xref: /linux/fs/nfs/file.c (revision 4413e16d9d21673bb5048a2e542f1aaa00015c2e)
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 
40 #define NFSDBG_FACILITY		NFSDBG_FILE
41 
42 static const struct vm_operations_struct nfs_file_vm_ops;
43 
44 /* Hack for future NFS swap support */
45 #ifndef IS_SWAPFILE
46 # define IS_SWAPFILE(inode)	(0)
47 #endif
48 
49 int nfs_check_flags(int flags)
50 {
51 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
52 		return -EINVAL;
53 
54 	return 0;
55 }
56 EXPORT_SYMBOL_GPL(nfs_check_flags);
57 
58 /*
59  * Open file
60  */
61 static int
62 nfs_file_open(struct inode *inode, struct file *filp)
63 {
64 	int res;
65 
66 	dprintk("NFS: open file(%s/%s)\n",
67 			filp->f_path.dentry->d_parent->d_name.name,
68 			filp->f_path.dentry->d_name.name);
69 
70 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 	res = nfs_check_flags(filp->f_flags);
72 	if (res)
73 		return res;
74 
75 	res = nfs_open(inode, filp);
76 	return res;
77 }
78 
79 int
80 nfs_file_release(struct inode *inode, struct file *filp)
81 {
82 	dprintk("NFS: release(%s/%s)\n",
83 			filp->f_path.dentry->d_parent->d_name.name,
84 			filp->f_path.dentry->d_name.name);
85 
86 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
87 	return nfs_release(inode, filp);
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  * @file - 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 	struct nfs_inode *nfsi = NFS_I(inode);
106 
107 	if (nfs_have_delegated_attributes(inode))
108 		goto out_noreval;
109 
110 	if (filp->f_flags & O_DIRECT)
111 		goto force_reval;
112 	if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
113 		goto force_reval;
114 	if (nfs_attribute_timeout(inode))
115 		goto force_reval;
116 out_noreval:
117 	return 0;
118 force_reval:
119 	return __nfs_revalidate_inode(server, inode);
120 }
121 
122 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
123 {
124 	dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
125 			filp->f_path.dentry->d_parent->d_name.name,
126 			filp->f_path.dentry->d_name.name,
127 			offset, origin);
128 
129 	/*
130 	 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
131 	 * the cached file length
132 	 */
133 	if (origin != SEEK_SET && origin != SEEK_CUR) {
134 		struct inode *inode = filp->f_mapping->host;
135 
136 		int retval = nfs_revalidate_file_size(inode, filp);
137 		if (retval < 0)
138 			return (loff_t)retval;
139 	}
140 
141 	return generic_file_llseek(filp, offset, origin);
142 }
143 EXPORT_SYMBOL_GPL(nfs_file_llseek);
144 
145 /*
146  * Flush all dirty pages, and check for write errors.
147  */
148 int
149 nfs_file_flush(struct file *file, fl_owner_t id)
150 {
151 	struct dentry	*dentry = file->f_path.dentry;
152 	struct inode	*inode = dentry->d_inode;
153 
154 	dprintk("NFS: flush(%s/%s)\n",
155 			dentry->d_parent->d_name.name,
156 			dentry->d_name.name);
157 
158 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
159 	if ((file->f_mode & FMODE_WRITE) == 0)
160 		return 0;
161 
162 	/*
163 	 * If we're holding a write delegation, then just start the i/o
164 	 * but don't wait for completion (or send a commit).
165 	 */
166 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
167 		return filemap_fdatawrite(file->f_mapping);
168 
169 	/* Flush writes to the server and return any errors */
170 	return vfs_fsync(file, 0);
171 }
172 EXPORT_SYMBOL_GPL(nfs_file_flush);
173 
174 ssize_t
175 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
176 		unsigned long nr_segs, loff_t pos)
177 {
178 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
179 	struct inode * inode = dentry->d_inode;
180 	ssize_t result;
181 
182 	if (iocb->ki_filp->f_flags & O_DIRECT)
183 		return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
184 
185 	dprintk("NFS: read(%s/%s, %lu@%lu)\n",
186 		dentry->d_parent->d_name.name, dentry->d_name.name,
187 		(unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
188 
189 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
190 	if (!result) {
191 		result = generic_file_aio_read(iocb, iov, nr_segs, pos);
192 		if (result > 0)
193 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
194 	}
195 	return result;
196 }
197 EXPORT_SYMBOL_GPL(nfs_file_read);
198 
199 ssize_t
200 nfs_file_splice_read(struct file *filp, loff_t *ppos,
201 		     struct pipe_inode_info *pipe, size_t count,
202 		     unsigned int flags)
203 {
204 	struct dentry *dentry = filp->f_path.dentry;
205 	struct inode *inode = dentry->d_inode;
206 	ssize_t res;
207 
208 	dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
209 		dentry->d_parent->d_name.name, dentry->d_name.name,
210 		(unsigned long) count, (unsigned long long) *ppos);
211 
212 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
213 	if (!res) {
214 		res = generic_file_splice_read(filp, ppos, pipe, count, flags);
215 		if (res > 0)
216 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
217 	}
218 	return res;
219 }
220 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
221 
222 int
223 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
224 {
225 	struct dentry *dentry = file->f_path.dentry;
226 	struct inode *inode = dentry->d_inode;
227 	int	status;
228 
229 	dprintk("NFS: mmap(%s/%s)\n",
230 		dentry->d_parent->d_name.name, dentry->d_name.name);
231 
232 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
233 	 *       so we call that before revalidating the mapping
234 	 */
235 	status = generic_file_mmap(file, vma);
236 	if (!status) {
237 		vma->vm_ops = &nfs_file_vm_ops;
238 		status = nfs_revalidate_mapping(inode, file->f_mapping);
239 	}
240 	return status;
241 }
242 EXPORT_SYMBOL_GPL(nfs_file_mmap);
243 
244 /*
245  * Flush any dirty pages for this process, and check for write errors.
246  * The return status from this call provides a reliable indication of
247  * whether any write errors occurred for this process.
248  *
249  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
250  * disk, but it retrieves and clears ctx->error after synching, despite
251  * the two being set at the same time in nfs_context_set_write_error().
252  * This is because the former is used to notify the _next_ call to
253  * nfs_file_write() that a write error occurred, and hence cause it to
254  * fall back to doing a synchronous write.
255  */
256 int
257 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
258 {
259 	struct dentry *dentry = file->f_path.dentry;
260 	struct nfs_open_context *ctx = nfs_file_open_context(file);
261 	struct inode *inode = dentry->d_inode;
262 	int have_error, status;
263 	int ret = 0;
264 
265 	dprintk("NFS: fsync file(%s/%s) datasync %d\n",
266 			dentry->d_parent->d_name.name, dentry->d_name.name,
267 			datasync);
268 
269 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
270 	have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
271 	status = nfs_commit_inode(inode, FLUSH_SYNC);
272 	if (status >= 0 && ret < 0)
273 		status = ret;
274 	have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
275 	if (have_error)
276 		ret = xchg(&ctx->error, 0);
277 	if (!ret && status < 0)
278 		ret = status;
279 	return ret;
280 }
281 EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
282 
283 static int
284 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
285 {
286 	int ret;
287 	struct inode *inode = file->f_path.dentry->d_inode;
288 
289 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
290 	if (ret != 0)
291 		goto out;
292 	mutex_lock(&inode->i_mutex);
293 	ret = nfs_file_fsync_commit(file, start, end, datasync);
294 	mutex_unlock(&inode->i_mutex);
295 out:
296 	return ret;
297 }
298 
299 /*
300  * Decide whether a read/modify/write cycle may be more efficient
301  * then a modify/write/read cycle when writing to a page in the
302  * page cache.
303  *
304  * The modify/write/read cycle may occur if a page is read before
305  * being completely filled by the writer.  In this situation, the
306  * page must be completely written to stable storage on the server
307  * before it can be refilled by reading in the page from the server.
308  * This can lead to expensive, small, FILE_SYNC mode writes being
309  * done.
310  *
311  * It may be more efficient to read the page first if the file is
312  * open for reading in addition to writing, the page is not marked
313  * as Uptodate, it is not dirty or waiting to be committed,
314  * indicating that it was previously allocated and then modified,
315  * that there were valid bytes of data in that range of the file,
316  * and that the new data won't completely replace the old data in
317  * that range of the file.
318  */
319 static int nfs_want_read_modify_write(struct file *file, struct page *page,
320 			loff_t pos, unsigned len)
321 {
322 	unsigned int pglen = nfs_page_length(page);
323 	unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
324 	unsigned int end = offset + len;
325 
326 	if ((file->f_mode & FMODE_READ) &&	/* open for read? */
327 	    !PageUptodate(page) &&		/* Uptodate? */
328 	    !PagePrivate(page) &&		/* i/o request already? */
329 	    pglen &&				/* valid bytes of file? */
330 	    (end < pglen || offset))		/* replace all valid bytes? */
331 		return 1;
332 	return 0;
333 }
334 
335 /*
336  * This does the "real" work of the write. We must allocate and lock the
337  * page to be sent back to the generic routine, which then copies the
338  * data from user space.
339  *
340  * If the writer ends up delaying the write, the writer needs to
341  * increment the page use counts until he is done with the page.
342  */
343 static int nfs_write_begin(struct file *file, struct address_space *mapping,
344 			loff_t pos, unsigned len, unsigned flags,
345 			struct page **pagep, void **fsdata)
346 {
347 	int ret;
348 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
349 	struct page *page;
350 	int once_thru = 0;
351 
352 	dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
353 		file->f_path.dentry->d_parent->d_name.name,
354 		file->f_path.dentry->d_name.name,
355 		mapping->host->i_ino, len, (long long) pos);
356 
357 start:
358 	/*
359 	 * Prevent starvation issues if someone is doing a consistency
360 	 * sync-to-disk
361 	 */
362 	ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
363 			nfs_wait_bit_killable, TASK_KILLABLE);
364 	if (ret)
365 		return ret;
366 
367 	page = grab_cache_page_write_begin(mapping, index, flags);
368 	if (!page)
369 		return -ENOMEM;
370 	*pagep = page;
371 
372 	ret = nfs_flush_incompatible(file, page);
373 	if (ret) {
374 		unlock_page(page);
375 		page_cache_release(page);
376 	} else if (!once_thru &&
377 		   nfs_want_read_modify_write(file, page, pos, len)) {
378 		once_thru = 1;
379 		ret = nfs_readpage(file, page);
380 		page_cache_release(page);
381 		if (!ret)
382 			goto start;
383 	}
384 	return ret;
385 }
386 
387 static int nfs_write_end(struct file *file, struct address_space *mapping,
388 			loff_t pos, unsigned len, unsigned copied,
389 			struct page *page, void *fsdata)
390 {
391 	unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
392 	int status;
393 
394 	dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
395 		file->f_path.dentry->d_parent->d_name.name,
396 		file->f_path.dentry->d_name.name,
397 		mapping->host->i_ino, len, (long long) pos);
398 
399 	/*
400 	 * Zero any uninitialised parts of the page, and then mark the page
401 	 * as up to date if it turns out that we're extending the file.
402 	 */
403 	if (!PageUptodate(page)) {
404 		unsigned pglen = nfs_page_length(page);
405 		unsigned end = offset + len;
406 
407 		if (pglen == 0) {
408 			zero_user_segments(page, 0, offset,
409 					end, PAGE_CACHE_SIZE);
410 			SetPageUptodate(page);
411 		} else if (end >= pglen) {
412 			zero_user_segment(page, end, PAGE_CACHE_SIZE);
413 			if (offset == 0)
414 				SetPageUptodate(page);
415 		} else
416 			zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
417 	}
418 
419 	status = nfs_updatepage(file, page, offset, copied);
420 
421 	unlock_page(page);
422 	page_cache_release(page);
423 
424 	if (status < 0)
425 		return status;
426 	NFS_I(mapping->host)->write_io += copied;
427 	return copied;
428 }
429 
430 /*
431  * Partially or wholly invalidate a page
432  * - Release the private state associated with a page if undergoing complete
433  *   page invalidation
434  * - Called if either PG_private or PG_fscache is set on the page
435  * - Caller holds page lock
436  */
437 static void nfs_invalidate_page(struct page *page, unsigned long offset)
438 {
439 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
440 
441 	if (offset != 0)
442 		return;
443 	/* Cancel any unstarted writes on this page */
444 	nfs_wb_page_cancel(page_file_mapping(page)->host, page);
445 
446 	nfs_fscache_invalidate_page(page, page->mapping->host);
447 }
448 
449 /*
450  * Attempt to release the private state associated with a page
451  * - Called if either PG_private or PG_fscache is set on the page
452  * - Caller holds page lock
453  * - Return true (may release page) or false (may not)
454  */
455 static int nfs_release_page(struct page *page, gfp_t gfp)
456 {
457 	struct address_space *mapping = page->mapping;
458 
459 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
460 
461 	/* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
462 	 * doing this memory reclaim for a fs-related allocation.
463 	 */
464 	if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
465 	    !(current->flags & PF_FSTRANS)) {
466 		int how = FLUSH_SYNC;
467 
468 		/* Don't let kswapd deadlock waiting for OOM RPC calls */
469 		if (current_is_kswapd())
470 			how = 0;
471 		nfs_commit_inode(mapping->host, how);
472 	}
473 	/* If PagePrivate() is set, then the page is not freeable */
474 	if (PagePrivate(page))
475 		return 0;
476 	return nfs_fscache_release_page(page, gfp);
477 }
478 
479 /*
480  * Attempt to clear the private state associated with a page when an error
481  * occurs that requires the cached contents of an inode to be written back or
482  * destroyed
483  * - Called if either PG_private or fscache is set on the page
484  * - Caller holds page lock
485  * - Return 0 if successful, -error otherwise
486  */
487 static int nfs_launder_page(struct page *page)
488 {
489 	struct inode *inode = page_file_mapping(page)->host;
490 	struct nfs_inode *nfsi = NFS_I(inode);
491 
492 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
493 		inode->i_ino, (long long)page_offset(page));
494 
495 	nfs_fscache_wait_on_page_write(nfsi, page);
496 	return nfs_wb_page(inode, page);
497 }
498 
499 #ifdef CONFIG_NFS_SWAP
500 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
501 						sector_t *span)
502 {
503 	*span = sis->pages;
504 	return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
505 }
506 
507 static void nfs_swap_deactivate(struct file *file)
508 {
509 	xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
510 }
511 #endif
512 
513 const struct address_space_operations nfs_file_aops = {
514 	.readpage = nfs_readpage,
515 	.readpages = nfs_readpages,
516 	.set_page_dirty = __set_page_dirty_nobuffers,
517 	.writepage = nfs_writepage,
518 	.writepages = nfs_writepages,
519 	.write_begin = nfs_write_begin,
520 	.write_end = nfs_write_end,
521 	.invalidatepage = nfs_invalidate_page,
522 	.releasepage = nfs_release_page,
523 	.direct_IO = nfs_direct_IO,
524 	.migratepage = nfs_migrate_page,
525 	.launder_page = nfs_launder_page,
526 	.error_remove_page = generic_error_remove_page,
527 #ifdef CONFIG_NFS_SWAP
528 	.swap_activate = nfs_swap_activate,
529 	.swap_deactivate = nfs_swap_deactivate,
530 #endif
531 };
532 
533 /*
534  * Notification that a PTE pointing to an NFS page is about to be made
535  * writable, implying that someone is about to modify the page through a
536  * shared-writable mapping
537  */
538 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
539 {
540 	struct page *page = vmf->page;
541 	struct file *filp = vma->vm_file;
542 	struct dentry *dentry = filp->f_path.dentry;
543 	unsigned pagelen;
544 	int ret = VM_FAULT_NOPAGE;
545 	struct address_space *mapping;
546 
547 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
548 		dentry->d_parent->d_name.name, dentry->d_name.name,
549 		filp->f_mapping->host->i_ino,
550 		(long long)page_offset(page));
551 
552 	/* make sure the cache has finished storing the page */
553 	nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
554 
555 	lock_page(page);
556 	mapping = page_file_mapping(page);
557 	if (mapping != dentry->d_inode->i_mapping)
558 		goto out_unlock;
559 
560 	wait_on_page_writeback(page);
561 
562 	pagelen = nfs_page_length(page);
563 	if (pagelen == 0)
564 		goto out_unlock;
565 
566 	ret = VM_FAULT_LOCKED;
567 	if (nfs_flush_incompatible(filp, page) == 0 &&
568 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
569 		goto out;
570 
571 	ret = VM_FAULT_SIGBUS;
572 out_unlock:
573 	unlock_page(page);
574 out:
575 	return ret;
576 }
577 
578 static const struct vm_operations_struct nfs_file_vm_ops = {
579 	.fault = filemap_fault,
580 	.page_mkwrite = nfs_vm_page_mkwrite,
581 };
582 
583 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
584 {
585 	struct nfs_open_context *ctx;
586 
587 	if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
588 		return 1;
589 	ctx = nfs_file_open_context(filp);
590 	if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
591 		return 1;
592 	return 0;
593 }
594 
595 ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
596 		       unsigned long nr_segs, loff_t pos)
597 {
598 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
599 	struct inode * inode = dentry->d_inode;
600 	unsigned long written = 0;
601 	ssize_t result;
602 	size_t count = iov_length(iov, nr_segs);
603 
604 	if (iocb->ki_filp->f_flags & O_DIRECT)
605 		return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
606 
607 	dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
608 		dentry->d_parent->d_name.name, dentry->d_name.name,
609 		(unsigned long) count, (long long) pos);
610 
611 	result = -EBUSY;
612 	if (IS_SWAPFILE(inode))
613 		goto out_swapfile;
614 	/*
615 	 * O_APPEND implies that we must revalidate the file length.
616 	 */
617 	if (iocb->ki_filp->f_flags & O_APPEND) {
618 		result = nfs_revalidate_file_size(inode, iocb->ki_filp);
619 		if (result)
620 			goto out;
621 	}
622 
623 	result = count;
624 	if (!count)
625 		goto out;
626 
627 	result = generic_file_aio_write(iocb, iov, nr_segs, pos);
628 	if (result > 0)
629 		written = result;
630 
631 	/* Return error values for O_DSYNC and IS_SYNC() */
632 	if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
633 		int err = vfs_fsync(iocb->ki_filp, 0);
634 		if (err < 0)
635 			result = err;
636 	}
637 	if (result > 0)
638 		nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
639 out:
640 	return result;
641 
642 out_swapfile:
643 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
644 	goto out;
645 }
646 EXPORT_SYMBOL_GPL(nfs_file_write);
647 
648 ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
649 			      struct file *filp, loff_t *ppos,
650 			      size_t count, unsigned int flags)
651 {
652 	struct dentry *dentry = filp->f_path.dentry;
653 	struct inode *inode = dentry->d_inode;
654 	unsigned long written = 0;
655 	ssize_t ret;
656 
657 	dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
658 		dentry->d_parent->d_name.name, dentry->d_name.name,
659 		(unsigned long) count, (unsigned long long) *ppos);
660 
661 	/*
662 	 * The combination of splice and an O_APPEND destination is disallowed.
663 	 */
664 
665 	ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
666 	if (ret > 0)
667 		written = ret;
668 
669 	if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
670 		int err = vfs_fsync(filp, 0);
671 		if (err < 0)
672 			ret = err;
673 	}
674 	if (ret > 0)
675 		nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
676 	return ret;
677 }
678 EXPORT_SYMBOL_GPL(nfs_file_splice_write);
679 
680 static int
681 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
682 {
683 	struct inode *inode = filp->f_mapping->host;
684 	int status = 0;
685 	unsigned int saved_type = fl->fl_type;
686 
687 	/* Try local locking first */
688 	posix_test_lock(filp, fl);
689 	if (fl->fl_type != F_UNLCK) {
690 		/* found a conflict */
691 		goto out;
692 	}
693 	fl->fl_type = saved_type;
694 
695 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
696 		goto out_noconflict;
697 
698 	if (is_local)
699 		goto out_noconflict;
700 
701 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
702 out:
703 	return status;
704 out_noconflict:
705 	fl->fl_type = F_UNLCK;
706 	goto out;
707 }
708 
709 static int do_vfs_lock(struct file *file, struct file_lock *fl)
710 {
711 	int res = 0;
712 	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
713 		case FL_POSIX:
714 			res = posix_lock_file_wait(file, fl);
715 			break;
716 		case FL_FLOCK:
717 			res = flock_lock_file_wait(file, fl);
718 			break;
719 		default:
720 			BUG();
721 	}
722 	return res;
723 }
724 
725 static int
726 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
727 {
728 	struct inode *inode = filp->f_mapping->host;
729 	int status;
730 
731 	/*
732 	 * Flush all pending writes before doing anything
733 	 * with locks..
734 	 */
735 	nfs_sync_mapping(filp->f_mapping);
736 
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 	/*
742 	 * Use local locking if mounted with "-onolock" or with appropriate
743 	 * "-olocal_lock="
744 	 */
745 	if (!is_local)
746 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
747 	else
748 		status = do_vfs_lock(filp, fl);
749 	return status;
750 }
751 
752 static int
753 is_time_granular(struct timespec *ts) {
754 	return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
755 }
756 
757 static int
758 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
759 {
760 	struct inode *inode = filp->f_mapping->host;
761 	int status;
762 
763 	/*
764 	 * Flush all pending writes before doing anything
765 	 * with locks..
766 	 */
767 	status = nfs_sync_mapping(filp->f_mapping);
768 	if (status != 0)
769 		goto out;
770 
771 	/*
772 	 * Use local locking if mounted with "-onolock" or with appropriate
773 	 * "-olocal_lock="
774 	 */
775 	if (!is_local)
776 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
777 	else
778 		status = do_vfs_lock(filp, fl);
779 	if (status < 0)
780 		goto out;
781 
782 	/*
783 	 * Revalidate the cache if the server has time stamps granular
784 	 * enough to detect subsecond changes.  Otherwise, clear the
785 	 * cache to prevent missing any changes.
786 	 *
787 	 * This makes locking act as a cache coherency point.
788 	 */
789 	nfs_sync_mapping(filp->f_mapping);
790 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
791 		if (is_time_granular(&NFS_SERVER(inode)->time_delta))
792 			__nfs_revalidate_inode(NFS_SERVER(inode), inode);
793 		else
794 			nfs_zap_caches(inode);
795 	}
796 out:
797 	return status;
798 }
799 
800 /*
801  * Lock a (portion of) a file
802  */
803 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
804 {
805 	struct inode *inode = filp->f_mapping->host;
806 	int ret = -ENOLCK;
807 	int is_local = 0;
808 
809 	dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
810 			filp->f_path.dentry->d_parent->d_name.name,
811 			filp->f_path.dentry->d_name.name,
812 			fl->fl_type, fl->fl_flags,
813 			(long long)fl->fl_start, (long long)fl->fl_end);
814 
815 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
816 
817 	/* No mandatory locks over NFS */
818 	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
819 		goto out_err;
820 
821 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
822 		is_local = 1;
823 
824 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
825 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
826 		if (ret < 0)
827 			goto out_err;
828 	}
829 
830 	if (IS_GETLK(cmd))
831 		ret = do_getlk(filp, cmd, fl, is_local);
832 	else if (fl->fl_type == F_UNLCK)
833 		ret = do_unlk(filp, cmd, fl, is_local);
834 	else
835 		ret = do_setlk(filp, cmd, fl, is_local);
836 out_err:
837 	return ret;
838 }
839 EXPORT_SYMBOL_GPL(nfs_lock);
840 
841 /*
842  * Lock a (portion of) a file
843  */
844 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
845 {
846 	struct inode *inode = filp->f_mapping->host;
847 	int is_local = 0;
848 
849 	dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
850 			filp->f_path.dentry->d_parent->d_name.name,
851 			filp->f_path.dentry->d_name.name,
852 			fl->fl_type, fl->fl_flags);
853 
854 	if (!(fl->fl_flags & FL_FLOCK))
855 		return -ENOLCK;
856 
857 	/*
858 	 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
859 	 * any standard. In principle we might be able to support LOCK_MAND
860 	 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
861 	 * NFS code is not set up for it.
862 	 */
863 	if (fl->fl_type & LOCK_MAND)
864 		return -EINVAL;
865 
866 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
867 		is_local = 1;
868 
869 	/* We're simulating flock() locks using posix locks on the server */
870 	fl->fl_owner = (fl_owner_t)filp;
871 	fl->fl_start = 0;
872 	fl->fl_end = OFFSET_MAX;
873 
874 	if (fl->fl_type == F_UNLCK)
875 		return do_unlk(filp, cmd, fl, is_local);
876 	return do_setlk(filp, cmd, fl, is_local);
877 }
878 EXPORT_SYMBOL_GPL(nfs_flock);
879 
880 /*
881  * There is no protocol support for leases, so we have no way to implement
882  * them correctly in the face of opens by other clients.
883  */
884 int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
885 {
886 	dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
887 			file->f_path.dentry->d_parent->d_name.name,
888 			file->f_path.dentry->d_name.name, arg);
889 	return -EINVAL;
890 }
891 EXPORT_SYMBOL_GPL(nfs_setlease);
892 
893 const struct file_operations nfs_file_operations = {
894 	.llseek		= nfs_file_llseek,
895 	.read		= do_sync_read,
896 	.write		= do_sync_write,
897 	.aio_read	= nfs_file_read,
898 	.aio_write	= nfs_file_write,
899 	.mmap		= nfs_file_mmap,
900 	.open		= nfs_file_open,
901 	.flush		= nfs_file_flush,
902 	.release	= nfs_file_release,
903 	.fsync		= nfs_file_fsync,
904 	.lock		= nfs_lock,
905 	.flock		= nfs_flock,
906 	.splice_read	= nfs_file_splice_read,
907 	.splice_write	= nfs_file_splice_write,
908 	.check_flags	= nfs_check_flags,
909 	.setlease	= nfs_setlease,
910 };
911 EXPORT_SYMBOL_GPL(nfs_file_operations);
912