xref: /openbmc/linux/fs/nfs/file.c (revision 0df499ea)
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, ret2;
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 	ret2 = file_check_and_advance_wb_err(file);
216 	if (ret2 < 0)
217 		return ret2;
218 	return ret;
219 }
220 
221 int
222 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
223 {
224 	struct nfs_open_context *ctx = nfs_file_open_context(file);
225 	struct inode *inode = file_inode(file);
226 	int ret;
227 
228 	trace_nfs_fsync_enter(inode);
229 
230 	for (;;) {
231 		ret = file_write_and_wait_range(file, start, end);
232 		if (ret != 0)
233 			break;
234 		ret = nfs_file_fsync_commit(file, datasync);
235 		if (ret != 0)
236 			break;
237 		ret = pnfs_sync_inode(inode, !!datasync);
238 		if (ret != 0)
239 			break;
240 		if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
241 			break;
242 		/*
243 		 * If nfs_file_fsync_commit detected a server reboot, then
244 		 * resend all dirty pages that might have been covered by
245 		 * the NFS_CONTEXT_RESEND_WRITES flag
246 		 */
247 		start = 0;
248 		end = LLONG_MAX;
249 	}
250 
251 	trace_nfs_fsync_exit(inode, ret);
252 	return ret;
253 }
254 EXPORT_SYMBOL_GPL(nfs_file_fsync);
255 
256 /*
257  * Decide whether a read/modify/write cycle may be more efficient
258  * then a modify/write/read cycle when writing to a page in the
259  * page cache.
260  *
261  * Some pNFS layout drivers can only read/write at a certain block
262  * granularity like all block devices and therefore we must perform
263  * read/modify/write whenever a page hasn't read yet and the data
264  * to be written there is not aligned to a block boundary and/or
265  * smaller than the block size.
266  *
267  * The modify/write/read cycle may occur if a page is read before
268  * being completely filled by the writer.  In this situation, the
269  * page must be completely written to stable storage on the server
270  * before it can be refilled by reading in the page from the server.
271  * This can lead to expensive, small, FILE_SYNC mode writes being
272  * done.
273  *
274  * It may be more efficient to read the page first if the file is
275  * open for reading in addition to writing, the page is not marked
276  * as Uptodate, it is not dirty or waiting to be committed,
277  * indicating that it was previously allocated and then modified,
278  * that there were valid bytes of data in that range of the file,
279  * and that the new data won't completely replace the old data in
280  * that range of the file.
281  */
282 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
283 {
284 	unsigned int pglen = nfs_page_length(page);
285 	unsigned int offset = pos & (PAGE_SIZE - 1);
286 	unsigned int end = offset + len;
287 
288 	return !pglen || (end >= pglen && !offset);
289 }
290 
291 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
292 			loff_t pos, unsigned int len)
293 {
294 	/*
295 	 * Up-to-date pages, those with ongoing or full-page write
296 	 * don't need read/modify/write
297 	 */
298 	if (PageUptodate(page) || PagePrivate(page) ||
299 	    nfs_full_page_write(page, pos, len))
300 		return false;
301 
302 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
303 		return true;
304 	/* Open for reading too? */
305 	if (file->f_mode & FMODE_READ)
306 		return true;
307 	return false;
308 }
309 
310 /*
311  * This does the "real" work of the write. We must allocate and lock the
312  * page to be sent back to the generic routine, which then copies the
313  * data from user space.
314  *
315  * If the writer ends up delaying the write, the writer needs to
316  * increment the page use counts until he is done with the page.
317  */
318 static int nfs_write_begin(struct file *file, struct address_space *mapping,
319 			loff_t pos, unsigned len,
320 			struct page **pagep, void **fsdata)
321 {
322 	int ret;
323 	pgoff_t index = pos >> PAGE_SHIFT;
324 	struct page *page;
325 	int once_thru = 0;
326 
327 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
328 		file, mapping->host->i_ino, len, (long long) pos);
329 
330 start:
331 	page = grab_cache_page_write_begin(mapping, index);
332 	if (!page)
333 		return -ENOMEM;
334 	*pagep = page;
335 
336 	ret = nfs_flush_incompatible(file, page);
337 	if (ret) {
338 		unlock_page(page);
339 		put_page(page);
340 	} else if (!once_thru &&
341 		   nfs_want_read_modify_write(file, page, pos, len)) {
342 		once_thru = 1;
343 		ret = nfs_read_folio(file, page_folio(page));
344 		put_page(page);
345 		if (!ret)
346 			goto start;
347 	}
348 	return ret;
349 }
350 
351 static int nfs_write_end(struct file *file, struct address_space *mapping,
352 			loff_t pos, unsigned len, unsigned copied,
353 			struct page *page, void *fsdata)
354 {
355 	unsigned offset = pos & (PAGE_SIZE - 1);
356 	struct nfs_open_context *ctx = nfs_file_open_context(file);
357 	int status;
358 
359 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
360 		file, mapping->host->i_ino, len, (long long) pos);
361 
362 	/*
363 	 * Zero any uninitialised parts of the page, and then mark the page
364 	 * as up to date if it turns out that we're extending the file.
365 	 */
366 	if (!PageUptodate(page)) {
367 		unsigned pglen = nfs_page_length(page);
368 		unsigned end = offset + copied;
369 
370 		if (pglen == 0) {
371 			zero_user_segments(page, 0, offset,
372 					end, PAGE_SIZE);
373 			SetPageUptodate(page);
374 		} else if (end >= pglen) {
375 			zero_user_segment(page, end, PAGE_SIZE);
376 			if (offset == 0)
377 				SetPageUptodate(page);
378 		} else
379 			zero_user_segment(page, pglen, PAGE_SIZE);
380 	}
381 
382 	status = nfs_updatepage(file, page, offset, copied);
383 
384 	unlock_page(page);
385 	put_page(page);
386 
387 	if (status < 0)
388 		return status;
389 	NFS_I(mapping->host)->write_io += copied;
390 
391 	if (nfs_ctx_key_to_expire(ctx, mapping->host))
392 		nfs_wb_all(mapping->host);
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 folio
419  * - Called if either private or fscache flags are set on the folio
420  * - Caller holds folio lock
421  * - Return true (may release folio) or false (may not)
422  */
423 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
424 {
425 	dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
426 
427 	/* If the private flag is set, then the folio is not freeable */
428 	if (folio_test_private(folio))
429 		return false;
430 	return nfs_fscache_release_folio(folio, gfp);
431 }
432 
433 static void nfs_check_dirty_writeback(struct folio *folio,
434 				bool *dirty, bool *writeback)
435 {
436 	struct nfs_inode *nfsi;
437 	struct address_space *mapping = folio->mapping;
438 
439 	/*
440 	 * Check if an unstable folio is currently being committed and
441 	 * if so, have the VM treat it as if the folio is under writeback
442 	 * so it will not block due to folios that will shortly be freeable.
443 	 */
444 	nfsi = NFS_I(mapping->host);
445 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
446 		*writeback = true;
447 		return;
448 	}
449 
450 	/*
451 	 * If the private flag is set, then the folio is not freeable
452 	 * and as the inode is not being committed, it's not going to
453 	 * be cleaned in the near future so treat it as dirty
454 	 */
455 	if (folio_test_private(folio))
456 		*dirty = true;
457 }
458 
459 /*
460  * Attempt to clear the private state associated with a page when an error
461  * occurs that requires the cached contents of an inode to be written back or
462  * destroyed
463  * - Called if either PG_private or fscache is set on the page
464  * - Caller holds page lock
465  * - Return 0 if successful, -error otherwise
466  */
467 static int nfs_launder_folio(struct folio *folio)
468 {
469 	struct inode *inode = folio->mapping->host;
470 
471 	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
472 		inode->i_ino, folio_pos(folio));
473 
474 	folio_wait_fscache(folio);
475 	return nfs_wb_page(inode, &folio->page);
476 }
477 
478 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
479 						sector_t *span)
480 {
481 	unsigned long blocks;
482 	long long isize;
483 	int ret;
484 	struct inode *inode = file_inode(file);
485 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
486 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
487 
488 	spin_lock(&inode->i_lock);
489 	blocks = inode->i_blocks;
490 	isize = inode->i_size;
491 	spin_unlock(&inode->i_lock);
492 	if (blocks*512 < isize) {
493 		pr_warn("swap activate: swapfile has holes\n");
494 		return -EINVAL;
495 	}
496 
497 	ret = rpc_clnt_swap_activate(clnt);
498 	if (ret)
499 		return ret;
500 	ret = add_swap_extent(sis, 0, sis->max, 0);
501 	if (ret < 0) {
502 		rpc_clnt_swap_deactivate(clnt);
503 		return ret;
504 	}
505 
506 	*span = sis->pages;
507 
508 	if (cl->rpc_ops->enable_swap)
509 		cl->rpc_ops->enable_swap(inode);
510 
511 	sis->flags |= SWP_FS_OPS;
512 	return ret;
513 }
514 
515 static void nfs_swap_deactivate(struct file *file)
516 {
517 	struct inode *inode = file_inode(file);
518 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
519 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
520 
521 	rpc_clnt_swap_deactivate(clnt);
522 	if (cl->rpc_ops->disable_swap)
523 		cl->rpc_ops->disable_swap(file_inode(file));
524 }
525 
526 const struct address_space_operations nfs_file_aops = {
527 	.read_folio = nfs_read_folio,
528 	.readahead = nfs_readahead,
529 	.dirty_folio = filemap_dirty_folio,
530 	.writepage = nfs_writepage,
531 	.writepages = nfs_writepages,
532 	.write_begin = nfs_write_begin,
533 	.write_end = nfs_write_end,
534 	.invalidate_folio = nfs_invalidate_folio,
535 	.release_folio = nfs_release_folio,
536 	.migrate_folio = nfs_migrate_folio,
537 	.launder_folio = nfs_launder_folio,
538 	.is_dirty_writeback = nfs_check_dirty_writeback,
539 	.error_remove_page = generic_error_remove_page,
540 	.swap_activate = nfs_swap_activate,
541 	.swap_deactivate = nfs_swap_deactivate,
542 	.swap_rw = nfs_swap_rw,
543 };
544 
545 /*
546  * Notification that a PTE pointing to an NFS page is about to be made
547  * writable, implying that someone is about to modify the page through a
548  * shared-writable mapping
549  */
550 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
551 {
552 	struct page *page = vmf->page;
553 	struct file *filp = vmf->vma->vm_file;
554 	struct inode *inode = file_inode(filp);
555 	unsigned pagelen;
556 	vm_fault_t ret = VM_FAULT_NOPAGE;
557 	struct address_space *mapping;
558 
559 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
560 		filp, filp->f_mapping->host->i_ino,
561 		(long long)page_offset(page));
562 
563 	sb_start_pagefault(inode->i_sb);
564 
565 	/* make sure the cache has finished storing the page */
566 	if (PageFsCache(page) &&
567 	    wait_on_page_fscache_killable(vmf->page) < 0) {
568 		ret = VM_FAULT_RETRY;
569 		goto out;
570 	}
571 
572 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
573 			nfs_wait_bit_killable, TASK_KILLABLE);
574 
575 	lock_page(page);
576 	mapping = page_file_mapping(page);
577 	if (mapping != inode->i_mapping)
578 		goto out_unlock;
579 
580 	wait_on_page_writeback(page);
581 
582 	pagelen = nfs_page_length(page);
583 	if (pagelen == 0)
584 		goto out_unlock;
585 
586 	ret = VM_FAULT_LOCKED;
587 	if (nfs_flush_incompatible(filp, page) == 0 &&
588 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
589 		goto out;
590 
591 	ret = VM_FAULT_SIGBUS;
592 out_unlock:
593 	unlock_page(page);
594 out:
595 	sb_end_pagefault(inode->i_sb);
596 	return ret;
597 }
598 
599 static const struct vm_operations_struct nfs_file_vm_ops = {
600 	.fault = filemap_fault,
601 	.map_pages = filemap_map_pages,
602 	.page_mkwrite = nfs_vm_page_mkwrite,
603 };
604 
605 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
606 {
607 	struct file *file = iocb->ki_filp;
608 	struct inode *inode = file_inode(file);
609 	unsigned int mntflags = NFS_SERVER(inode)->flags;
610 	ssize_t result, written;
611 	errseq_t since;
612 	int error;
613 
614 	result = nfs_key_timeout_notify(file, inode);
615 	if (result)
616 		return result;
617 
618 	if (iocb->ki_flags & IOCB_DIRECT)
619 		return nfs_file_direct_write(iocb, from, false);
620 
621 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
622 		file, iov_iter_count(from), (long long) iocb->ki_pos);
623 
624 	if (IS_SWAPFILE(inode))
625 		goto out_swapfile;
626 	/*
627 	 * O_APPEND implies that we must revalidate the file length.
628 	 */
629 	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
630 		result = nfs_revalidate_file_size(inode, file);
631 		if (result)
632 			return result;
633 	}
634 
635 	nfs_clear_invalid_mapping(file->f_mapping);
636 
637 	since = filemap_sample_wb_err(file->f_mapping);
638 	nfs_start_io_write(inode);
639 	result = generic_write_checks(iocb, from);
640 	if (result > 0) {
641 		current->backing_dev_info = inode_to_bdi(inode);
642 		result = generic_perform_write(iocb, from);
643 		current->backing_dev_info = NULL;
644 	}
645 	nfs_end_io_write(inode);
646 	if (result <= 0)
647 		goto out;
648 
649 	written = result;
650 	iocb->ki_pos += written;
651 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
652 
653 	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
654 		result = filemap_fdatawrite_range(file->f_mapping,
655 						  iocb->ki_pos - written,
656 						  iocb->ki_pos - 1);
657 		if (result < 0)
658 			goto out;
659 	}
660 	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
661 		result = filemap_fdatawait_range(file->f_mapping,
662 						 iocb->ki_pos - written,
663 						 iocb->ki_pos - 1);
664 	}
665 	result = generic_write_sync(iocb, written);
666 	if (result < 0)
667 		return result;
668 
669 out:
670 	/* Return error values */
671 	error = filemap_check_wb_err(file->f_mapping, since);
672 	switch (error) {
673 	default:
674 		break;
675 	case -EDQUOT:
676 	case -EFBIG:
677 	case -ENOSPC:
678 		nfs_wb_all(inode);
679 		error = file_check_and_advance_wb_err(file);
680 		if (error < 0)
681 			result = error;
682 	}
683 	return result;
684 
685 out_swapfile:
686 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
687 	return -ETXTBSY;
688 }
689 EXPORT_SYMBOL_GPL(nfs_file_write);
690 
691 static int
692 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
693 {
694 	struct inode *inode = filp->f_mapping->host;
695 	int status = 0;
696 	unsigned int saved_type = fl->fl_type;
697 
698 	/* Try local locking first */
699 	posix_test_lock(filp, fl);
700 	if (fl->fl_type != F_UNLCK) {
701 		/* found a conflict */
702 		goto out;
703 	}
704 	fl->fl_type = saved_type;
705 
706 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
707 		goto out_noconflict;
708 
709 	if (is_local)
710 		goto out_noconflict;
711 
712 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
713 out:
714 	return status;
715 out_noconflict:
716 	fl->fl_type = F_UNLCK;
717 	goto out;
718 }
719 
720 static int
721 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
722 {
723 	struct inode *inode = filp->f_mapping->host;
724 	struct nfs_lock_context *l_ctx;
725 	int status;
726 
727 	/*
728 	 * Flush all pending writes before doing anything
729 	 * with locks..
730 	 */
731 	nfs_wb_all(inode);
732 
733 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
734 	if (!IS_ERR(l_ctx)) {
735 		status = nfs_iocounter_wait(l_ctx);
736 		nfs_put_lock_context(l_ctx);
737 		/*  NOTE: special case
738 		 * 	If we're signalled while cleaning up locks on process exit, we
739 		 * 	still need to complete the unlock.
740 		 */
741 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
742 			return status;
743 	}
744 
745 	/*
746 	 * Use local locking if mounted with "-onolock" or with appropriate
747 	 * "-olocal_lock="
748 	 */
749 	if (!is_local)
750 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
751 	else
752 		status = locks_lock_file_wait(filp, fl);
753 	return status;
754 }
755 
756 static int
757 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
758 {
759 	struct inode *inode = filp->f_mapping->host;
760 	int status;
761 
762 	/*
763 	 * Flush all pending writes before doing anything
764 	 * with locks..
765 	 */
766 	status = nfs_sync_mapping(filp->f_mapping);
767 	if (status != 0)
768 		goto out;
769 
770 	/*
771 	 * Use local locking if mounted with "-onolock" or with appropriate
772 	 * "-olocal_lock="
773 	 */
774 	if (!is_local)
775 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
776 	else
777 		status = locks_lock_file_wait(filp, fl);
778 	if (status < 0)
779 		goto out;
780 
781 	/*
782 	 * Invalidate cache to prevent missing any changes.  If
783 	 * the file is mapped, clear the page cache as well so
784 	 * those mappings will be loaded.
785 	 *
786 	 * This makes locking act as a cache coherency point.
787 	 */
788 	nfs_sync_mapping(filp->f_mapping);
789 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
790 		nfs_zap_caches(inode);
791 		if (mapping_mapped(filp->f_mapping))
792 			nfs_revalidate_mapping(inode, filp->f_mapping);
793 	}
794 out:
795 	return status;
796 }
797 
798 /*
799  * Lock a (portion of) a file
800  */
801 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
802 {
803 	struct inode *inode = filp->f_mapping->host;
804 	int ret = -ENOLCK;
805 	int is_local = 0;
806 
807 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
808 			filp, fl->fl_type, fl->fl_flags,
809 			(long long)fl->fl_start, (long long)fl->fl_end);
810 
811 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
812 
813 	if (fl->fl_flags & FL_RECLAIM)
814 		return -ENOGRACE;
815 
816 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
817 		is_local = 1;
818 
819 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
820 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
821 		if (ret < 0)
822 			goto out_err;
823 	}
824 
825 	if (IS_GETLK(cmd))
826 		ret = do_getlk(filp, cmd, fl, is_local);
827 	else if (fl->fl_type == F_UNLCK)
828 		ret = do_unlk(filp, cmd, fl, is_local);
829 	else
830 		ret = do_setlk(filp, cmd, fl, is_local);
831 out_err:
832 	return ret;
833 }
834 EXPORT_SYMBOL_GPL(nfs_lock);
835 
836 /*
837  * Lock a (portion of) a file
838  */
839 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
840 {
841 	struct inode *inode = filp->f_mapping->host;
842 	int is_local = 0;
843 
844 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
845 			filp, fl->fl_type, fl->fl_flags);
846 
847 	if (!(fl->fl_flags & FL_FLOCK))
848 		return -ENOLCK;
849 
850 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
851 		is_local = 1;
852 
853 	/* We're simulating flock() locks using posix locks on the server */
854 	if (fl->fl_type == F_UNLCK)
855 		return do_unlk(filp, cmd, fl, is_local);
856 	return do_setlk(filp, cmd, fl, is_local);
857 }
858 EXPORT_SYMBOL_GPL(nfs_flock);
859 
860 const struct file_operations nfs_file_operations = {
861 	.llseek		= nfs_file_llseek,
862 	.read_iter	= nfs_file_read,
863 	.write_iter	= nfs_file_write,
864 	.mmap		= nfs_file_mmap,
865 	.open		= nfs_file_open,
866 	.flush		= nfs_file_flush,
867 	.release	= nfs_file_release,
868 	.fsync		= nfs_file_fsync,
869 	.lock		= nfs_lock,
870 	.flock		= nfs_flock,
871 	.splice_read	= generic_file_splice_read,
872 	.splice_write	= iter_file_splice_write,
873 	.check_flags	= nfs_check_flags,
874 	.setlease	= simple_nosetlease,
875 };
876 EXPORT_SYMBOL_GPL(nfs_file_operations);
877