xref: /openbmc/linux/fs/nfs/file.c (revision 92a2c6b2)
1 /*
2  *  linux/fs/nfs/file.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  *
6  *  Changes Copyright (C) 1994 by Florian La Roche
7  *   - Do not copy data too often around in the kernel.
8  *   - In nfs_file_read the return value of kmalloc wasn't checked.
9  *   - Put in a better version of read look-ahead buffering. Original idea
10  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
11  *
12  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
13  *
14  *  Total rewrite of read side for new NFS buffer cache.. Linus.
15  *
16  *  nfs regular file handling functions
17  */
18 
19 #include <linux/module.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/fcntl.h>
24 #include <linux/stat.h>
25 #include <linux/nfs_fs.h>
26 #include <linux/nfs_mount.h>
27 #include <linux/mm.h>
28 #include <linux/pagemap.h>
29 #include <linux/aio.h>
30 #include <linux/gfp.h>
31 #include <linux/swap.h>
32 
33 #include <asm/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 	return nfs_release(inode, filp);
87 }
88 EXPORT_SYMBOL_GPL(nfs_file_release);
89 
90 /**
91  * nfs_revalidate_size - Revalidate the file size
92  * @inode - pointer to inode struct
93  * @file - pointer to struct file
94  *
95  * Revalidates the file length. This is basically a wrapper around
96  * nfs_revalidate_inode() that takes into account the fact that we may
97  * have cached writes (in which case we don't care about the server's
98  * idea of what the file length is), or O_DIRECT (in which case we
99  * shouldn't trust the cache).
100  */
101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102 {
103 	struct nfs_server *server = NFS_SERVER(inode);
104 	struct nfs_inode *nfsi = NFS_I(inode);
105 
106 	if (nfs_have_delegated_attributes(inode))
107 		goto out_noreval;
108 
109 	if (filp->f_flags & O_DIRECT)
110 		goto force_reval;
111 	if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112 		goto force_reval;
113 	if (nfs_attribute_timeout(inode))
114 		goto force_reval;
115 out_noreval:
116 	return 0;
117 force_reval:
118 	return __nfs_revalidate_inode(server, inode);
119 }
120 
121 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
122 {
123 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 			filp, offset, whence);
125 
126 	/*
127 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
128 	 * the cached file length
129 	 */
130 	if (whence != SEEK_SET && whence != SEEK_CUR) {
131 		struct inode *inode = filp->f_mapping->host;
132 
133 		int retval = nfs_revalidate_file_size(inode, filp);
134 		if (retval < 0)
135 			return (loff_t)retval;
136 	}
137 
138 	return generic_file_llseek(filp, offset, whence);
139 }
140 EXPORT_SYMBOL_GPL(nfs_file_llseek);
141 
142 /*
143  * Flush all dirty pages, and check for write errors.
144  */
145 int
146 nfs_file_flush(struct file *file, fl_owner_t id)
147 {
148 	struct inode	*inode = file_inode(file);
149 
150 	dprintk("NFS: flush(%pD2)\n", file);
151 
152 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
153 	if ((file->f_mode & FMODE_WRITE) == 0)
154 		return 0;
155 
156 	/*
157 	 * If we're holding a write delegation, then just start the i/o
158 	 * but don't wait for completion (or send a commit).
159 	 */
160 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
161 		return filemap_fdatawrite(file->f_mapping);
162 
163 	/* Flush writes to the server and return any errors */
164 	return vfs_fsync(file, 0);
165 }
166 EXPORT_SYMBOL_GPL(nfs_file_flush);
167 
168 ssize_t
169 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
170 {
171 	struct inode *inode = file_inode(iocb->ki_filp);
172 	ssize_t result;
173 
174 	if (iocb->ki_filp->f_flags & O_DIRECT)
175 		return nfs_file_direct_read(iocb, to, iocb->ki_pos);
176 
177 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
178 		iocb->ki_filp,
179 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
180 
181 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
182 	if (!result) {
183 		result = generic_file_read_iter(iocb, to);
184 		if (result > 0)
185 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 	}
187 	return result;
188 }
189 EXPORT_SYMBOL_GPL(nfs_file_read);
190 
191 ssize_t
192 nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 		     struct pipe_inode_info *pipe, size_t count,
194 		     unsigned int flags)
195 {
196 	struct inode *inode = file_inode(filp);
197 	ssize_t res;
198 
199 	dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 		filp, (unsigned long) count, (unsigned long long) *ppos);
201 
202 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
203 	if (!res) {
204 		res = generic_file_splice_read(filp, ppos, pipe, count, flags);
205 		if (res > 0)
206 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 	}
208 	return res;
209 }
210 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
211 
212 int
213 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214 {
215 	struct inode *inode = file_inode(file);
216 	int	status;
217 
218 	dprintk("NFS: mmap(%pD2)\n", file);
219 
220 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 	 *       so we call that before revalidating the mapping
222 	 */
223 	status = generic_file_mmap(file, vma);
224 	if (!status) {
225 		vma->vm_ops = &nfs_file_vm_ops;
226 		status = nfs_revalidate_mapping(inode, file->f_mapping);
227 	}
228 	return status;
229 }
230 EXPORT_SYMBOL_GPL(nfs_file_mmap);
231 
232 /*
233  * Flush any dirty pages for this process, and check for write errors.
234  * The return status from this call provides a reliable indication of
235  * whether any write errors occurred for this process.
236  *
237  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238  * disk, but it retrieves and clears ctx->error after synching, despite
239  * the two being set at the same time in nfs_context_set_write_error().
240  * This is because the former is used to notify the _next_ call to
241  * nfs_file_write() that a write error occurred, and hence cause it to
242  * fall back to doing a synchronous write.
243  */
244 int
245 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
246 {
247 	struct nfs_open_context *ctx = nfs_file_open_context(file);
248 	struct inode *inode = file_inode(file);
249 	int have_error, do_resend, status;
250 	int ret = 0;
251 
252 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
253 
254 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
255 	do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
256 	have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 	status = nfs_commit_inode(inode, FLUSH_SYNC);
258 	have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
259 	if (have_error) {
260 		ret = xchg(&ctx->error, 0);
261 		if (ret)
262 			goto out;
263 	}
264 	if (status < 0) {
265 		ret = status;
266 		goto out;
267 	}
268 	do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 	if (do_resend)
270 		ret = -EAGAIN;
271 out:
272 	return ret;
273 }
274 EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
275 
276 static int
277 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278 {
279 	int ret;
280 	struct inode *inode = file_inode(file);
281 
282 	trace_nfs_fsync_enter(inode);
283 
284 	do {
285 		ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 		if (ret != 0)
287 			break;
288 		mutex_lock(&inode->i_mutex);
289 		ret = nfs_file_fsync_commit(file, start, end, datasync);
290 		mutex_unlock(&inode->i_mutex);
291 		/*
292 		 * If nfs_file_fsync_commit detected a server reboot, then
293 		 * resend all dirty pages that might have been covered by
294 		 * the NFS_CONTEXT_RESEND_WRITES flag
295 		 */
296 		start = 0;
297 		end = LLONG_MAX;
298 	} while (ret == -EAGAIN);
299 
300 	trace_nfs_fsync_exit(inode, ret);
301 	return ret;
302 }
303 
304 /*
305  * Decide whether a read/modify/write cycle may be more efficient
306  * then a modify/write/read cycle when writing to a page in the
307  * page cache.
308  *
309  * The modify/write/read cycle may occur if a page is read before
310  * being completely filled by the writer.  In this situation, the
311  * page must be completely written to stable storage on the server
312  * before it can be refilled by reading in the page from the server.
313  * This can lead to expensive, small, FILE_SYNC mode writes being
314  * done.
315  *
316  * It may be more efficient to read the page first if the file is
317  * open for reading in addition to writing, the page is not marked
318  * as Uptodate, it is not dirty or waiting to be committed,
319  * indicating that it was previously allocated and then modified,
320  * that there were valid bytes of data in that range of the file,
321  * and that the new data won't completely replace the old data in
322  * that range of the file.
323  */
324 static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 			loff_t pos, unsigned len)
326 {
327 	unsigned int pglen = nfs_page_length(page);
328 	unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 	unsigned int end = offset + len;
330 
331 	if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
332 		if (!PageUptodate(page))
333 			return 1;
334 		return 0;
335 	}
336 
337 	if ((file->f_mode & FMODE_READ) &&	/* open for read? */
338 	    !PageUptodate(page) &&		/* Uptodate? */
339 	    !PagePrivate(page) &&		/* i/o request already? */
340 	    pglen &&				/* valid bytes of file? */
341 	    (end < pglen || offset))		/* replace all valid bytes? */
342 		return 1;
343 	return 0;
344 }
345 
346 /*
347  * This does the "real" work of the write. We must allocate and lock the
348  * page to be sent back to the generic routine, which then copies the
349  * data from user space.
350  *
351  * If the writer ends up delaying the write, the writer needs to
352  * increment the page use counts until he is done with the page.
353  */
354 static int nfs_write_begin(struct file *file, struct address_space *mapping,
355 			loff_t pos, unsigned len, unsigned flags,
356 			struct page **pagep, void **fsdata)
357 {
358 	int ret;
359 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
360 	struct page *page;
361 	int once_thru = 0;
362 
363 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
364 		file, mapping->host->i_ino, len, (long long) pos);
365 
366 start:
367 	/*
368 	 * Prevent starvation issues if someone is doing a consistency
369 	 * sync-to-disk
370 	 */
371 	ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
372 				 nfs_wait_bit_killable, TASK_KILLABLE);
373 	if (ret)
374 		return ret;
375 
376 	page = grab_cache_page_write_begin(mapping, index, flags);
377 	if (!page)
378 		return -ENOMEM;
379 	*pagep = page;
380 
381 	ret = nfs_flush_incompatible(file, page);
382 	if (ret) {
383 		unlock_page(page);
384 		page_cache_release(page);
385 	} else if (!once_thru &&
386 		   nfs_want_read_modify_write(file, page, pos, len)) {
387 		once_thru = 1;
388 		ret = nfs_readpage(file, page);
389 		page_cache_release(page);
390 		if (!ret)
391 			goto start;
392 	}
393 	return ret;
394 }
395 
396 static int nfs_write_end(struct file *file, struct address_space *mapping,
397 			loff_t pos, unsigned len, unsigned copied,
398 			struct page *page, void *fsdata)
399 {
400 	unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
401 	struct nfs_open_context *ctx = nfs_file_open_context(file);
402 	int status;
403 
404 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
405 		file, mapping->host->i_ino, len, (long long) pos);
406 
407 	/*
408 	 * Zero any uninitialised parts of the page, and then mark the page
409 	 * as up to date if it turns out that we're extending the file.
410 	 */
411 	if (!PageUptodate(page)) {
412 		unsigned pglen = nfs_page_length(page);
413 		unsigned end = offset + len;
414 
415 		if (pglen == 0) {
416 			zero_user_segments(page, 0, offset,
417 					end, PAGE_CACHE_SIZE);
418 			SetPageUptodate(page);
419 		} else if (end >= pglen) {
420 			zero_user_segment(page, end, PAGE_CACHE_SIZE);
421 			if (offset == 0)
422 				SetPageUptodate(page);
423 		} else
424 			zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
425 	}
426 
427 	status = nfs_updatepage(file, page, offset, copied);
428 
429 	unlock_page(page);
430 	page_cache_release(page);
431 
432 	if (status < 0)
433 		return status;
434 	NFS_I(mapping->host)->write_io += copied;
435 
436 	if (nfs_ctx_key_to_expire(ctx)) {
437 		status = nfs_wb_all(mapping->host);
438 		if (status < 0)
439 			return status;
440 	}
441 
442 	return copied;
443 }
444 
445 /*
446  * Partially or wholly invalidate a page
447  * - Release the private state associated with a page if undergoing complete
448  *   page invalidation
449  * - Called if either PG_private or PG_fscache is set on the page
450  * - Caller holds page lock
451  */
452 static void nfs_invalidate_page(struct page *page, unsigned int offset,
453 				unsigned int length)
454 {
455 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
456 		 page, offset, length);
457 
458 	if (offset != 0 || length < PAGE_CACHE_SIZE)
459 		return;
460 	/* Cancel any unstarted writes on this page */
461 	nfs_wb_page_cancel(page_file_mapping(page)->host, page);
462 
463 	nfs_fscache_invalidate_page(page, page->mapping->host);
464 }
465 
466 /*
467  * Attempt to release the private state associated with a page
468  * - Called if either PG_private or PG_fscache is set on the page
469  * - Caller holds page lock
470  * - Return true (may release page) or false (may not)
471  */
472 static int nfs_release_page(struct page *page, gfp_t gfp)
473 {
474 	struct address_space *mapping = page->mapping;
475 
476 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
477 
478 	/* Always try to initiate a 'commit' if relevant, but only
479 	 * wait for it if __GFP_WAIT is set.  Even then, only wait 1
480 	 * second and only if the 'bdi' is not congested.
481 	 * Waiting indefinitely can cause deadlocks when the NFS
482 	 * server is on this machine, when a new TCP connection is
483 	 * needed and in other rare cases.  There is no particular
484 	 * need to wait extensively here.  A short wait has the
485 	 * benefit that someone else can worry about the freezer.
486 	 */
487 	if (mapping) {
488 		struct nfs_server *nfss = NFS_SERVER(mapping->host);
489 		nfs_commit_inode(mapping->host, 0);
490 		if ((gfp & __GFP_WAIT) &&
491 		    !bdi_write_congested(&nfss->backing_dev_info)) {
492 			wait_on_page_bit_killable_timeout(page, PG_private,
493 							  HZ);
494 			if (PagePrivate(page))
495 				set_bdi_congested(&nfss->backing_dev_info,
496 						  BLK_RW_ASYNC);
497 		}
498 	}
499 	/* If PagePrivate() is set, then the page is not freeable */
500 	if (PagePrivate(page))
501 		return 0;
502 	return nfs_fscache_release_page(page, gfp);
503 }
504 
505 static void nfs_check_dirty_writeback(struct page *page,
506 				bool *dirty, bool *writeback)
507 {
508 	struct nfs_inode *nfsi;
509 	struct address_space *mapping = page_file_mapping(page);
510 
511 	if (!mapping || PageSwapCache(page))
512 		return;
513 
514 	/*
515 	 * Check if an unstable page is currently being committed and
516 	 * if so, have the VM treat it as if the page is under writeback
517 	 * so it will not block due to pages that will shortly be freeable.
518 	 */
519 	nfsi = NFS_I(mapping->host);
520 	if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
521 		*writeback = true;
522 		return;
523 	}
524 
525 	/*
526 	 * If PagePrivate() is set, then the page is not freeable and as the
527 	 * inode is not being committed, it's not going to be cleaned in the
528 	 * near future so treat it as dirty
529 	 */
530 	if (PagePrivate(page))
531 		*dirty = true;
532 }
533 
534 /*
535  * Attempt to clear the private state associated with a page when an error
536  * occurs that requires the cached contents of an inode to be written back or
537  * destroyed
538  * - Called if either PG_private or fscache is set on the page
539  * - Caller holds page lock
540  * - Return 0 if successful, -error otherwise
541  */
542 static int nfs_launder_page(struct page *page)
543 {
544 	struct inode *inode = page_file_mapping(page)->host;
545 	struct nfs_inode *nfsi = NFS_I(inode);
546 
547 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
548 		inode->i_ino, (long long)page_offset(page));
549 
550 	nfs_fscache_wait_on_page_write(nfsi, page);
551 	return nfs_wb_page(inode, page);
552 }
553 
554 #ifdef CONFIG_NFS_SWAP
555 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
556 						sector_t *span)
557 {
558 	int ret;
559 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
560 
561 	*span = sis->pages;
562 
563 	rcu_read_lock();
564 	ret = xs_swapper(rcu_dereference(clnt->cl_xprt), 1);
565 	rcu_read_unlock();
566 
567 	return ret;
568 }
569 
570 static void nfs_swap_deactivate(struct file *file)
571 {
572 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
573 
574 	rcu_read_lock();
575 	xs_swapper(rcu_dereference(clnt->cl_xprt), 0);
576 	rcu_read_unlock();
577 }
578 #endif
579 
580 const struct address_space_operations nfs_file_aops = {
581 	.readpage = nfs_readpage,
582 	.readpages = nfs_readpages,
583 	.set_page_dirty = __set_page_dirty_nobuffers,
584 	.writepage = nfs_writepage,
585 	.writepages = nfs_writepages,
586 	.write_begin = nfs_write_begin,
587 	.write_end = nfs_write_end,
588 	.invalidatepage = nfs_invalidate_page,
589 	.releasepage = nfs_release_page,
590 	.direct_IO = nfs_direct_IO,
591 	.migratepage = nfs_migrate_page,
592 	.launder_page = nfs_launder_page,
593 	.is_dirty_writeback = nfs_check_dirty_writeback,
594 	.error_remove_page = generic_error_remove_page,
595 #ifdef CONFIG_NFS_SWAP
596 	.swap_activate = nfs_swap_activate,
597 	.swap_deactivate = nfs_swap_deactivate,
598 #endif
599 };
600 
601 /*
602  * Notification that a PTE pointing to an NFS page is about to be made
603  * writable, implying that someone is about to modify the page through a
604  * shared-writable mapping
605  */
606 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
607 {
608 	struct page *page = vmf->page;
609 	struct file *filp = vma->vm_file;
610 	struct inode *inode = file_inode(filp);
611 	unsigned pagelen;
612 	int ret = VM_FAULT_NOPAGE;
613 	struct address_space *mapping;
614 
615 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
616 		filp, filp->f_mapping->host->i_ino,
617 		(long long)page_offset(page));
618 
619 	/* make sure the cache has finished storing the page */
620 	nfs_fscache_wait_on_page_write(NFS_I(inode), page);
621 
622 	lock_page(page);
623 	mapping = page_file_mapping(page);
624 	if (mapping != inode->i_mapping)
625 		goto out_unlock;
626 
627 	wait_on_page_writeback(page);
628 
629 	pagelen = nfs_page_length(page);
630 	if (pagelen == 0)
631 		goto out_unlock;
632 
633 	ret = VM_FAULT_LOCKED;
634 	if (nfs_flush_incompatible(filp, page) == 0 &&
635 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
636 		goto out;
637 
638 	ret = VM_FAULT_SIGBUS;
639 out_unlock:
640 	unlock_page(page);
641 out:
642 	return ret;
643 }
644 
645 static const struct vm_operations_struct nfs_file_vm_ops = {
646 	.fault = filemap_fault,
647 	.map_pages = filemap_map_pages,
648 	.page_mkwrite = nfs_vm_page_mkwrite,
649 };
650 
651 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
652 {
653 	struct nfs_open_context *ctx;
654 
655 	if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
656 		return 1;
657 	ctx = nfs_file_open_context(filp);
658 	if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
659 	    nfs_ctx_key_to_expire(ctx))
660 		return 1;
661 	return 0;
662 }
663 
664 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
665 {
666 	struct file *file = iocb->ki_filp;
667 	struct inode *inode = file_inode(file);
668 	unsigned long written = 0;
669 	ssize_t result;
670 	size_t count = iov_iter_count(from);
671 	loff_t pos = iocb->ki_pos;
672 
673 	result = nfs_key_timeout_notify(file, inode);
674 	if (result)
675 		return result;
676 
677 	if (file->f_flags & O_DIRECT)
678 		return nfs_file_direct_write(iocb, from, pos);
679 
680 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
681 		file, count, (long long) pos);
682 
683 	result = -EBUSY;
684 	if (IS_SWAPFILE(inode))
685 		goto out_swapfile;
686 	/*
687 	 * O_APPEND implies that we must revalidate the file length.
688 	 */
689 	if (file->f_flags & O_APPEND) {
690 		result = nfs_revalidate_file_size(inode, file);
691 		if (result)
692 			goto out;
693 	}
694 
695 	result = count;
696 	if (!count)
697 		goto out;
698 
699 	result = generic_file_write_iter(iocb, from);
700 	if (result > 0)
701 		written = result;
702 
703 	/* Return error values for O_DSYNC and IS_SYNC() */
704 	if (result >= 0 && nfs_need_sync_write(file, inode)) {
705 		int err = vfs_fsync(file, 0);
706 		if (err < 0)
707 			result = err;
708 	}
709 	if (result > 0)
710 		nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
711 out:
712 	return result;
713 
714 out_swapfile:
715 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
716 	goto out;
717 }
718 EXPORT_SYMBOL_GPL(nfs_file_write);
719 
720 static int
721 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
722 {
723 	struct inode *inode = filp->f_mapping->host;
724 	int status = 0;
725 	unsigned int saved_type = fl->fl_type;
726 
727 	/* Try local locking first */
728 	posix_test_lock(filp, fl);
729 	if (fl->fl_type != F_UNLCK) {
730 		/* found a conflict */
731 		goto out;
732 	}
733 	fl->fl_type = saved_type;
734 
735 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
736 		goto out_noconflict;
737 
738 	if (is_local)
739 		goto out_noconflict;
740 
741 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
742 out:
743 	return status;
744 out_noconflict:
745 	fl->fl_type = F_UNLCK;
746 	goto out;
747 }
748 
749 static int do_vfs_lock(struct file *file, struct file_lock *fl)
750 {
751 	int res = 0;
752 	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
753 		case FL_POSIX:
754 			res = posix_lock_file_wait(file, fl);
755 			break;
756 		case FL_FLOCK:
757 			res = flock_lock_file_wait(file, fl);
758 			break;
759 		default:
760 			BUG();
761 	}
762 	return res;
763 }
764 
765 static int
766 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
767 {
768 	struct inode *inode = filp->f_mapping->host;
769 	struct nfs_lock_context *l_ctx;
770 	int status;
771 
772 	/*
773 	 * Flush all pending writes before doing anything
774 	 * with locks..
775 	 */
776 	nfs_sync_mapping(filp->f_mapping);
777 
778 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
779 	if (!IS_ERR(l_ctx)) {
780 		status = nfs_iocounter_wait(&l_ctx->io_count);
781 		nfs_put_lock_context(l_ctx);
782 		if (status < 0)
783 			return status;
784 	}
785 
786 	/* NOTE: special case
787 	 * 	If we're signalled while cleaning up locks on process exit, we
788 	 * 	still need to complete the unlock.
789 	 */
790 	/*
791 	 * Use local locking if mounted with "-onolock" or with appropriate
792 	 * "-olocal_lock="
793 	 */
794 	if (!is_local)
795 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
796 	else
797 		status = do_vfs_lock(filp, fl);
798 	return status;
799 }
800 
801 static int
802 is_time_granular(struct timespec *ts) {
803 	return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
804 }
805 
806 static int
807 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
808 {
809 	struct inode *inode = filp->f_mapping->host;
810 	int status;
811 
812 	/*
813 	 * Flush all pending writes before doing anything
814 	 * with locks..
815 	 */
816 	status = nfs_sync_mapping(filp->f_mapping);
817 	if (status != 0)
818 		goto out;
819 
820 	/*
821 	 * Use local locking if mounted with "-onolock" or with appropriate
822 	 * "-olocal_lock="
823 	 */
824 	if (!is_local)
825 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
826 	else
827 		status = do_vfs_lock(filp, fl);
828 	if (status < 0)
829 		goto out;
830 
831 	/*
832 	 * Revalidate the cache if the server has time stamps granular
833 	 * enough to detect subsecond changes.  Otherwise, clear the
834 	 * cache to prevent missing any changes.
835 	 *
836 	 * This makes locking act as a cache coherency point.
837 	 */
838 	nfs_sync_mapping(filp->f_mapping);
839 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
840 		if (is_time_granular(&NFS_SERVER(inode)->time_delta))
841 			__nfs_revalidate_inode(NFS_SERVER(inode), inode);
842 		else
843 			nfs_zap_caches(inode);
844 	}
845 out:
846 	return status;
847 }
848 
849 /*
850  * Lock a (portion of) a file
851  */
852 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
853 {
854 	struct inode *inode = filp->f_mapping->host;
855 	int ret = -ENOLCK;
856 	int is_local = 0;
857 
858 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
859 			filp, fl->fl_type, fl->fl_flags,
860 			(long long)fl->fl_start, (long long)fl->fl_end);
861 
862 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
863 
864 	/* No mandatory locks over NFS */
865 	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
866 		goto out_err;
867 
868 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
869 		is_local = 1;
870 
871 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
872 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
873 		if (ret < 0)
874 			goto out_err;
875 	}
876 
877 	if (IS_GETLK(cmd))
878 		ret = do_getlk(filp, cmd, fl, is_local);
879 	else if (fl->fl_type == F_UNLCK)
880 		ret = do_unlk(filp, cmd, fl, is_local);
881 	else
882 		ret = do_setlk(filp, cmd, fl, is_local);
883 out_err:
884 	return ret;
885 }
886 EXPORT_SYMBOL_GPL(nfs_lock);
887 
888 /*
889  * Lock a (portion of) a file
890  */
891 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
892 {
893 	struct inode *inode = filp->f_mapping->host;
894 	int is_local = 0;
895 
896 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
897 			filp, fl->fl_type, fl->fl_flags);
898 
899 	if (!(fl->fl_flags & FL_FLOCK))
900 		return -ENOLCK;
901 
902 	/*
903 	 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
904 	 * any standard. In principle we might be able to support LOCK_MAND
905 	 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
906 	 * NFS code is not set up for it.
907 	 */
908 	if (fl->fl_type & LOCK_MAND)
909 		return -EINVAL;
910 
911 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
912 		is_local = 1;
913 
914 	/* We're simulating flock() locks using posix locks on the server */
915 	if (fl->fl_type == F_UNLCK)
916 		return do_unlk(filp, cmd, fl, is_local);
917 	return do_setlk(filp, cmd, fl, is_local);
918 }
919 EXPORT_SYMBOL_GPL(nfs_flock);
920 
921 const struct file_operations nfs_file_operations = {
922 	.llseek		= nfs_file_llseek,
923 	.read		= new_sync_read,
924 	.write		= new_sync_write,
925 	.read_iter	= nfs_file_read,
926 	.write_iter	= nfs_file_write,
927 	.mmap		= nfs_file_mmap,
928 	.open		= nfs_file_open,
929 	.flush		= nfs_file_flush,
930 	.release	= nfs_file_release,
931 	.fsync		= nfs_file_fsync,
932 	.lock		= nfs_lock,
933 	.flock		= nfs_flock,
934 	.splice_read	= nfs_file_splice_read,
935 	.splice_write	= iter_file_splice_write,
936 	.check_flags	= nfs_check_flags,
937 	.setlease	= simple_nosetlease,
938 };
939 EXPORT_SYMBOL_GPL(nfs_file_operations);
940