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/pagemap.h>
28 #include <linux/aio.h>
29 #include <linux/gfp.h>
30 #include <linux/swap.h>
31 
32 #include <asm/uaccess.h>
33 
34 #include "delegation.h"
35 #include "internal.h"
36 #include "iostat.h"
37 #include "fscache.h"
38 #include "pnfs.h"
39 
40 #define NFSDBG_FACILITY		NFSDBG_FILE
41 
42 static const struct vm_operations_struct nfs_file_vm_ops;
43 
44 const struct inode_operations nfs_file_inode_operations = {
45 	.permission	= nfs_permission,
46 	.getattr	= nfs_getattr,
47 	.setattr	= nfs_setattr,
48 };
49 
50 #ifdef CONFIG_NFS_V3
51 const struct inode_operations nfs3_file_inode_operations = {
52 	.permission	= nfs_permission,
53 	.getattr	= nfs_getattr,
54 	.setattr	= nfs_setattr,
55 	.listxattr	= nfs3_listxattr,
56 	.getxattr	= nfs3_getxattr,
57 	.setxattr	= nfs3_setxattr,
58 	.removexattr	= nfs3_removexattr,
59 };
60 #endif  /* CONFIG_NFS_v3 */
61 
62 /* Hack for future NFS swap support */
63 #ifndef IS_SWAPFILE
64 # define IS_SWAPFILE(inode)	(0)
65 #endif
66 
nfs_check_flags(int flags)67 static int nfs_check_flags(int flags)
68 {
69 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
70 		return -EINVAL;
71 
72 	return 0;
73 }
74 
75 /*
76  * Open file
77  */
78 static int
nfs_file_open(struct inode * inode,struct file * filp)79 nfs_file_open(struct inode *inode, struct file *filp)
80 {
81 	int res;
82 
83 	dprintk("NFS: open file(%s/%s)\n",
84 			filp->f_path.dentry->d_parent->d_name.name,
85 			filp->f_path.dentry->d_name.name);
86 
87 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
88 	res = nfs_check_flags(filp->f_flags);
89 	if (res)
90 		return res;
91 
92 	res = nfs_open(inode, filp);
93 	return res;
94 }
95 
96 static int
nfs_file_release(struct inode * inode,struct file * filp)97 nfs_file_release(struct inode *inode, struct file *filp)
98 {
99 	dprintk("NFS: release(%s/%s)\n",
100 			filp->f_path.dentry->d_parent->d_name.name,
101 			filp->f_path.dentry->d_name.name);
102 
103 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
104 	return nfs_release(inode, filp);
105 }
106 
107 /**
108  * nfs_revalidate_size - Revalidate the file size
109  * @inode - pointer to inode struct
110  * @file - pointer to struct file
111  *
112  * Revalidates the file length. This is basically a wrapper around
113  * nfs_revalidate_inode() that takes into account the fact that we may
114  * have cached writes (in which case we don't care about the server's
115  * idea of what the file length is), or O_DIRECT (in which case we
116  * shouldn't trust the cache).
117  */
nfs_revalidate_file_size(struct inode * inode,struct file * filp)118 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
119 {
120 	struct nfs_server *server = NFS_SERVER(inode);
121 	struct nfs_inode *nfsi = NFS_I(inode);
122 
123 	if (nfs_have_delegated_attributes(inode))
124 		goto out_noreval;
125 
126 	if (filp->f_flags & O_DIRECT)
127 		goto force_reval;
128 	if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
129 		goto force_reval;
130 	if (nfs_attribute_timeout(inode))
131 		goto force_reval;
132 out_noreval:
133 	return 0;
134 force_reval:
135 	return __nfs_revalidate_inode(server, inode);
136 }
137 
nfs_file_llseek(struct file * filp,loff_t offset,int origin)138 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
139 {
140 	dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
141 			filp->f_path.dentry->d_parent->d_name.name,
142 			filp->f_path.dentry->d_name.name,
143 			offset, origin);
144 
145 	/*
146 	 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
147 	 * the cached file length
148 	 */
149 	if (origin != SEEK_SET && origin != SEEK_CUR) {
150 		struct inode *inode = filp->f_mapping->host;
151 
152 		int retval = nfs_revalidate_file_size(inode, filp);
153 		if (retval < 0)
154 			return (loff_t)retval;
155 	}
156 
157 	return generic_file_llseek(filp, offset, origin);
158 }
159 
160 /*
161  * Flush all dirty pages, and check for write errors.
162  */
163 static int
nfs_file_flush(struct file * file,fl_owner_t id)164 nfs_file_flush(struct file *file, fl_owner_t id)
165 {
166 	struct dentry	*dentry = file->f_path.dentry;
167 	struct inode	*inode = dentry->d_inode;
168 
169 	dprintk("NFS: flush(%s/%s)\n",
170 			dentry->d_parent->d_name.name,
171 			dentry->d_name.name);
172 
173 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
174 	if ((file->f_mode & FMODE_WRITE) == 0)
175 		return 0;
176 
177 	/* Flush writes to the server and return any errors */
178 	return vfs_fsync(file, 0);
179 }
180 
181 static ssize_t
nfs_file_read(struct kiocb * iocb,const struct iovec * iov,unsigned long nr_segs,loff_t pos)182 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
183 		unsigned long nr_segs, loff_t pos)
184 {
185 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
186 	struct inode * inode = dentry->d_inode;
187 	ssize_t result;
188 
189 	if (iocb->ki_filp->f_flags & O_DIRECT)
190 		return nfs_file_direct_read(iocb, iov, nr_segs, pos);
191 
192 	dprintk("NFS: read(%s/%s, %lu@%lu)\n",
193 		dentry->d_parent->d_name.name, dentry->d_name.name,
194 		(unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
195 
196 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
197 	if (!result) {
198 		result = generic_file_aio_read(iocb, iov, nr_segs, pos);
199 		if (result > 0)
200 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
201 	}
202 	return result;
203 }
204 
205 static ssize_t
nfs_file_splice_read(struct file * filp,loff_t * ppos,struct pipe_inode_info * pipe,size_t count,unsigned int flags)206 nfs_file_splice_read(struct file *filp, loff_t *ppos,
207 		     struct pipe_inode_info *pipe, size_t count,
208 		     unsigned int flags)
209 {
210 	struct dentry *dentry = filp->f_path.dentry;
211 	struct inode *inode = dentry->d_inode;
212 	ssize_t res;
213 
214 	dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
215 		dentry->d_parent->d_name.name, dentry->d_name.name,
216 		(unsigned long) count, (unsigned long long) *ppos);
217 
218 	res = nfs_revalidate_mapping(inode, filp->f_mapping);
219 	if (!res) {
220 		res = generic_file_splice_read(filp, ppos, pipe, count, flags);
221 		if (res > 0)
222 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
223 	}
224 	return res;
225 }
226 
227 static int
nfs_file_mmap(struct file * file,struct vm_area_struct * vma)228 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
229 {
230 	struct dentry *dentry = file->f_path.dentry;
231 	struct inode *inode = dentry->d_inode;
232 	int	status;
233 
234 	dprintk("NFS: mmap(%s/%s)\n",
235 		dentry->d_parent->d_name.name, dentry->d_name.name);
236 
237 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
238 	 *       so we call that before revalidating the mapping
239 	 */
240 	status = generic_file_mmap(file, vma);
241 	if (!status) {
242 		vma->vm_ops = &nfs_file_vm_ops;
243 		status = nfs_revalidate_mapping(inode, file->f_mapping);
244 	}
245 	return status;
246 }
247 
248 /*
249  * Flush any dirty pages for this process, and check for write errors.
250  * The return status from this call provides a reliable indication of
251  * whether any write errors occurred for this process.
252  *
253  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
254  * disk, but it retrieves and clears ctx->error after synching, despite
255  * the two being set at the same time in nfs_context_set_write_error().
256  * This is because the former is used to notify the _next_ call to
257  * nfs_file_write() that a write error occurred, and hence cause it to
258  * fall back to doing a synchronous write.
259  */
260 static int
nfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)261 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
262 {
263 	struct dentry *dentry = file->f_path.dentry;
264 	struct nfs_open_context *ctx = nfs_file_open_context(file);
265 	struct inode *inode = dentry->d_inode;
266 	int have_error, status;
267 	int ret = 0;
268 
269 	dprintk("NFS: fsync file(%s/%s) datasync %d\n",
270 			dentry->d_parent->d_name.name, dentry->d_name.name,
271 			datasync);
272 
273 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
274 	mutex_lock(&inode->i_mutex);
275 
276 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
277 	have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
278 	status = nfs_commit_inode(inode, FLUSH_SYNC);
279 	if (status >= 0 && ret < 0)
280 		status = ret;
281 	have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
282 	if (have_error)
283 		ret = xchg(&ctx->error, 0);
284 	if (!ret && status < 0)
285 		ret = status;
286 	if (!ret && !datasync)
287 		/* application has asked for meta-data sync */
288 		ret = pnfs_layoutcommit_inode(inode, true);
289 	mutex_unlock(&inode->i_mutex);
290 	return ret;
291 }
292 
293 /*
294  * Decide whether a read/modify/write cycle may be more efficient
295  * then a modify/write/read cycle when writing to a page in the
296  * page cache.
297  *
298  * The modify/write/read cycle may occur if a page is read before
299  * being completely filled by the writer.  In this situation, the
300  * page must be completely written to stable storage on the server
301  * before it can be refilled by reading in the page from the server.
302  * This can lead to expensive, small, FILE_SYNC mode writes being
303  * done.
304  *
305  * It may be more efficient to read the page first if the file is
306  * open for reading in addition to writing, the page is not marked
307  * as Uptodate, it is not dirty or waiting to be committed,
308  * indicating that it was previously allocated and then modified,
309  * that there were valid bytes of data in that range of the file,
310  * and that the new data won't completely replace the old data in
311  * that range of the file.
312  */
nfs_want_read_modify_write(struct file * file,struct page * page,loff_t pos,unsigned len)313 static int nfs_want_read_modify_write(struct file *file, struct page *page,
314 			loff_t pos, unsigned len)
315 {
316 	unsigned int pglen = nfs_page_length(page);
317 	unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
318 	unsigned int end = offset + len;
319 
320 	if ((file->f_mode & FMODE_READ) &&	/* open for read? */
321 	    !PageUptodate(page) &&		/* Uptodate? */
322 	    !PagePrivate(page) &&		/* i/o request already? */
323 	    pglen &&				/* valid bytes of file? */
324 	    (end < pglen || offset))		/* replace all valid bytes? */
325 		return 1;
326 	return 0;
327 }
328 
329 /*
330  * This does the "real" work of the write. We must allocate and lock the
331  * page to be sent back to the generic routine, which then copies the
332  * data from user space.
333  *
334  * If the writer ends up delaying the write, the writer needs to
335  * increment the page use counts until he is done with the page.
336  */
nfs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)337 static int nfs_write_begin(struct file *file, struct address_space *mapping,
338 			loff_t pos, unsigned len, unsigned flags,
339 			struct page **pagep, void **fsdata)
340 {
341 	int ret;
342 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
343 	struct page *page;
344 	int once_thru = 0;
345 
346 	dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
347 		file->f_path.dentry->d_parent->d_name.name,
348 		file->f_path.dentry->d_name.name,
349 		mapping->host->i_ino, len, (long long) pos);
350 
351 start:
352 	/*
353 	 * Prevent starvation issues if someone is doing a consistency
354 	 * sync-to-disk
355 	 */
356 	ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
357 			nfs_wait_bit_killable, TASK_KILLABLE);
358 	if (ret)
359 		return ret;
360 
361 	page = grab_cache_page_write_begin(mapping, index, flags);
362 	if (!page)
363 		return -ENOMEM;
364 	*pagep = page;
365 
366 	ret = nfs_flush_incompatible(file, page);
367 	if (ret) {
368 		unlock_page(page);
369 		page_cache_release(page);
370 	} else if (!once_thru &&
371 		   nfs_want_read_modify_write(file, page, pos, len)) {
372 		once_thru = 1;
373 		ret = nfs_readpage(file, page);
374 		page_cache_release(page);
375 		if (!ret)
376 			goto start;
377 	}
378 	return ret;
379 }
380 
nfs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)381 static int nfs_write_end(struct file *file, struct address_space *mapping,
382 			loff_t pos, unsigned len, unsigned copied,
383 			struct page *page, void *fsdata)
384 {
385 	unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
386 	int status;
387 
388 	dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
389 		file->f_path.dentry->d_parent->d_name.name,
390 		file->f_path.dentry->d_name.name,
391 		mapping->host->i_ino, len, (long long) pos);
392 
393 	/*
394 	 * Zero any uninitialised parts of the page, and then mark the page
395 	 * as up to date if it turns out that we're extending the file.
396 	 */
397 	if (!PageUptodate(page)) {
398 		unsigned pglen = nfs_page_length(page);
399 		unsigned end = offset + len;
400 
401 		if (pglen == 0) {
402 			zero_user_segments(page, 0, offset,
403 					end, PAGE_CACHE_SIZE);
404 			SetPageUptodate(page);
405 		} else if (end >= pglen) {
406 			zero_user_segment(page, end, PAGE_CACHE_SIZE);
407 			if (offset == 0)
408 				SetPageUptodate(page);
409 		} else
410 			zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
411 	}
412 
413 	status = nfs_updatepage(file, page, offset, copied);
414 
415 	unlock_page(page);
416 	page_cache_release(page);
417 
418 	if (status < 0)
419 		return status;
420 	return copied;
421 }
422 
423 /*
424  * Partially or wholly invalidate a page
425  * - Release the private state associated with a page if undergoing complete
426  *   page invalidation
427  * - Called if either PG_private or PG_fscache is set on the page
428  * - Caller holds page lock
429  */
nfs_invalidate_page(struct page * page,unsigned long offset)430 static void nfs_invalidate_page(struct page *page, unsigned long offset)
431 {
432 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
433 
434 	if (offset != 0)
435 		return;
436 	/* Cancel any unstarted writes on this page */
437 	nfs_wb_page_cancel(page->mapping->host, page);
438 
439 	nfs_fscache_invalidate_page(page, page->mapping->host);
440 }
441 
442 /*
443  * Attempt to release the private state associated with a page
444  * - Called if either PG_private or PG_fscache is set on the page
445  * - Caller holds page lock
446  * - Return true (may release page) or false (may not)
447  */
nfs_release_page(struct page * page,gfp_t gfp)448 static int nfs_release_page(struct page *page, gfp_t gfp)
449 {
450 	struct address_space *mapping = page->mapping;
451 
452 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
453 
454 	/* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
455 	 * doing this memory reclaim for a fs-related allocation.
456 	 */
457 	if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
458 	    !(current->flags & PF_FSTRANS)) {
459 		int how = FLUSH_SYNC;
460 
461 		/* Don't let kswapd deadlock waiting for OOM RPC calls */
462 		if (current_is_kswapd())
463 			how = 0;
464 		nfs_commit_inode(mapping->host, how);
465 	}
466 	/* If PagePrivate() is set, then the page is not freeable */
467 	if (PagePrivate(page))
468 		return 0;
469 	return nfs_fscache_release_page(page, gfp);
470 }
471 
472 /*
473  * Attempt to clear the private state associated with a page when an error
474  * occurs that requires the cached contents of an inode to be written back or
475  * destroyed
476  * - Called if either PG_private or fscache is set on the page
477  * - Caller holds page lock
478  * - Return 0 if successful, -error otherwise
479  */
nfs_launder_page(struct page * page)480 static int nfs_launder_page(struct page *page)
481 {
482 	struct inode *inode = page->mapping->host;
483 	struct nfs_inode *nfsi = NFS_I(inode);
484 
485 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
486 		inode->i_ino, (long long)page_offset(page));
487 
488 	nfs_fscache_wait_on_page_write(nfsi, page);
489 	return nfs_wb_page(inode, page);
490 }
491 
492 const struct address_space_operations nfs_file_aops = {
493 	.readpage = nfs_readpage,
494 	.readpages = nfs_readpages,
495 	.set_page_dirty = __set_page_dirty_nobuffers,
496 	.writepage = nfs_writepage,
497 	.writepages = nfs_writepages,
498 	.write_begin = nfs_write_begin,
499 	.write_end = nfs_write_end,
500 	.invalidatepage = nfs_invalidate_page,
501 	.releasepage = nfs_release_page,
502 	.direct_IO = nfs_direct_IO,
503 	.migratepage = nfs_migrate_page,
504 	.launder_page = nfs_launder_page,
505 	.error_remove_page = generic_error_remove_page,
506 };
507 
508 /*
509  * Notification that a PTE pointing to an NFS page is about to be made
510  * writable, implying that someone is about to modify the page through a
511  * shared-writable mapping
512  */
nfs_vm_page_mkwrite(struct vm_area_struct * vma,struct vm_fault * vmf)513 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
514 {
515 	struct page *page = vmf->page;
516 	struct file *filp = vma->vm_file;
517 	struct dentry *dentry = filp->f_path.dentry;
518 	unsigned pagelen;
519 	int ret = VM_FAULT_NOPAGE;
520 	struct address_space *mapping;
521 
522 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
523 		dentry->d_parent->d_name.name, dentry->d_name.name,
524 		filp->f_mapping->host->i_ino,
525 		(long long)page_offset(page));
526 
527 	/* make sure the cache has finished storing the page */
528 	nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
529 
530 	lock_page(page);
531 	mapping = page->mapping;
532 	if (mapping != dentry->d_inode->i_mapping)
533 		goto out_unlock;
534 
535 	wait_on_page_writeback(page);
536 
537 	pagelen = nfs_page_length(page);
538 	if (pagelen == 0)
539 		goto out_unlock;
540 
541 	ret = VM_FAULT_LOCKED;
542 	if (nfs_flush_incompatible(filp, page) == 0 &&
543 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
544 		goto out;
545 
546 	ret = VM_FAULT_SIGBUS;
547 out_unlock:
548 	unlock_page(page);
549 out:
550 	return ret;
551 }
552 
553 static const struct vm_operations_struct nfs_file_vm_ops = {
554 	.fault = filemap_fault,
555 	.page_mkwrite = nfs_vm_page_mkwrite,
556 };
557 
nfs_need_sync_write(struct file * filp,struct inode * inode)558 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
559 {
560 	struct nfs_open_context *ctx;
561 
562 	if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
563 		return 1;
564 	ctx = nfs_file_open_context(filp);
565 	if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
566 		return 1;
567 	return 0;
568 }
569 
nfs_file_write(struct kiocb * iocb,const struct iovec * iov,unsigned long nr_segs,loff_t pos)570 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
571 				unsigned long nr_segs, loff_t pos)
572 {
573 	struct dentry * dentry = iocb->ki_filp->f_path.dentry;
574 	struct inode * inode = dentry->d_inode;
575 	unsigned long written = 0;
576 	ssize_t result;
577 	size_t count = iov_length(iov, nr_segs);
578 
579 	if (iocb->ki_filp->f_flags & O_DIRECT)
580 		return nfs_file_direct_write(iocb, iov, nr_segs, pos);
581 
582 	dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
583 		dentry->d_parent->d_name.name, dentry->d_name.name,
584 		(unsigned long) count, (long long) pos);
585 
586 	result = -EBUSY;
587 	if (IS_SWAPFILE(inode))
588 		goto out_swapfile;
589 	/*
590 	 * O_APPEND implies that we must revalidate the file length.
591 	 */
592 	if (iocb->ki_filp->f_flags & O_APPEND) {
593 		result = nfs_revalidate_file_size(inode, iocb->ki_filp);
594 		if (result)
595 			goto out;
596 	}
597 
598 	result = count;
599 	if (!count)
600 		goto out;
601 
602 	result = generic_file_aio_write(iocb, iov, nr_segs, pos);
603 	if (result > 0)
604 		written = result;
605 
606 	/* Return error values for O_DSYNC and IS_SYNC() */
607 	if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
608 		int err = vfs_fsync(iocb->ki_filp, 0);
609 		if (err < 0)
610 			result = err;
611 	}
612 	if (result > 0)
613 		nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
614 out:
615 	return result;
616 
617 out_swapfile:
618 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
619 	goto out;
620 }
621 
nfs_file_splice_write(struct pipe_inode_info * pipe,struct file * filp,loff_t * ppos,size_t count,unsigned int flags)622 static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
623 				     struct file *filp, loff_t *ppos,
624 				     size_t count, unsigned int flags)
625 {
626 	struct dentry *dentry = filp->f_path.dentry;
627 	struct inode *inode = dentry->d_inode;
628 	unsigned long written = 0;
629 	ssize_t ret;
630 
631 	dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
632 		dentry->d_parent->d_name.name, dentry->d_name.name,
633 		(unsigned long) count, (unsigned long long) *ppos);
634 
635 	/*
636 	 * The combination of splice and an O_APPEND destination is disallowed.
637 	 */
638 
639 	ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
640 	if (ret > 0)
641 		written = ret;
642 
643 	if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
644 		int err = vfs_fsync(filp, 0);
645 		if (err < 0)
646 			ret = err;
647 	}
648 	if (ret > 0)
649 		nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
650 	return ret;
651 }
652 
653 static int
do_getlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)654 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
655 {
656 	struct inode *inode = filp->f_mapping->host;
657 	int status = 0;
658 	unsigned int saved_type = fl->fl_type;
659 
660 	/* Try local locking first */
661 	posix_test_lock(filp, fl);
662 	if (fl->fl_type != F_UNLCK) {
663 		/* found a conflict */
664 		goto out;
665 	}
666 	fl->fl_type = saved_type;
667 
668 	if (nfs_have_delegation(inode, FMODE_READ))
669 		goto out_noconflict;
670 
671 	if (is_local)
672 		goto out_noconflict;
673 
674 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
675 out:
676 	return status;
677 out_noconflict:
678 	fl->fl_type = F_UNLCK;
679 	goto out;
680 }
681 
do_vfs_lock(struct file * file,struct file_lock * fl)682 static int do_vfs_lock(struct file *file, struct file_lock *fl)
683 {
684 	int res = 0;
685 	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
686 		case FL_POSIX:
687 			res = posix_lock_file_wait(file, fl);
688 			break;
689 		case FL_FLOCK:
690 			res = flock_lock_file_wait(file, fl);
691 			break;
692 		default:
693 			BUG();
694 	}
695 	return res;
696 }
697 
698 static int
do_unlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)699 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
700 {
701 	struct inode *inode = filp->f_mapping->host;
702 	int status;
703 
704 	/*
705 	 * Flush all pending writes before doing anything
706 	 * with locks..
707 	 */
708 	nfs_sync_mapping(filp->f_mapping);
709 
710 	/* NOTE: special case
711 	 * 	If we're signalled while cleaning up locks on process exit, we
712 	 * 	still need to complete the unlock.
713 	 */
714 	/*
715 	 * Use local locking if mounted with "-onolock" or with appropriate
716 	 * "-olocal_lock="
717 	 */
718 	if (!is_local)
719 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
720 	else
721 		status = do_vfs_lock(filp, fl);
722 	return status;
723 }
724 
725 static int
is_time_granular(struct timespec * ts)726 is_time_granular(struct timespec *ts) {
727 	return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
728 }
729 
730 static int
do_setlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)731 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
732 {
733 	struct inode *inode = filp->f_mapping->host;
734 	int status;
735 
736 	/*
737 	 * Flush all pending writes before doing anything
738 	 * with locks..
739 	 */
740 	status = nfs_sync_mapping(filp->f_mapping);
741 	if (status != 0)
742 		goto out;
743 
744 	/*
745 	 * Use local locking if mounted with "-onolock" or with appropriate
746 	 * "-olocal_lock="
747 	 */
748 	if (!is_local)
749 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
750 	else
751 		status = do_vfs_lock(filp, fl);
752 	if (status < 0)
753 		goto out;
754 
755 	/*
756 	 * Revalidate the cache if the server has time stamps granular
757 	 * enough to detect subsecond changes.  Otherwise, clear the
758 	 * cache to prevent missing any changes.
759 	 *
760 	 * This makes locking act as a cache coherency point.
761 	 */
762 	nfs_sync_mapping(filp->f_mapping);
763 	if (!nfs_have_delegation(inode, FMODE_READ)) {
764 		if (is_time_granular(&NFS_SERVER(inode)->time_delta))
765 			__nfs_revalidate_inode(NFS_SERVER(inode), inode);
766 		else
767 			nfs_zap_caches(inode);
768 	}
769 out:
770 	return status;
771 }
772 
773 /*
774  * Lock a (portion of) a file
775  */
nfs_lock(struct file * filp,int cmd,struct file_lock * fl)776 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
777 {
778 	struct inode *inode = filp->f_mapping->host;
779 	int ret = -ENOLCK;
780 	int is_local = 0;
781 
782 	dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
783 			filp->f_path.dentry->d_parent->d_name.name,
784 			filp->f_path.dentry->d_name.name,
785 			fl->fl_type, fl->fl_flags,
786 			(long long)fl->fl_start, (long long)fl->fl_end);
787 
788 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
789 
790 	/* No mandatory locks over NFS */
791 	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
792 		goto out_err;
793 
794 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
795 		is_local = 1;
796 
797 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
798 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
799 		if (ret < 0)
800 			goto out_err;
801 	}
802 
803 	if (IS_GETLK(cmd))
804 		ret = do_getlk(filp, cmd, fl, is_local);
805 	else if (fl->fl_type == F_UNLCK)
806 		ret = do_unlk(filp, cmd, fl, is_local);
807 	else
808 		ret = do_setlk(filp, cmd, fl, is_local);
809 out_err:
810 	return ret;
811 }
812 
813 /*
814  * Lock a (portion of) a file
815  */
nfs_flock(struct file * filp,int cmd,struct file_lock * fl)816 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
817 {
818 	struct inode *inode = filp->f_mapping->host;
819 	int is_local = 0;
820 
821 	dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
822 			filp->f_path.dentry->d_parent->d_name.name,
823 			filp->f_path.dentry->d_name.name,
824 			fl->fl_type, fl->fl_flags);
825 
826 	if (!(fl->fl_flags & FL_FLOCK))
827 		return -ENOLCK;
828 
829 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
830 		is_local = 1;
831 
832 	/* We're simulating flock() locks using posix locks on the server */
833 	fl->fl_owner = (fl_owner_t)filp;
834 	fl->fl_start = 0;
835 	fl->fl_end = OFFSET_MAX;
836 
837 	if (fl->fl_type == F_UNLCK)
838 		return do_unlk(filp, cmd, fl, is_local);
839 	return do_setlk(filp, cmd, fl, is_local);
840 }
841 
842 /*
843  * There is no protocol support for leases, so we have no way to implement
844  * them correctly in the face of opens by other clients.
845  */
nfs_setlease(struct file * file,long arg,struct file_lock ** fl)846 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
847 {
848 	dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
849 			file->f_path.dentry->d_parent->d_name.name,
850 			file->f_path.dentry->d_name.name, arg);
851 	return -EINVAL;
852 }
853 
854 const struct file_operations nfs_file_operations = {
855 	.llseek		= nfs_file_llseek,
856 	.read		= do_sync_read,
857 	.write		= do_sync_write,
858 	.aio_read	= nfs_file_read,
859 	.aio_write	= nfs_file_write,
860 	.mmap		= nfs_file_mmap,
861 	.open		= nfs_file_open,
862 	.flush		= nfs_file_flush,
863 	.release	= nfs_file_release,
864 	.fsync		= nfs_file_fsync,
865 	.lock		= nfs_lock,
866 	.flock		= nfs_flock,
867 	.splice_read	= nfs_file_splice_read,
868 	.splice_write	= nfs_file_splice_write,
869 	.check_flags	= nfs_check_flags,
870 	.setlease	= nfs_setlease,
871 };
872 
873 #ifdef CONFIG_NFS_V4
874 static int
nfs4_file_open(struct inode * inode,struct file * filp)875 nfs4_file_open(struct inode *inode, struct file *filp)
876 {
877 	/*
878 	 * NFSv4 opens are handled in d_lookup and d_revalidate. If we get to
879 	 * this point, then something is very wrong
880 	 */
881 	dprintk("NFS: %s called! inode=%p filp=%p\n", __func__, inode, filp);
882 	return -ENOTDIR;
883 }
884 
885 const struct file_operations nfs4_file_operations = {
886 	.llseek		= nfs_file_llseek,
887 	.read		= do_sync_read,
888 	.write		= do_sync_write,
889 	.aio_read	= nfs_file_read,
890 	.aio_write	= nfs_file_write,
891 	.mmap		= nfs_file_mmap,
892 	.open		= nfs4_file_open,
893 	.flush		= nfs_file_flush,
894 	.release	= nfs_file_release,
895 	.fsync		= nfs_file_fsync,
896 	.lock		= nfs_lock,
897 	.flock		= nfs_flock,
898 	.splice_read	= nfs_file_splice_read,
899 	.splice_write	= nfs_file_splice_write,
900 	.check_flags	= nfs_check_flags,
901 	.setlease	= nfs_setlease,
902 };
903 #endif /* CONFIG_NFS_V4 */
904