xref: /openbmc/linux/fs/nfs/file.c (revision b6dcefde)
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/time.h>
20 #include <linux/kernel.h>
21 #include <linux/errno.h>
22 #include <linux/fcntl.h>
23 #include <linux/stat.h>
24 #include <linux/nfs_fs.h>
25 #include <linux/nfs_mount.h>
26 #include <linux/mm.h>
27 #include <linux/slab.h>
28 #include <linux/pagemap.h>
29 #include <linux/aio.h>
30 
31 #include <asm/uaccess.h>
32 #include <asm/system.h>
33 
34 #include "delegation.h"
35 #include "internal.h"
36 #include "iostat.h"
37 #include "fscache.h"
38 
39 #define NFSDBG_FACILITY		NFSDBG_FILE
40 
41 static int nfs_file_open(struct inode *, struct file *);
42 static int nfs_file_release(struct inode *, struct file *);
43 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
44 static int  nfs_file_mmap(struct file *, struct vm_area_struct *);
45 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
46 					struct pipe_inode_info *pipe,
47 					size_t count, unsigned int flags);
48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
49 				unsigned long nr_segs, loff_t pos);
50 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
51 					struct file *filp, loff_t *ppos,
52 					size_t count, unsigned int flags);
53 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
54 				unsigned long nr_segs, loff_t pos);
55 static int  nfs_file_flush(struct file *, fl_owner_t id);
56 static int  nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
57 static int nfs_check_flags(int flags);
58 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
59 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
60 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
61 
62 static const struct vm_operations_struct nfs_file_vm_ops;
63 
64 const struct file_operations nfs_file_operations = {
65 	.llseek		= nfs_file_llseek,
66 	.read		= do_sync_read,
67 	.write		= do_sync_write,
68 	.aio_read	= nfs_file_read,
69 	.aio_write	= nfs_file_write,
70 	.mmap		= nfs_file_mmap,
71 	.open		= nfs_file_open,
72 	.flush		= nfs_file_flush,
73 	.release	= nfs_file_release,
74 	.fsync		= nfs_file_fsync,
75 	.lock		= nfs_lock,
76 	.flock		= nfs_flock,
77 	.splice_read	= nfs_file_splice_read,
78 	.splice_write	= nfs_file_splice_write,
79 	.check_flags	= nfs_check_flags,
80 	.setlease	= nfs_setlease,
81 };
82 
83 const struct inode_operations nfs_file_inode_operations = {
84 	.permission	= nfs_permission,
85 	.getattr	= nfs_getattr,
86 	.setattr	= nfs_setattr,
87 };
88 
89 #ifdef CONFIG_NFS_V3
90 const struct inode_operations nfs3_file_inode_operations = {
91 	.permission	= nfs_permission,
92 	.getattr	= nfs_getattr,
93 	.setattr	= nfs_setattr,
94 	.listxattr	= nfs3_listxattr,
95 	.getxattr	= nfs3_getxattr,
96 	.setxattr	= nfs3_setxattr,
97 	.removexattr	= nfs3_removexattr,
98 };
99 #endif  /* CONFIG_NFS_v3 */
100 
101 /* Hack for future NFS swap support */
102 #ifndef IS_SWAPFILE
103 # define IS_SWAPFILE(inode)	(0)
104 #endif
105 
106 static int nfs_check_flags(int flags)
107 {
108 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
109 		return -EINVAL;
110 
111 	return 0;
112 }
113 
114 /*
115  * Open file
116  */
117 static int
118 nfs_file_open(struct inode *inode, struct file *filp)
119 {
120 	int res;
121 
122 	dprintk("NFS: open file(%s/%s)\n",
123 			filp->f_path.dentry->d_parent->d_name.name,
124 			filp->f_path.dentry->d_name.name);
125 
126 	res = nfs_check_flags(filp->f_flags);
127 	if (res)
128 		return res;
129 
130 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
131 	res = nfs_open(inode, filp);
132 	return res;
133 }
134 
135 static int
136 nfs_file_release(struct inode *inode, struct file *filp)
137 {
138 	struct dentry *dentry = filp->f_path.dentry;
139 
140 	dprintk("NFS: release(%s/%s)\n",
141 			dentry->d_parent->d_name.name,
142 			dentry->d_name.name);
143 
144 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
145 	return nfs_release(inode, filp);
146 }
147 
148 /**
149  * nfs_revalidate_size - Revalidate the file size
150  * @inode - pointer to inode struct
151  * @file - pointer to struct file
152  *
153  * Revalidates the file length. This is basically a wrapper around
154  * nfs_revalidate_inode() that takes into account the fact that we may
155  * have cached writes (in which case we don't care about the server's
156  * idea of what the file length is), or O_DIRECT (in which case we
157  * shouldn't trust the cache).
158  */
159 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
160 {
161 	struct nfs_server *server = NFS_SERVER(inode);
162 	struct nfs_inode *nfsi = NFS_I(inode);
163 
164 	if (server->flags & NFS_MOUNT_NOAC)
165 		goto force_reval;
166 	if (filp->f_flags & O_DIRECT)
167 		goto force_reval;
168 	if (nfsi->npages != 0)
169 		return 0;
170 	if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
171 		return 0;
172 force_reval:
173 	return __nfs_revalidate_inode(server, inode);
174 }
175 
176 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
177 {
178 	loff_t loff;
179 
180 	dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
181 			filp->f_path.dentry->d_parent->d_name.name,
182 			filp->f_path.dentry->d_name.name,
183 			offset, origin);
184 
185 	/* origin == SEEK_END => we must revalidate the cached file length */
186 	if (origin == SEEK_END) {
187 		struct inode *inode = filp->f_mapping->host;
188 
189 		int retval = nfs_revalidate_file_size(inode, filp);
190 		if (retval < 0)
191 			return (loff_t)retval;
192 
193 		spin_lock(&inode->i_lock);
194 		loff = generic_file_llseek_unlocked(filp, offset, origin);
195 		spin_unlock(&inode->i_lock);
196 	} else
197 		loff = generic_file_llseek_unlocked(filp, offset, origin);
198 	return loff;
199 }
200 
201 /*
202  * Helper for nfs_file_flush() and nfs_file_fsync()
203  *
204  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
205  * disk, but it retrieves and clears ctx->error after synching, despite
206  * the two being set at the same time in nfs_context_set_write_error().
207  * This is because the former is used to notify the _next_ call to
208  * nfs_file_write() that a write error occured, and hence cause it to
209  * fall back to doing a synchronous write.
210  */
211 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
212 {
213 	int have_error, status;
214 	int ret = 0;
215 
216 	have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
217 	status = nfs_wb_all(inode);
218 	have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
219 	if (have_error)
220 		ret = xchg(&ctx->error, 0);
221 	if (!ret)
222 		ret = status;
223 	return ret;
224 }
225 
226 /*
227  * Flush all dirty pages, and check for write errors.
228  */
229 static int
230 nfs_file_flush(struct file *file, fl_owner_t id)
231 {
232 	struct nfs_open_context *ctx = nfs_file_open_context(file);
233 	struct dentry	*dentry = file->f_path.dentry;
234 	struct inode	*inode = dentry->d_inode;
235 
236 	dprintk("NFS: flush(%s/%s)\n",
237 			dentry->d_parent->d_name.name,
238 			dentry->d_name.name);
239 
240 	if ((file->f_mode & FMODE_WRITE) == 0)
241 		return 0;
242 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
243 
244 	/* Flush writes to the server and return any errors */
245 	return nfs_do_fsync(ctx, inode);
246 }
247 
248 static ssize_t
249 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
250 		unsigned long nr_segs, loff_t pos)
251 {
252 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
253 	struct inode * inode = dentry->d_inode;
254 	ssize_t result;
255 	size_t count = iov_length(iov, nr_segs);
256 
257 	if (iocb->ki_filp->f_flags & O_DIRECT)
258 		return nfs_file_direct_read(iocb, iov, nr_segs, pos);
259 
260 	dprintk("NFS: read(%s/%s, %lu@%lu)\n",
261 		dentry->d_parent->d_name.name, dentry->d_name.name,
262 		(unsigned long) count, (unsigned long) pos);
263 
264 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
265 	nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
266 	if (!result)
267 		result = generic_file_aio_read(iocb, iov, nr_segs, pos);
268 	return result;
269 }
270 
271 static ssize_t
272 nfs_file_splice_read(struct file *filp, loff_t *ppos,
273 		     struct pipe_inode_info *pipe, size_t count,
274 		     unsigned int flags)
275 {
276 	struct dentry *dentry = filp->f_path.dentry;
277 	struct inode *inode = dentry->d_inode;
278 	ssize_t res;
279 
280 	dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
281 		dentry->d_parent->d_name.name, dentry->d_name.name,
282 		(unsigned long) count, (unsigned long long) *ppos);
283 
284 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
285 	if (!res)
286 		res = generic_file_splice_read(filp, ppos, pipe, count, flags);
287 	return res;
288 }
289 
290 static int
291 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
292 {
293 	struct dentry *dentry = file->f_path.dentry;
294 	struct inode *inode = dentry->d_inode;
295 	int	status;
296 
297 	dprintk("NFS: mmap(%s/%s)\n",
298 		dentry->d_parent->d_name.name, dentry->d_name.name);
299 
300 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
301 	 *       so we call that before revalidating the mapping
302 	 */
303 	status = generic_file_mmap(file, vma);
304 	if (!status) {
305 		vma->vm_ops = &nfs_file_vm_ops;
306 		status = nfs_revalidate_mapping(inode, file->f_mapping);
307 	}
308 	return status;
309 }
310 
311 /*
312  * Flush any dirty pages for this process, and check for write errors.
313  * The return status from this call provides a reliable indication of
314  * whether any write errors occurred for this process.
315  */
316 static int
317 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
318 {
319 	struct nfs_open_context *ctx = nfs_file_open_context(file);
320 	struct inode *inode = dentry->d_inode;
321 
322 	dprintk("NFS: fsync file(%s/%s) datasync %d\n",
323 			dentry->d_parent->d_name.name, dentry->d_name.name,
324 			datasync);
325 
326 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
327 	return nfs_do_fsync(ctx, inode);
328 }
329 
330 /*
331  * Decide whether a read/modify/write cycle may be more efficient
332  * then a modify/write/read cycle when writing to a page in the
333  * page cache.
334  *
335  * The modify/write/read cycle may occur if a page is read before
336  * being completely filled by the writer.  In this situation, the
337  * page must be completely written to stable storage on the server
338  * before it can be refilled by reading in the page from the server.
339  * This can lead to expensive, small, FILE_SYNC mode writes being
340  * done.
341  *
342  * It may be more efficient to read the page first if the file is
343  * open for reading in addition to writing, the page is not marked
344  * as Uptodate, it is not dirty or waiting to be committed,
345  * indicating that it was previously allocated and then modified,
346  * that there were valid bytes of data in that range of the file,
347  * and that the new data won't completely replace the old data in
348  * that range of the file.
349  */
350 static int nfs_want_read_modify_write(struct file *file, struct page *page,
351 			loff_t pos, unsigned len)
352 {
353 	unsigned int pglen = nfs_page_length(page);
354 	unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
355 	unsigned int end = offset + len;
356 
357 	if ((file->f_mode & FMODE_READ) &&	/* open for read? */
358 	    !PageUptodate(page) &&		/* Uptodate? */
359 	    !PagePrivate(page) &&		/* i/o request already? */
360 	    pglen &&				/* valid bytes of file? */
361 	    (end < pglen || offset))		/* replace all valid bytes? */
362 		return 1;
363 	return 0;
364 }
365 
366 /*
367  * This does the "real" work of the write. We must allocate and lock the
368  * page to be sent back to the generic routine, which then copies the
369  * data from user space.
370  *
371  * If the writer ends up delaying the write, the writer needs to
372  * increment the page use counts until he is done with the page.
373  */
374 static int nfs_write_begin(struct file *file, struct address_space *mapping,
375 			loff_t pos, unsigned len, unsigned flags,
376 			struct page **pagep, void **fsdata)
377 {
378 	int ret;
379 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
380 	struct page *page;
381 	int once_thru = 0;
382 
383 	dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
384 		file->f_path.dentry->d_parent->d_name.name,
385 		file->f_path.dentry->d_name.name,
386 		mapping->host->i_ino, len, (long long) pos);
387 
388 start:
389 	/*
390 	 * Prevent starvation issues if someone is doing a consistency
391 	 * sync-to-disk
392 	 */
393 	ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
394 			nfs_wait_bit_killable, TASK_KILLABLE);
395 	if (ret)
396 		return ret;
397 
398 	page = grab_cache_page_write_begin(mapping, index, flags);
399 	if (!page)
400 		return -ENOMEM;
401 	*pagep = page;
402 
403 	ret = nfs_flush_incompatible(file, page);
404 	if (ret) {
405 		unlock_page(page);
406 		page_cache_release(page);
407 	} else if (!once_thru &&
408 		   nfs_want_read_modify_write(file, page, pos, len)) {
409 		once_thru = 1;
410 		ret = nfs_readpage(file, page);
411 		page_cache_release(page);
412 		if (!ret)
413 			goto start;
414 	}
415 	return ret;
416 }
417 
418 static int nfs_write_end(struct file *file, struct address_space *mapping,
419 			loff_t pos, unsigned len, unsigned copied,
420 			struct page *page, void *fsdata)
421 {
422 	unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
423 	int status;
424 
425 	dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
426 		file->f_path.dentry->d_parent->d_name.name,
427 		file->f_path.dentry->d_name.name,
428 		mapping->host->i_ino, len, (long long) pos);
429 
430 	/*
431 	 * Zero any uninitialised parts of the page, and then mark the page
432 	 * as up to date if it turns out that we're extending the file.
433 	 */
434 	if (!PageUptodate(page)) {
435 		unsigned pglen = nfs_page_length(page);
436 		unsigned end = offset + len;
437 
438 		if (pglen == 0) {
439 			zero_user_segments(page, 0, offset,
440 					end, PAGE_CACHE_SIZE);
441 			SetPageUptodate(page);
442 		} else if (end >= pglen) {
443 			zero_user_segment(page, end, PAGE_CACHE_SIZE);
444 			if (offset == 0)
445 				SetPageUptodate(page);
446 		} else
447 			zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
448 	}
449 
450 	status = nfs_updatepage(file, page, offset, copied);
451 
452 	unlock_page(page);
453 	page_cache_release(page);
454 
455 	if (status < 0)
456 		return status;
457 	return copied;
458 }
459 
460 /*
461  * Partially or wholly invalidate a page
462  * - Release the private state associated with a page if undergoing complete
463  *   page invalidation
464  * - Called if either PG_private or PG_fscache is set on the page
465  * - Caller holds page lock
466  */
467 static void nfs_invalidate_page(struct page *page, unsigned long offset)
468 {
469 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
470 
471 	if (offset != 0)
472 		return;
473 	/* Cancel any unstarted writes on this page */
474 	nfs_wb_page_cancel(page->mapping->host, page);
475 
476 	nfs_fscache_invalidate_page(page, page->mapping->host);
477 }
478 
479 /*
480  * Attempt to release the private state associated with a page
481  * - Called if either PG_private or PG_fscache is set on the page
482  * - Caller holds page lock
483  * - Return true (may release page) or false (may not)
484  */
485 static int nfs_release_page(struct page *page, gfp_t gfp)
486 {
487 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
488 
489 	if (gfp & __GFP_WAIT)
490 		nfs_wb_page(page->mapping->host, page);
491 	/* If PagePrivate() is set, then the page is not freeable */
492 	if (PagePrivate(page))
493 		return 0;
494 	return nfs_fscache_release_page(page, gfp);
495 }
496 
497 /*
498  * Attempt to clear the private state associated with a page when an error
499  * occurs that requires the cached contents of an inode to be written back or
500  * destroyed
501  * - Called if either PG_private or fscache is set on the page
502  * - Caller holds page lock
503  * - Return 0 if successful, -error otherwise
504  */
505 static int nfs_launder_page(struct page *page)
506 {
507 	struct inode *inode = page->mapping->host;
508 	struct nfs_inode *nfsi = NFS_I(inode);
509 
510 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
511 		inode->i_ino, (long long)page_offset(page));
512 
513 	nfs_fscache_wait_on_page_write(nfsi, page);
514 	return nfs_wb_page(inode, page);
515 }
516 
517 const struct address_space_operations nfs_file_aops = {
518 	.readpage = nfs_readpage,
519 	.readpages = nfs_readpages,
520 	.set_page_dirty = __set_page_dirty_nobuffers,
521 	.writepage = nfs_writepage,
522 	.writepages = nfs_writepages,
523 	.write_begin = nfs_write_begin,
524 	.write_end = nfs_write_end,
525 	.invalidatepage = nfs_invalidate_page,
526 	.releasepage = nfs_release_page,
527 	.direct_IO = nfs_direct_IO,
528 	.migratepage = nfs_migrate_page,
529 	.launder_page = nfs_launder_page,
530 	.error_remove_page = generic_error_remove_page,
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 = -EINVAL;
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->mapping;
557 	if (mapping != dentry->d_inode->i_mapping)
558 		goto out_unlock;
559 
560 	ret = 0;
561 	pagelen = nfs_page_length(page);
562 	if (pagelen == 0)
563 		goto out_unlock;
564 
565 	ret = nfs_flush_incompatible(filp, page);
566 	if (ret != 0)
567 		goto out_unlock;
568 
569 	ret = nfs_updatepage(filp, page, 0, pagelen);
570 out_unlock:
571 	if (!ret)
572 		return VM_FAULT_LOCKED;
573 	unlock_page(page);
574 	return VM_FAULT_SIGBUS;
575 }
576 
577 static const struct vm_operations_struct nfs_file_vm_ops = {
578 	.fault = filemap_fault,
579 	.page_mkwrite = nfs_vm_page_mkwrite,
580 };
581 
582 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
583 {
584 	struct nfs_open_context *ctx;
585 
586 	if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
587 		return 1;
588 	ctx = nfs_file_open_context(filp);
589 	if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
590 		return 1;
591 	return 0;
592 }
593 
594 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
595 				unsigned long nr_segs, loff_t pos)
596 {
597 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
598 	struct inode * inode = dentry->d_inode;
599 	ssize_t result;
600 	size_t count = iov_length(iov, nr_segs);
601 
602 	if (iocb->ki_filp->f_flags & O_DIRECT)
603 		return nfs_file_direct_write(iocb, iov, nr_segs, pos);
604 
605 	dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
606 		dentry->d_parent->d_name.name, dentry->d_name.name,
607 		(unsigned long) count, (long long) pos);
608 
609 	result = -EBUSY;
610 	if (IS_SWAPFILE(inode))
611 		goto out_swapfile;
612 	/*
613 	 * O_APPEND implies that we must revalidate the file length.
614 	 */
615 	if (iocb->ki_filp->f_flags & O_APPEND) {
616 		result = nfs_revalidate_file_size(inode, iocb->ki_filp);
617 		if (result)
618 			goto out;
619 	}
620 
621 	result = count;
622 	if (!count)
623 		goto out;
624 
625 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
626 	result = generic_file_aio_write(iocb, iov, nr_segs, pos);
627 	/* Return error values for O_DSYNC and IS_SYNC() */
628 	if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
629 		int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
630 		if (err < 0)
631 			result = err;
632 	}
633 out:
634 	return result;
635 
636 out_swapfile:
637 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
638 	goto out;
639 }
640 
641 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
642 				     struct file *filp, loff_t *ppos,
643 				     size_t count, unsigned int flags)
644 {
645 	struct dentry *dentry = filp->f_path.dentry;
646 	struct inode *inode = dentry->d_inode;
647 	ssize_t ret;
648 
649 	dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
650 		dentry->d_parent->d_name.name, dentry->d_name.name,
651 		(unsigned long) count, (unsigned long long) *ppos);
652 
653 	/*
654 	 * The combination of splice and an O_APPEND destination is disallowed.
655 	 */
656 
657 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
658 
659 	ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
660 	if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
661 		int err = nfs_do_fsync(nfs_file_open_context(filp), inode);
662 		if (err < 0)
663 			ret = err;
664 	}
665 	return ret;
666 }
667 
668 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
669 {
670 	struct inode *inode = filp->f_mapping->host;
671 	int status = 0;
672 
673 	/* Try local locking first */
674 	posix_test_lock(filp, fl);
675 	if (fl->fl_type != F_UNLCK) {
676 		/* found a conflict */
677 		goto out;
678 	}
679 
680 	if (nfs_have_delegation(inode, FMODE_READ))
681 		goto out_noconflict;
682 
683 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
684 		goto out_noconflict;
685 
686 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
687 out:
688 	return status;
689 out_noconflict:
690 	fl->fl_type = F_UNLCK;
691 	goto out;
692 }
693 
694 static int do_vfs_lock(struct file *file, struct file_lock *fl)
695 {
696 	int res = 0;
697 	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
698 		case FL_POSIX:
699 			res = posix_lock_file_wait(file, fl);
700 			break;
701 		case FL_FLOCK:
702 			res = flock_lock_file_wait(file, fl);
703 			break;
704 		default:
705 			BUG();
706 	}
707 	if (res < 0)
708 		dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
709 			" - error %d!\n",
710 				__func__, res);
711 	return res;
712 }
713 
714 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
715 {
716 	struct inode *inode = filp->f_mapping->host;
717 	int status;
718 
719 	/*
720 	 * Flush all pending writes before doing anything
721 	 * with locks..
722 	 */
723 	nfs_sync_mapping(filp->f_mapping);
724 
725 	/* NOTE: special case
726 	 * 	If we're signalled while cleaning up locks on process exit, we
727 	 * 	still need to complete the unlock.
728 	 */
729 	/* Use local locking if mounted with "-onolock" */
730 	if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
731 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
732 	else
733 		status = do_vfs_lock(filp, fl);
734 	return status;
735 }
736 
737 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
738 {
739 	struct inode *inode = filp->f_mapping->host;
740 	int status;
741 
742 	/*
743 	 * Flush all pending writes before doing anything
744 	 * with locks..
745 	 */
746 	status = nfs_sync_mapping(filp->f_mapping);
747 	if (status != 0)
748 		goto out;
749 
750 	/* Use local locking if mounted with "-onolock" */
751 	if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
752 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
753 	else
754 		status = do_vfs_lock(filp, fl);
755 	if (status < 0)
756 		goto out;
757 	/*
758 	 * Make sure we clear the cache whenever we try to get the lock.
759 	 * This makes locking act as a cache coherency point.
760 	 */
761 	nfs_sync_mapping(filp->f_mapping);
762 	if (!nfs_have_delegation(inode, FMODE_READ))
763 		nfs_zap_caches(inode);
764 out:
765 	return status;
766 }
767 
768 /*
769  * Lock a (portion of) a file
770  */
771 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
772 {
773 	struct inode *inode = filp->f_mapping->host;
774 	int ret = -ENOLCK;
775 
776 	dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
777 			filp->f_path.dentry->d_parent->d_name.name,
778 			filp->f_path.dentry->d_name.name,
779 			fl->fl_type, fl->fl_flags,
780 			(long long)fl->fl_start, (long long)fl->fl_end);
781 
782 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
783 
784 	/* No mandatory locks over NFS */
785 	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
786 		goto out_err;
787 
788 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
789 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
790 		if (ret < 0)
791 			goto out_err;
792 	}
793 
794 	if (IS_GETLK(cmd))
795 		ret = do_getlk(filp, cmd, fl);
796 	else if (fl->fl_type == F_UNLCK)
797 		ret = do_unlk(filp, cmd, fl);
798 	else
799 		ret = do_setlk(filp, cmd, fl);
800 out_err:
801 	return ret;
802 }
803 
804 /*
805  * Lock a (portion of) a file
806  */
807 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
808 {
809 	dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\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 
814 	if (!(fl->fl_flags & FL_FLOCK))
815 		return -ENOLCK;
816 
817 	/* We're simulating flock() locks using posix locks on the server */
818 	fl->fl_owner = (fl_owner_t)filp;
819 	fl->fl_start = 0;
820 	fl->fl_end = OFFSET_MAX;
821 
822 	if (fl->fl_type == F_UNLCK)
823 		return do_unlk(filp, cmd, fl);
824 	return do_setlk(filp, cmd, fl);
825 }
826 
827 /*
828  * There is no protocol support for leases, so we have no way to implement
829  * them correctly in the face of opens by other clients.
830  */
831 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
832 {
833 	dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
834 			file->f_path.dentry->d_parent->d_name.name,
835 			file->f_path.dentry->d_name.name, arg);
836 
837 	return -EINVAL;
838 }
839