committed and
449 	 * if so, have the VM treat it as if the page is under writeback
450 	 * so it will not block due to pages that will shortly be freeable.
451 	 */
452 	nfsi = NFS_I(mapping->host);
453 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
454 		*writeback = true;
455 		return;
456 	}
457 
458 	/*
459 	 * If PagePrivate() is set, then the page is not freeable and as the
460 	 * inode is not being committed, it's not going to be cleaned in the
461 	 * near future so treat it as dirty
462 	 */
463 	if (PagePrivate(page))
464 		*dirty = true;
465 }
466 
467 /*
468  * Attempt to clear the private state associated with a page when an error
469  * occurs that requires the cached contents of an inode to be written back or
470  * destroyed
471  * - Called if either PG_private or fscache is set on the page
472  * - Caller holds page lock
473  * - Return 0 if successful, -error otherwise
474  */
475 static int nfs_launder_page(struct page *page)
476 {
477 	struct inode *inode = page_file_mapping(page)->host;
478 
479 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
480 		inode->i_ino, (long long)page_offset(page));
481 
482 	wait_on_page_fscache(page);
483 	return nfs_wb_page(inode, page);
484 }
485 
486 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
487 						sector_t *span)
488 {
489 	unsigned long blocks;
490 	long long isize;
491 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
492 	struct inode *inode = file->f_mapping->host;
493 
494 	spin_lock(&inode->i_lock);
495 	blocks = inode->i_blocks;
496 	isize = inode->i_size;
497 	spin_unlock(&inode->i_lock);
498 	if (blocks*512 < isize) {
499 		pr_warn("swap activate: swapfile has holes\n");
500 		return -EINVAL;
501 	}
502 
503 	*span = sis->pages;
504 
505 	return rpc_clnt_swap_activate(clnt);
506 }
507 
508 static void nfs_swap_deactivate(struct file *file)
509 {
510 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
511 
512 	rpc_clnt_swap_deactivate(clnt);
513 }
514 
515 const struct address_space_operations nfs_file_aops = {
516 	.readpage = nfs_readpage,
517 	.readpages = nfs_readpages,
518 	.set_page_dirty = __set_page_dirty_nobuffers,
519 	.writepage = nfs_writepage,
520 	.writepages = nfs_writepages,
521 	.write_begin = nfs_write_begin,
522 	.write_end = nfs_write_end,
523 	.invalidatepage = nfs_invalidate_page,
524 	.releasepage = nfs_release_page,
525 	.direct_IO = nfs_direct_IO,
526 #ifdef CONFIG_MIGRATION
527 	.migratepage = nfs_migrate_page,
528 #endif
529 	.launder_page = nfs_launder_page,
530 	.is_dirty_writeback = nfs_check_dirty_writeback,
531 	.error_remove_page = generic_error_remove_page,
532 	.swap_activate = nfs_swap_activate,
533 	.swap_deactivate = nfs_swap_deactivate,
534 };
535 
536 /*
537  * Notification that a PTE pointing to an NFS page is about to be made
538  * writable, implying that someone is about to modify the page through a
539  * shared-writable mapping
540  */
541 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
542 {
543 	struct page *page = vmf->page;
544 	struct file *filp = vmf->vma->vm_file;
545 	struct inode *inode = file_inode(filp);
546 	unsigned pagelen;
547 	vm_fault_t ret = VM_FAULT_NOPAGE;
548 	struct address_space *mapping;
549 
550 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
551 		filp, filp->f_mapping->host->i_ino,
552 		(long long)page_offset(page));
553 
554 	sb_start_pagefault(inode->i_sb);
555 
556 	/* make sure the cache has finished storing the page */
557 	if (PageFsCache(page) &&
558 	    wait_on_page_fscache_killable(vmf->page) < 0) {
559 		ret = VM_FAULT_RETRY;
560 		goto out;
561 	}
562 
563 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
564 			nfs_wait_bit_killable, TASK_KILLABLE);
565 
566 	lock_page(page);
567 	mapping = page_file_mapping(page);
568 	if (mapping != inode->i_mapping)
569 		goto out_unlock;
570 
571 	wait_on_page_writeback(page);
572 
573 	pagelen = nfs_page_length(page);
574 	if (pagelen == 0)
575 		goto out_unlock;
576 
577 	ret = VM_FAULT_LOCKED;
578 	if (nfs_flush_incompatible(filp, page) == 0 &&
579 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
580 		goto out;
581 
582 	ret = VM_FAULT_SIGBUS;
583 out_unlock:
584 	unlock_page(page);
585 out:
586 	sb_end_pagefault(inode->i_sb);
587 	return ret;
588 }
589 
590 static const struct vm_operations_struct nfs_file_vm_ops = {
591 	.fault = filemap_fault,
592 	.map_pages = filemap_map_pages,
593 	.page_mkwrite = nfs_vm_page_mkwrite,
594 };
595 
596 static int nfs_need_check_write(struct file *filp, struct inode *inode,
597 				int error)
598 {
599 	struct nfs_open_context *ctx;
600 
601 	ctx = nfs_file_open_context(filp);
602 	if (nfs_error_is_fatal_on_server(error) ||
603 	    nfs_ctx_key_to_expire(ctx, inode))
604 		return 1;
605 	return 0;
606 }
607 
608 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
609 {
610 	struct file *file = iocb->ki_filp;
611 	struct inode *inode = file_inode(file);
612 	unsigned int mntflags = NFS_SERVER(inode)->flags;
613 	ssize_t result, written;
614 	errseq_t since;
615 	int error;
616 
617 	result = nfs_key_timeout_notify(file, inode);
618 	if (result)
619 		return result;
620 
621 	if (iocb->ki_flags & IOCB_DIRECT)
622 		return nfs_file_direct_write(iocb, from);
623 
624 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
625 		file, iov_iter_count(from), (long long) iocb->ki_pos);
626 
627 	if (IS_SWAPFILE(inode))
628 		goto out_swapfile;
629 	/*
630 	 * O_APPEND implies that we must revalidate the file length.
631 	 */
632 	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
633 		result = nfs_revalidate_file_size(inode, file);
634 		if (result)
635 			goto out;
636 	}
637 
638 	nfs_clear_invalid_mapping(file->f_mapping);
639 
640 	since = filemap_sample_wb_err(file->f_mapping);
641 	nfs_start_io_write(inode);
642 	result = generic_write_checks(iocb, from);
643 	if (result > 0) {
644 		current->backing_dev_info = inode_to_bdi(inode);
645 		result = generic_perform_write(file, from, iocb->ki_pos);
646 		current->backing_dev_info = NULL;
647 	}
648 	nfs_end_io_write(inode);
649 	if (result <= 0)
650 		goto out;
651 
652 	written = result;
653 	iocb->ki_pos += written;
654 
655 	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
656 		result = filemap_fdatawrite_range(file->f_mapping,
657 						  iocb->ki_pos - written,
658 						  iocb->ki_pos - 1);
659 		if (result < 0)
660 			goto out;
661 	}
662 	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
663 		result = filemap_fdatawait_range(file->f_mapping,
664 						 iocb->ki_pos - written,
665 						 iocb->ki_pos - 1);
666 		if (result < 0)
667 			goto out;
668 	}
669 	result = generic_write_sync(iocb, written);
670 	if (result < 0)
671 		goto out;
672 
673 	/* Return error values */
674 	error = filemap_check_wb_err(file->f_mapping, since);
675 	if (nfs_need_check_write(file, inode, error)) {
676 		int err = nfs_wb_all(inode);
677 		if (err < 0)
678 			result = err;
679 	}
680 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
681 out:
682 	return result;
683 
684 out_swapfile:
685 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
686 	return -ETXTBSY;
687 }
688 EXPORT_SYMBOL_GPL(nfs_file_write);
689 
690 static int
691 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
692 {
693 	struct inode *inode = filp->f_mapping->host;
694 	int status = 0;
695 	unsigned int saved_type = fl->fl_type;
696 
697 	/* Try local locking first */
698 	posix_test_lock(filp, fl);
699 	if (fl->fl_type != F_UNLCK) {
700 		/* found a conflict */
701 		goto out;
702 	}
703 	fl->fl_type = saved_type;
704 
705 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
706 		goto out_noconflict;
707 
708 	if (is_local)
709 		goto out_noconflict;
710 
711 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
712 out:
713 	return status;
714 out_noconflict:
715 	fl->fl_type = F_UNLCK;
716 	goto out;
717 }
718 
719 static int
720 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
721 {
722 	struct inode *inode = filp->f_mapping->host;
723 	struct nfs_lock_context *l_ctx;
724 	int status;
725 
726 	/*
727 	 * Flush all pending writes before doing anything
728 	 * with locks..
729 	 */
730 	nfs_wb_all(inode);
731 
732 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
733 	if (!IS_ERR(l_ctx)) {
734 		status = nfs_iocounter_wait(l_ctx);
735 		nfs_put_lock_context(l_ctx);
736 		/*  NOTE: special case
737 		 * 	If we're signalled while cleaning up locks on process exit, we
738 		 * 	still need to complete the unlock.
739 		 */
740 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
741 			return status;
742 	}
743 
744 	/*
745 	 * Use local locking if mounted with "-onolock" or with appropriate
746 	 * "-olocal_lock="
747 	 */
748 	if (!is_local)
749 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
750 	else
751 		status = locks_lock_file_wait(filp, fl);
752 	return status;
753 }
754 
755 static int
756 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
757 {
758 	struct inode *inode = filp->f_mapping->host;
759 	int status;
760 
761 	/*
762 	 * Flush all pending writes before doing anything
763 	 * with locks..
764 	 */
765 	status = nfs_sync_mapping(filp->f_mapping);
766 	if (status != 0)
767 		goto out;
768 
769 	/*
770 	 * Use local locking if mounted with "-onolock" or with appropriate
771 	 * "-olocal_lock="
772 	 */
773 	if (!is_local)
774 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
775 	else
776 		status = locks_lock_file_wait(filp, fl);
777 	if (status < 0)
778 		goto out;
779 
780 	/*
781 	 * Invalidate cache to prevent missing any changes.  If
782 	 * the file is mapped, clear the page cache as well so
783 	 * those mappings will be loaded.
784 	 *
785 	 * This makes locking act as a cache coherency point.
786 	 */
787 	nfs_sync_mapping(filp->f_mapping);
788 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
789 		nfs_zap_caches(inode);
790 		if (mapping_mapped(filp->f_mapping))
791 			nfs_revalidate_mapping(inode, filp->f_mapping);
792 	}
793 out:
794 	return status;
795 }
796 
797 /*
798  * Lock a (portion of) a file
799  */
800 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
801 {
802 	struct inode *inode = filp->f_mapping->host;
803 	int ret = -ENOLCK;
804 	int is_local = 0;
805 
806 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
807 			filp, fl->fl_type, fl->fl_flags,
808 			(long long)fl->fl_start, (long long)fl->fl_end);
809 
810 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
811 
812 	if (fl->fl_flags & FL_RECLAIM)
813 		return -ENOGRACE;
814 
815 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
816 		is_local = 1;
817 
818 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
819 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
820 		if (ret < 0)
821 			goto out_err;
822 	}
823 
824 	if (IS_GETLK(cmd))
825 		ret = do_getlk(filp, cmd, fl, is_local);
826 	else if (fl->fl_type == F_UNLCK)
827 		ret = do_unlk(filp, cmd, fl, is_local);
828 	else
829 		ret = do_setlk(filp, cmd, fl, is_local);
830 out_err:
831 	return ret;
832 }
833 EXPORT_SYMBOL_GPL(nfs_lock);
834 
835 /*
836  * Lock a (portion of) a file
837  */
838 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
839 {
840 	struct inode *inode = filp->f_mapping->host;
841 	int is_local = 0;
842 
843 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
844 			filp, fl->fl_type, fl->fl_flags);
845 
846 	if (!(fl->fl_flags & FL_FLOCK))
847 		return -ENOLCK;
848 
849 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
850 		is_local = 1;
851 
852 	/* We're simulating flock() locks using posix locks on the server */
853 	if (fl->fl_type == F_UNLCK)
854 		return do_unlk(filp, cmd, fl, is_local);
855 	return do_setlk(filp, cmd, fl, is_local);
856 }
857 EXPORT_SYMBOL_GPL(nfs_flock);
858 
859 const struct file_operations nfs_file_operations = {
860 	.llseek		= nfs_file_llseek,
861 	.read_iter	= nfs_file_read,
862 	.write_iter	= nfs_file_write,
863 	.mmap		= nfs_file_mmap,
864 	.open		= nfs_file_open,
865 	.flush		= nfs_file_flush,
866 	.release	= nfs_file_release,
867 	.fsync		= nfs_file_fsync,
868 	.lock		= nfs_lock,
869 	.flock		= nfs_flock,
870 	.splice_read	= generic_file_splice_read,
871 	.splice_write	= iter_file_splice_write,
872 	.check_flags	= nfs_check_flags,
873 	.setlease	= simple_nosetlease,
874 };
875 EXPORT_SYMBOL_GPL(nfs_file_operations);
876