• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/file.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  Changes Copyright (C) 1994 by Florian La Roche
8  *   - Do not copy data too often around in the kernel.
9  *   - In nfs_file_read the return value of kmalloc wasn't checked.
10  *   - Put in a better version of read look-ahead buffering. Original idea
11  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
12  *
13  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
14  *
15  *  Total rewrite of read side for new NFS buffer cache.. Linus.
16  *
17  *  nfs regular file handling functions
18  */
19 
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/stat.h>
26 #include <linux/nfs_fs.h>
27 #include <linux/nfs_mount.h>
28 #include <linux/mm.h>
29 #include <linux/pagemap.h>
30 #include <linux/gfp.h>
31 #include <linux/swap.h>
32 
33 #include <linux/uaccess.h>
34 
35 #include "delegation.h"
36 #include "internal.h"
37 #include "iostat.h"
38 #include "fscache.h"
39 #include "pnfs.h"
40 
41 #include "nfstrace.h"
42 
43 #define NFSDBG_FACILITY		NFSDBG_FILE
44 
45 static const struct vm_operations_struct nfs_file_vm_ops;
46 
47 /* Hack for future NFS swap support */
48 #ifndef IS_SWAPFILE
49 # define IS_SWAPFILE(inode)	(0)
50 #endif
51 
nfs_check_flags(int flags)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
nfs_file_open(struct inode * inode,struct file * filp)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
nfs_file_release(struct inode * inode,struct file * filp)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 	nfs_file_clear_open_context(filp);
87 	return 0;
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  * @filp: 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  */
nfs_revalidate_file_size(struct inode * inode,struct file * filp)102 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103 {
104 	struct nfs_server *server = NFS_SERVER(inode);
105 
106 	if (filp->f_flags & O_DIRECT)
107 		goto force_reval;
108 	if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
109 		goto force_reval;
110 	return 0;
111 force_reval:
112 	return __nfs_revalidate_inode(server, inode);
113 }
114 
nfs_file_llseek(struct file * filp,loff_t offset,int whence)115 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
116 {
117 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
118 			filp, offset, whence);
119 
120 	/*
121 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
122 	 * the cached file length
123 	 */
124 	if (whence != SEEK_SET && whence != SEEK_CUR) {
125 		struct inode *inode = filp->f_mapping->host;
126 
127 		int retval = nfs_revalidate_file_size(inode, filp);
128 		if (retval < 0)
129 			return (loff_t)retval;
130 	}
131 
132 	return generic_file_llseek(filp, offset, whence);
133 }
134 EXPORT_SYMBOL_GPL(nfs_file_llseek);
135 
136 /*
137  * Flush all dirty pages, and check for write errors.
138  */
139 static int
nfs_file_flush(struct file * file,fl_owner_t id)140 nfs_file_flush(struct file *file, fl_owner_t id)
141 {
142 	struct inode	*inode = file_inode(file);
143 	errseq_t since;
144 
145 	dprintk("NFS: flush(%pD2)\n", file);
146 
147 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
148 	if ((file->f_mode & FMODE_WRITE) == 0)
149 		return 0;
150 
151 	/* Flush writes to the server and return any errors */
152 	since = filemap_sample_wb_err(file->f_mapping);
153 	nfs_wb_all(inode);
154 	return filemap_check_wb_err(file->f_mapping, since);
155 }
156 
157 ssize_t
nfs_file_read(struct kiocb * iocb,struct iov_iter * to)158 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
159 {
160 	struct inode *inode = file_inode(iocb->ki_filp);
161 	ssize_t result;
162 
163 	if (iocb->ki_flags & IOCB_DIRECT)
164 		return nfs_file_direct_read(iocb, to, false);
165 
166 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
167 		iocb->ki_filp,
168 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
169 
170 	nfs_start_io_read(inode);
171 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
172 	if (!result) {
173 		result = generic_file_read_iter(iocb, to);
174 		if (result > 0)
175 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
176 	}
177 	nfs_end_io_read(inode);
178 	return result;
179 }
180 EXPORT_SYMBOL_GPL(nfs_file_read);
181 
182 int
nfs_file_mmap(struct file * file,struct vm_area_struct * vma)183 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
184 {
185 	struct inode *inode = file_inode(file);
186 	int	status;
187 
188 	dprintk("NFS: mmap(%pD2)\n", file);
189 
190 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
191 	 *       so we call that before revalidating the mapping
192 	 */
193 	status = generic_file_mmap(file, vma);
194 	if (!status) {
195 		vma->vm_ops = &nfs_file_vm_ops;
196 		status = nfs_revalidate_mapping(inode, file->f_mapping);
197 	}
198 	return status;
199 }
200 EXPORT_SYMBOL_GPL(nfs_file_mmap);
201 
202 /*
203  * Flush any dirty pages for this process, and check for write errors.
204  * The return status from this call provides a reliable indication of
205  * whether any write errors occurred for this process.
206  */
207 static int
nfs_file_fsync_commit(struct file * file,int datasync)208 nfs_file_fsync_commit(struct file *file, int datasync)
209 {
210 	struct nfs_open_context *ctx = nfs_file_open_context(file);
211 	struct inode *inode = file_inode(file);
212 	int do_resend, status;
213 	int ret = 0;
214 
215 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
216 
217 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
218 	do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
219 	status = nfs_commit_inode(inode, FLUSH_SYNC);
220 	if (status == 0)
221 		status = file_check_and_advance_wb_err(file);
222 	if (status < 0) {
223 		ret = status;
224 		goto out;
225 	}
226 	do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
227 	if (do_resend)
228 		ret = -EAGAIN;
229 out:
230 	return ret;
231 }
232 
233 int
nfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)234 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
235 {
236 	int ret;
237 	struct inode *inode = file_inode(file);
238 
239 	trace_nfs_fsync_enter(inode);
240 
241 	do {
242 		ret = file_write_and_wait_range(file, start, end);
243 		if (ret != 0)
244 			break;
245 		ret = nfs_file_fsync_commit(file, datasync);
246 		if (!ret)
247 			ret = pnfs_sync_inode(inode, !!datasync);
248 		/*
249 		 * If nfs_file_fsync_commit detected a server reboot, then
250 		 * resend all dirty pages that might have been covered by
251 		 * the NFS_CONTEXT_RESEND_WRITES flag
252 		 */
253 		start = 0;
254 		end = LLONG_MAX;
255 	} while (ret == -EAGAIN);
256 
257 	trace_nfs_fsync_exit(inode, ret);
258 	return ret;
259 }
260 EXPORT_SYMBOL_GPL(nfs_file_fsync);
261 
262 /*
263  * Decide whether a read/modify/write cycle may be more efficient
264  * then a modify/write/read cycle when writing to a page in the
265  * page cache.
266  *
267  * Some pNFS layout drivers can only read/write at a certain block
268  * granularity like all block devices and therefore we must perform
269  * read/modify/write whenever a page hasn't read yet and the data
270  * to be written there is not aligned to a block boundary and/or
271  * smaller than the block size.
272  *
273  * The modify/write/read cycle may occur if a page is read before
274  * being completely filled by the writer.  In this situation, the
275  * page must be completely written to stable storage on the server
276  * before it can be refilled by reading in the page from the server.
277  * This can lead to expensive, small, FILE_SYNC mode writes being
278  * done.
279  *
280  * It may be more efficient to read the page first if the file is
281  * open for reading in addition to writing, the page is not marked
282  * as Uptodate, it is not dirty or waiting to be committed,
283  * indicating that it was previously allocated and then modified,
284  * that there were valid bytes of data in that range of the file,
285  * and that the new data won't completely replace the old data in
286  * that range of the file.
287  */
nfs_full_page_write(struct page * page,loff_t pos,unsigned int len)288 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
289 {
290 	unsigned int pglen = nfs_page_length(page);
291 	unsigned int offset = pos & (PAGE_SIZE - 1);
292 	unsigned int end = offset + len;
293 
294 	return !pglen || (end >= pglen && !offset);
295 }
296 
nfs_want_read_modify_write(struct file * file,struct page * page,loff_t pos,unsigned int len)297 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
298 			loff_t pos, unsigned int len)
299 {
300 	/*
301 	 * Up-to-date pages, those with ongoing or full-page write
302 	 * don't need read/modify/write
303 	 */
304 	if (PageUptodate(page) || PagePrivate(page) ||
305 	    nfs_full_page_write(page, pos, len))
306 		return false;
307 
308 	if (pnfs_ld_read_whole_page(file->f_mapping->host))
309 		return true;
310 	/* Open for reading too? */
311 	if (file->f_mode & FMODE_READ)
312 		return true;
313 	return false;
314 }
315 
316 /*
317  * This does the "real" work of the write. We must allocate and lock the
318  * page to be sent back to the generic routine, which then copies the
319  * data from user space.
320  *
321  * If the writer ends up delaying the write, the writer needs to
322  * increment the page use counts until he is done with the page.
323  */
nfs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)324 static int nfs_write_begin(struct file *file, struct address_space *mapping,
325 			loff_t pos, unsigned len, unsigned flags,
326 			struct page **pagep, void **fsdata)
327 {
328 	int ret;
329 	pgoff_t index = pos >> PAGE_SHIFT;
330 	struct page *page;
331 	int once_thru = 0;
332 
333 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
334 		file, mapping->host->i_ino, len, (long long) pos);
335 
336 start:
337 	page = grab_cache_page_write_begin(mapping, index, flags);
338 	if (!page)
339 		return -ENOMEM;
340 	*pagep = page;
341 
342 	ret = nfs_flush_incompatible(file, page);
343 	if (ret) {
344 		unlock_page(page);
345 		put_page(page);
346 	} else if (!once_thru &&
347 		   nfs_want_read_modify_write(file, page, pos, len)) {
348 		once_thru = 1;
349 		ret = nfs_readpage(file, page);
350 		put_page(page);
351 		if (!ret)
352 			goto start;
353 	}
354 	return ret;
355 }
356 
nfs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)357 static int nfs_write_end(struct file *file, struct address_space *mapping,
358 			loff_t pos, unsigned len, unsigned copied,
359 			struct page *page, void *fsdata)
360 {
361 	unsigned offset = pos & (PAGE_SIZE - 1);
362 	struct nfs_open_context *ctx = nfs_file_open_context(file);
363 	int status;
364 
365 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
366 		file, mapping->host->i_ino, len, (long long) pos);
367 
368 	/*
369 	 * Zero any uninitialised parts of the page, and then mark the page
370 	 * as up to date if it turns out that we're extending the file.
371 	 */
372 	if (!PageUptodate(page)) {
373 		unsigned pglen = nfs_page_length(page);
374 		unsigned end = offset + copied;
375 
376 		if (pglen == 0) {
377 			zero_user_segments(page, 0, offset,
378 					end, PAGE_SIZE);
379 			SetPageUptodate(page);
380 		} else if (end >= pglen) {
381 			zero_user_segment(page, end, PAGE_SIZE);
382 			if (offset == 0)
383 				SetPageUptodate(page);
384 		} else
385 			zero_user_segment(page, pglen, PAGE_SIZE);
386 	}
387 
388 	status = nfs_updatepage(file, page, offset, copied);
389 
390 	unlock_page(page);
391 	put_page(page);
392 
393 	if (status < 0)
394 		return status;
395 	NFS_I(mapping->host)->write_io += copied;
396 
397 	if (nfs_ctx_key_to_expire(ctx, mapping->host))
398 		nfs_wb_all(mapping->host);
399 
400 	return copied;
401 }
402 
403 /*
404  * Partially or wholly invalidate a page
405  * - Release the private state associated with a page if undergoing complete
406  *   page invalidation
407  * - Called if either PG_private or PG_fscache is set on the page
408  * - Caller holds page lock
409  */
nfs_invalidate_page(struct page * page,unsigned int offset,unsigned int length)410 static void nfs_invalidate_page(struct page *page, unsigned int offset,
411 				unsigned int length)
412 {
413 	dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
414 		 page, offset, length);
415 
416 	if (offset != 0 || length < PAGE_SIZE)
417 		return;
418 	/* Cancel any unstarted writes on this page */
419 	nfs_wb_page_cancel(page_file_mapping(page)->host, page);
420 
421 	nfs_fscache_invalidate_page(page, page->mapping->host);
422 }
423 
424 /*
425  * Attempt to release the private state associated with a page
426  * - Called if either PG_private or PG_fscache is set on the page
427  * - Caller holds page lock
428  * - Return true (may release page) or false (may not)
429  */
nfs_release_page(struct page * page,gfp_t gfp)430 static int nfs_release_page(struct page *page, gfp_t gfp)
431 {
432 	dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
433 
434 	/* If PagePrivate() is set, then the page is not freeable */
435 	if (PagePrivate(page))
436 		return 0;
437 	return nfs_fscache_release_page(page, gfp);
438 }
439 
nfs_check_dirty_writeback(struct page * page,bool * dirty,bool * writeback)440 static void nfs_check_dirty_writeback(struct page *page,
441 				bool *dirty, bool *writeback)
442 {
443 	struct nfs_inode *nfsi;
444 	struct address_space *mapping = page_file_mapping(page);
445 
446 	if (!mapping || PageSwapCache(page))
447 		return;
448 
449 	/*
450 	 * Check if an unstable page is currently being committed and
451 	 * if so, have the VM treat it as if the page is under writeback
452 	 * so it will not block due to pages that will shortly be freeable.
453 	 */
454 	nfsi = NFS_I(mapping->host);
455 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
456 		*writeback = true;
457 		return;
458 	}
459 
460 	/*
461 	 * If PagePrivate() is set, then the page is not freeable and as the
462 	 * inode is not being committed, it's not going to be cleaned in the
463 	 * near future so treat it as dirty
464 	 */
465 	if (PagePrivate(page))
466 		*dirty = true;
467 }
468 
469 /*
470  * Attempt to clear the private state associated with a page when an error
471  * occurs that requires the cached contents of an inode to be written back or
472  * destroyed
473  * - Called if either PG_private or fscache is set on the page
474  * - Caller holds page lock
475  * - Return 0 if successful, -error otherwise
476  */
nfs_launder_page(struct page * page)477 static int nfs_launder_page(struct page *page)
478 {
479 	struct inode *inode = page_file_mapping(page)->host;
480 	struct nfs_inode *nfsi = NFS_I(inode);
481 
482 	dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
483 		inode->i_ino, (long long)page_offset(page));
484 
485 	nfs_fscache_wait_on_page_write(nfsi, page);
486 	return nfs_wb_page(inode, page);
487 }
488 
nfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)489 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
490 						sector_t *span)
491 {
492 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
493 
494 	*span = sis->pages;
495 
496 	return rpc_clnt_swap_activate(clnt);
497 }
498 
nfs_swap_deactivate(struct file * file)499 static void nfs_swap_deactivate(struct file *file)
500 {
501 	struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
502 
503 	rpc_clnt_swap_deactivate(clnt);
504 }
505 
506 const struct address_space_operations nfs_file_aops = {
507 	.readpage = nfs_readpage,
508 	.readpages = nfs_readpages,
509 	.set_page_dirty = __set_page_dirty_nobuffers,
510 	.writepage = nfs_writepage,
511 	.writepages = nfs_writepages,
512 	.write_begin = nfs_write_begin,
513 	.write_end = nfs_write_end,
514 	.invalidatepage = nfs_invalidate_page,
515 	.releasepage = nfs_release_page,
516 	.direct_IO = nfs_direct_IO,
517 #ifdef CONFIG_MIGRATION
518 	.migratepage = nfs_migrate_page,
519 #endif
520 	.launder_page = nfs_launder_page,
521 	.is_dirty_writeback = nfs_check_dirty_writeback,
522 	.error_remove_page = generic_error_remove_page,
523 	.swap_activate = nfs_swap_activate,
524 	.swap_deactivate = nfs_swap_deactivate,
525 };
526 
527 /*
528  * Notification that a PTE pointing to an NFS page is about to be made
529  * writable, implying that someone is about to modify the page through a
530  * shared-writable mapping
531  */
nfs_vm_page_mkwrite(struct vm_fault * vmf)532 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
533 {
534 	struct page *page = vmf->page;
535 	struct file *filp = vmf->vma->vm_file;
536 	struct inode *inode = file_inode(filp);
537 	unsigned pagelen;
538 	vm_fault_t ret = VM_FAULT_NOPAGE;
539 	struct address_space *mapping;
540 
541 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
542 		filp, filp->f_mapping->host->i_ino,
543 		(long long)page_offset(page));
544 
545 	sb_start_pagefault(inode->i_sb);
546 
547 	/* make sure the cache has finished storing the page */
548 	nfs_fscache_wait_on_page_write(NFS_I(inode), page);
549 
550 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
551 			nfs_wait_bit_killable, TASK_KILLABLE);
552 
553 	lock_page(page);
554 	mapping = page_file_mapping(page);
555 	if (mapping != inode->i_mapping)
556 		goto out_unlock;
557 
558 	wait_on_page_writeback(page);
559 
560 	pagelen = nfs_page_length(page);
561 	if (pagelen == 0)
562 		goto out_unlock;
563 
564 	ret = VM_FAULT_LOCKED;
565 	if (nfs_flush_incompatible(filp, page) == 0 &&
566 	    nfs_updatepage(filp, page, 0, pagelen) == 0)
567 		goto out;
568 
569 	ret = VM_FAULT_SIGBUS;
570 out_unlock:
571 	unlock_page(page);
572 out:
573 	sb_end_pagefault(inode->i_sb);
574 	return ret;
575 }
576 
577 static const struct vm_operations_struct nfs_file_vm_ops = {
578 	.fault = filemap_fault,
579 	.map_pages = filemap_map_pages,
580 	.page_mkwrite = nfs_vm_page_mkwrite,
581 };
582 
nfs_need_check_write(struct file * filp,struct inode * inode,int error)583 static int nfs_need_check_write(struct file *filp, struct inode *inode,
584 				int error)
585 {
586 	struct nfs_open_context *ctx;
587 
588 	ctx = nfs_file_open_context(filp);
589 	if (nfs_error_is_fatal_on_server(error) ||
590 	    nfs_ctx_key_to_expire(ctx, inode))
591 		return 1;
592 	return 0;
593 }
594 
nfs_file_write(struct kiocb * iocb,struct iov_iter * from)595 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
596 {
597 	struct file *file = iocb->ki_filp;
598 	struct inode *inode = file_inode(file);
599 	unsigned long written = 0;
600 	ssize_t result;
601 	errseq_t since;
602 	int error;
603 
604 	result = nfs_key_timeout_notify(file, inode);
605 	if (result)
606 		return result;
607 
608 	if (iocb->ki_flags & IOCB_DIRECT)
609 		return nfs_file_direct_write(iocb, from, false);
610 
611 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
612 		file, iov_iter_count(from), (long long) iocb->ki_pos);
613 
614 	if (IS_SWAPFILE(inode))
615 		goto out_swapfile;
616 	/*
617 	 * O_APPEND implies that we must revalidate the file length.
618 	 */
619 	if (iocb->ki_flags & IOCB_APPEND) {
620 		result = nfs_revalidate_file_size(inode, file);
621 		if (result)
622 			goto out;
623 	}
624 	if (iocb->ki_pos > i_size_read(inode))
625 		nfs_revalidate_mapping(inode, file->f_mapping);
626 
627 	since = filemap_sample_wb_err(file->f_mapping);
628 	nfs_start_io_write(inode);
629 	result = generic_write_checks(iocb, from);
630 	if (result > 0) {
631 		current->backing_dev_info = inode_to_bdi(inode);
632 		result = generic_perform_write(file, from, iocb->ki_pos);
633 		current->backing_dev_info = NULL;
634 	}
635 	nfs_end_io_write(inode);
636 	if (result <= 0)
637 		goto out;
638 
639 	written = result;
640 	iocb->ki_pos += written;
641 	result = generic_write_sync(iocb, written);
642 	if (result < 0)
643 		goto out;
644 
645 	/* Return error values */
646 	error = filemap_check_wb_err(file->f_mapping, since);
647 	if (nfs_need_check_write(file, inode, error)) {
648 		int err = nfs_wb_all(inode);
649 		if (err < 0)
650 			result = err;
651 	}
652 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
653 out:
654 	return result;
655 
656 out_swapfile:
657 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
658 	return -EBUSY;
659 }
660 EXPORT_SYMBOL_GPL(nfs_file_write);
661 
662 static int
do_getlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)663 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
664 {
665 	struct inode *inode = filp->f_mapping->host;
666 	int status = 0;
667 	unsigned int saved_type = fl->fl_type;
668 
669 	/* Try local locking first */
670 	posix_test_lock(filp, fl);
671 	if (fl->fl_type != F_UNLCK) {
672 		/* found a conflict */
673 		goto out;
674 	}
675 	fl->fl_type = saved_type;
676 
677 	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
678 		goto out_noconflict;
679 
680 	if (is_local)
681 		goto out_noconflict;
682 
683 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
684 out:
685 	return status;
686 out_noconflict:
687 	fl->fl_type = F_UNLCK;
688 	goto out;
689 }
690 
691 static int
do_unlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)692 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
693 {
694 	struct inode *inode = filp->f_mapping->host;
695 	struct nfs_lock_context *l_ctx;
696 	int status;
697 
698 	/*
699 	 * Flush all pending writes before doing anything
700 	 * with locks..
701 	 */
702 	nfs_wb_all(inode);
703 
704 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
705 	if (!IS_ERR(l_ctx)) {
706 		status = nfs_iocounter_wait(l_ctx);
707 		nfs_put_lock_context(l_ctx);
708 		/*  NOTE: special case
709 		 * 	If we're signalled while cleaning up locks on process exit, we
710 		 * 	still need to complete the unlock.
711 		 */
712 		if (status < 0 && !(fl->fl_flags & FL_CLOSE))
713 			return status;
714 	}
715 
716 	/*
717 	 * Use local locking if mounted with "-onolock" or with appropriate
718 	 * "-olocal_lock="
719 	 */
720 	if (!is_local)
721 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
722 	else
723 		status = locks_lock_file_wait(filp, fl);
724 	return status;
725 }
726 
727 static int
do_setlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)728 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
729 {
730 	struct inode *inode = filp->f_mapping->host;
731 	int status;
732 
733 	/*
734 	 * Flush all pending writes before doing anything
735 	 * with locks..
736 	 */
737 	status = nfs_sync_mapping(filp->f_mapping);
738 	if (status != 0)
739 		goto out;
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 = locks_lock_file_wait(filp, fl);
749 	if (status < 0)
750 		goto out;
751 
752 	/*
753 	 * Invalidate cache to prevent missing any changes.  If
754 	 * the file is mapped, clear the page cache as well so
755 	 * those mappings will be loaded.
756 	 *
757 	 * This makes locking act as a cache coherency point.
758 	 */
759 	nfs_sync_mapping(filp->f_mapping);
760 	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
761 		nfs_zap_caches(inode);
762 		if (mapping_mapped(filp->f_mapping))
763 			nfs_revalidate_mapping(inode, filp->f_mapping);
764 	}
765 out:
766 	return status;
767 }
768 
769 /*
770  * Lock a (portion of) a file
771  */
nfs_lock(struct file * filp,int cmd,struct file_lock * fl)772 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
773 {
774 	struct inode *inode = filp->f_mapping->host;
775 	int ret = -ENOLCK;
776 	int is_local = 0;
777 
778 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
779 			filp, 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_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
789 		is_local = 1;
790 
791 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
792 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
793 		if (ret < 0)
794 			goto out_err;
795 	}
796 
797 	if (IS_GETLK(cmd))
798 		ret = do_getlk(filp, cmd, fl, is_local);
799 	else if (fl->fl_type == F_UNLCK)
800 		ret = do_unlk(filp, cmd, fl, is_local);
801 	else
802 		ret = do_setlk(filp, cmd, fl, is_local);
803 out_err:
804 	return ret;
805 }
806 EXPORT_SYMBOL_GPL(nfs_lock);
807 
808 /*
809  * Lock a (portion of) a file
810  */
nfs_flock(struct file * filp,int cmd,struct file_lock * fl)811 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
812 {
813 	struct inode *inode = filp->f_mapping->host;
814 	int is_local = 0;
815 
816 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
817 			filp, fl->fl_type, fl->fl_flags);
818 
819 	if (!(fl->fl_flags & FL_FLOCK))
820 		return -ENOLCK;
821 
822 	/*
823 	 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
824 	 * any standard. In principle we might be able to support LOCK_MAND
825 	 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
826 	 * NFS code is not set up for it.
827 	 */
828 	if (fl->fl_type & LOCK_MAND)
829 		return -EINVAL;
830 
831 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
832 		is_local = 1;
833 
834 	/* We're simulating flock() locks using posix locks on the server */
835 	if (fl->fl_type == F_UNLCK)
836 		return do_unlk(filp, cmd, fl, is_local);
837 	return do_setlk(filp, cmd, fl, is_local);
838 }
839 EXPORT_SYMBOL_GPL(nfs_flock);
840 
841 const struct file_operations nfs_file_operations = {
842 	.llseek		= nfs_file_llseek,
843 	.read_iter	= nfs_file_read,
844 	.write_iter	= nfs_file_write,
845 	.mmap		= nfs_file_mmap,
846 	.open		= nfs_file_open,
847 	.flush		= nfs_file_flush,
848 	.release	= nfs_file_release,
849 	.fsync		= nfs_file_fsync,
850 	.lock		= nfs_lock,
851 	.flock		= nfs_flock,
852 	.splice_read	= generic_file_splice_read,
853 	.splice_write	= iter_file_splice_write,
854 	.check_flags	= nfs_check_flags,
855 	.setlease	= simple_nosetlease,
856 };
857 EXPORT_SYMBOL_GPL(nfs_file_operations);
858