xref: /linux/fs/nfs/file.c (revision baf67f6aa9d29512809f1b1fbab624fce57fd16d)
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 #include <linux/filelock.h>
35 
36 #include "delegation.h"
37 #include "internal.h"
38 #include "iostat.h"
39 #include "fscache.h"
40 #include "pnfs.h"
41 
42 #include "nfstrace.h"
43 
44 #define NFSDBG_FACILITY		NFSDBG_FILE
45 
46 static const struct vm_operations_struct nfs_file_vm_ops;
47 
nfs_check_flags(int flags)48 int nfs_check_flags(int flags)
49 {
50 	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
51 		return -EINVAL;
52 
53 	return 0;
54 }
55 EXPORT_SYMBOL_GPL(nfs_check_flags);
56 
57 /*
58  * Open file
59  */
60 static int
nfs_file_open(struct inode * inode,struct file * filp)61 nfs_file_open(struct inode *inode, struct file *filp)
62 {
63 	int res;
64 
65 	dprintk("NFS: open file(%pD2)\n", filp);
66 
67 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
68 	res = nfs_check_flags(filp->f_flags);
69 	if (res)
70 		return res;
71 
72 	res = nfs_open(inode, filp);
73 	if (res == 0)
74 		filp->f_mode |= FMODE_CAN_ODIRECT;
75 	return res;
76 }
77 
78 int
nfs_file_release(struct inode * inode,struct file * filp)79 nfs_file_release(struct inode *inode, struct file *filp)
80 {
81 	dprintk("NFS: release(%pD2)\n", filp);
82 
83 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
84 	nfs_file_clear_open_context(filp);
85 	nfs_fscache_release_file(inode, filp);
86 	return 0;
87 }
88 EXPORT_SYMBOL_GPL(nfs_file_release);
89 
90 /**
91  * nfs_revalidate_file_size - Revalidate the file size
92  * @inode: pointer to inode struct
93  * @filp: pointer to struct file
94  *
95  * Revalidates the file length. This is basically a wrapper around
96  * nfs_revalidate_inode() that takes into account the fact that we may
97  * have cached writes (in which case we don't care about the server's
98  * idea of what the file length is), or O_DIRECT (in which case we
99  * shouldn't trust the cache).
100  */
nfs_revalidate_file_size(struct inode * inode,struct file * filp)101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102 {
103 	struct nfs_server *server = NFS_SERVER(inode);
104 
105 	if (filp->f_flags & O_DIRECT)
106 		goto force_reval;
107 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
108 		goto force_reval;
109 	return 0;
110 force_reval:
111 	return __nfs_revalidate_inode(server, inode);
112 }
113 
nfs_file_llseek(struct file * filp,loff_t offset,int whence)114 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
115 {
116 	dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
117 			filp, offset, whence);
118 
119 	/*
120 	 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
121 	 * the cached file length
122 	 */
123 	if (whence != SEEK_SET && whence != SEEK_CUR) {
124 		struct inode *inode = filp->f_mapping->host;
125 
126 		int retval = nfs_revalidate_file_size(inode, filp);
127 		if (retval < 0)
128 			return (loff_t)retval;
129 	}
130 
131 	return generic_file_llseek(filp, offset, whence);
132 }
133 EXPORT_SYMBOL_GPL(nfs_file_llseek);
134 
135 /*
136  * Flush all dirty pages, and check for write errors.
137  */
138 static int
nfs_file_flush(struct file * file,fl_owner_t id)139 nfs_file_flush(struct file *file, fl_owner_t id)
140 {
141 	struct inode	*inode = file_inode(file);
142 	errseq_t since;
143 
144 	dprintk("NFS: flush(%pD2)\n", file);
145 
146 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
147 	if ((file->f_mode & FMODE_WRITE) == 0)
148 		return 0;
149 
150 	/* Flush writes to the server and return any errors */
151 	since = filemap_sample_wb_err(file->f_mapping);
152 	nfs_wb_all(inode);
153 	return filemap_check_wb_err(file->f_mapping, since);
154 }
155 
156 ssize_t
nfs_file_read(struct kiocb * iocb,struct iov_iter * to)157 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
158 {
159 	struct inode *inode = file_inode(iocb->ki_filp);
160 	ssize_t result;
161 
162 	if (iocb->ki_flags & IOCB_DIRECT)
163 		return nfs_file_direct_read(iocb, to, false);
164 
165 	dprintk("NFS: read(%pD2, %zu@%lu)\n",
166 		iocb->ki_filp,
167 		iov_iter_count(to), (unsigned long) iocb->ki_pos);
168 
169 	result = nfs_start_io_read(inode);
170 	if (result)
171 		return result;
172 
173 	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
174 	if (!result) {
175 		result = generic_file_read_iter(iocb, to);
176 		if (result > 0)
177 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
178 	}
179 	nfs_end_io_read(inode);
180 	return result;
181 }
182 EXPORT_SYMBOL_GPL(nfs_file_read);
183 
184 ssize_t
nfs_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)185 nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
186 		     size_t len, unsigned int flags)
187 {
188 	struct inode *inode = file_inode(in);
189 	ssize_t result;
190 
191 	dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
192 
193 	result = nfs_start_io_read(inode);
194 	if (result)
195 		return result;
196 
197 	result = nfs_revalidate_mapping(inode, in->f_mapping);
198 	if (!result) {
199 		result = filemap_splice_read(in, ppos, pipe, len, flags);
200 		if (result > 0)
201 			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
202 	}
203 	nfs_end_io_read(inode);
204 	return result;
205 }
206 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
207 
208 int
nfs_file_mmap(struct file * file,struct vm_area_struct * vma)209 nfs_file_mmap(struct file *file, struct vm_area_struct *vma)
210 {
211 	struct inode *inode = file_inode(file);
212 	int	status;
213 
214 	dprintk("NFS: mmap(%pD2)\n", file);
215 
216 	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
217 	 *       so we call that before revalidating the mapping
218 	 */
219 	status = generic_file_mmap(file, vma);
220 	if (!status) {
221 		vma->vm_ops = &nfs_file_vm_ops;
222 		status = nfs_revalidate_mapping(inode, file->f_mapping);
223 	}
224 	return status;
225 }
226 EXPORT_SYMBOL_GPL(nfs_file_mmap);
227 
228 /*
229  * Flush any dirty pages for this process, and check for write errors.
230  * The return status from this call provides a reliable indication of
231  * whether any write errors occurred for this process.
232  */
233 static int
nfs_file_fsync_commit(struct file * file,int datasync)234 nfs_file_fsync_commit(struct file *file, int datasync)
235 {
236 	struct inode *inode = file_inode(file);
237 	int ret, ret2;
238 
239 	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
240 
241 	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
242 	ret = nfs_commit_inode(inode, FLUSH_SYNC);
243 	ret2 = file_check_and_advance_wb_err(file);
244 	if (ret2 < 0)
245 		return ret2;
246 	return ret;
247 }
248 
249 int
nfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)250 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
251 {
252 	struct inode *inode = file_inode(file);
253 	struct nfs_inode *nfsi = NFS_I(inode);
254 	long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
255 	long nredirtied;
256 	int ret;
257 
258 	trace_nfs_fsync_enter(inode);
259 
260 	for (;;) {
261 		ret = file_write_and_wait_range(file, start, end);
262 		if (ret != 0)
263 			break;
264 		ret = nfs_file_fsync_commit(file, datasync);
265 		if (ret != 0)
266 			break;
267 		ret = pnfs_sync_inode(inode, !!datasync);
268 		if (ret != 0)
269 			break;
270 		nredirtied = atomic_long_read(&nfsi->redirtied_pages);
271 		if (nredirtied == save_nredirtied)
272 			break;
273 		save_nredirtied = nredirtied;
274 	}
275 
276 	trace_nfs_fsync_exit(inode, ret);
277 	return ret;
278 }
279 EXPORT_SYMBOL_GPL(nfs_file_fsync);
280 
281 /*
282  * Decide whether a read/modify/write cycle may be more efficient
283  * then a modify/write/read cycle when writing to a page in the
284  * page cache.
285  *
286  * Some pNFS layout drivers can only read/write at a certain block
287  * granularity like all block devices and therefore we must perform
288  * read/modify/write whenever a page hasn't read yet and the data
289  * to be written there is not aligned to a block boundary and/or
290  * smaller than the block size.
291  *
292  * The modify/write/read cycle may occur if a page is read before
293  * being completely filled by the writer.  In this situation, the
294  * page must be completely written to stable storage on the server
295  * before it can be refilled by reading in the page from the server.
296  * This can lead to expensive, small, FILE_SYNC mode writes being
297  * done.
298  *
299  * It may be more efficient to read the page first if the file is
300  * open for reading in addition to writing, the page is not marked
301  * as Uptodate, it is not dirty or waiting to be committed,
302  * indicating that it was previously allocated and then modified,
303  * that there were valid bytes of data in that range of the file,
304  * and that the new data won't completely replace the old data in
305  * that range of the file.
306  */
nfs_folio_is_full_write(struct folio * folio,loff_t pos,unsigned int len)307 static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
308 				    unsigned int len)
309 {
310 	unsigned int pglen = nfs_folio_length(folio);
311 	unsigned int offset = offset_in_folio(folio, pos);
312 	unsigned int end = offset + len;
313 
314 	return !pglen || (end >= pglen && !offset);
315 }
316 
nfs_want_read_modify_write(struct file * file,struct folio * folio,loff_t pos,unsigned int len)317 static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
318 				       loff_t pos, unsigned int len)
319 {
320 	/*
321 	 * Up-to-date pages, those with ongoing or full-page write
322 	 * don't need read/modify/write
323 	 */
324 	if (folio_test_uptodate(folio) || folio_test_private(folio) ||
325 	    nfs_folio_is_full_write(folio, pos, len))
326 		return false;
327 
328 	if (pnfs_ld_read_whole_page(file_inode(file)))
329 		return true;
330 	/* Open for reading too? */
331 	if (file->f_mode & FMODE_READ)
332 		return true;
333 	return false;
334 }
335 
336 /*
337  * This does the "real" work of the write. We must allocate and lock the
338  * page to be sent back to the generic routine, which then copies the
339  * data from user space.
340  *
341  * If the writer ends up delaying the write, the writer needs to
342  * increment the page use counts until he is done with the page.
343  */
nfs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)344 static int nfs_write_begin(struct file *file, struct address_space *mapping,
345 			   loff_t pos, unsigned len, struct folio **foliop,
346 			   void **fsdata)
347 {
348 	fgf_t fgp = FGP_WRITEBEGIN;
349 	struct folio *folio;
350 	int once_thru = 0;
351 	int ret;
352 
353 	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
354 		file, mapping->host->i_ino, len, (long long) pos);
355 
356 	fgp |= fgf_set_order(len);
357 start:
358 	folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp,
359 				    mapping_gfp_mask(mapping));
360 	if (IS_ERR(folio))
361 		return PTR_ERR(folio);
362 	*foliop = folio;
363 
364 	ret = nfs_flush_incompatible(file, folio);
365 	if (ret) {
366 		folio_unlock(folio);
367 		folio_put(folio);
368 	} else if (!once_thru &&
369 		   nfs_want_read_modify_write(file, folio, pos, len)) {
370 		once_thru = 1;
371 		ret = nfs_read_folio(file, folio);
372 		folio_put(folio);
373 		if (!ret)
374 			goto start;
375 	}
376 	return ret;
377 }
378 
nfs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)379 static int nfs_write_end(struct file *file, struct address_space *mapping,
380 			 loff_t pos, unsigned len, unsigned copied,
381 			 struct folio *folio, void *fsdata)
382 {
383 	struct nfs_open_context *ctx = nfs_file_open_context(file);
384 	unsigned offset = offset_in_folio(folio, pos);
385 	int status;
386 
387 	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
388 		file, mapping->host->i_ino, len, (long long) pos);
389 
390 	/*
391 	 * Zero any uninitialised parts of the page, and then mark the page
392 	 * as up to date if it turns out that we're extending the file.
393 	 */
394 	if (!folio_test_uptodate(folio)) {
395 		size_t fsize = folio_size(folio);
396 		unsigned pglen = nfs_folio_length(folio);
397 		unsigned end = offset + copied;
398 
399 		if (pglen == 0) {
400 			folio_zero_segments(folio, 0, offset, end, fsize);
401 			folio_mark_uptodate(folio);
402 		} else if (end >= pglen) {
403 			folio_zero_segment(folio, end, fsize);
404 			if (offset == 0)
405 				folio_mark_uptodate(folio);
406 		} else
407 			folio_zero_segment(folio, pglen, fsize);
408 	}
409 
410 	status = nfs_update_folio(file, folio, offset, copied);
411 
412 	folio_unlock(folio);
413 	folio_put(folio);
414 
415 	if (status < 0)
416 		return status;
417 	NFS_I(mapping->host)->write_io += copied;
418 
419 	if (nfs_ctx_key_to_expire(ctx, mapping->host))
420 		nfs_wb_all(mapping->host);
421 
422 	return copied;
423 }
424 
425 /*
426  * Partially or wholly invalidate a page
427  * - Release the private state associated with a page if undergoing complete
428  *   page invalidation
429  * - Called if either PG_private or PG_fscache is set on the page
430  * - Caller holds page lock
431  */
nfs_invalidate_folio(struct folio * folio,size_t offset,size_t length)432 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
433 				size_t length)
434 {
435 	struct inode *inode = folio->mapping->host;
436 	dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
437 		 folio->index, offset, length);
438 
439 	if (offset != 0 || length < folio_size(folio))
440 		return;
441 	/* Cancel any unstarted writes on this page */
442 	nfs_wb_folio_cancel(inode, folio);
443 	folio_wait_private_2(folio); /* [DEPRECATED] */
444 	trace_nfs_invalidate_folio(inode, folio_pos(folio) + offset, length);
445 }
446 
447 /*
448  * Attempt to release the private state associated with a folio
449  * - Called if either private or fscache flags are set on the folio
450  * - Caller holds folio lock
451  * - Return true (may release folio) or false (may not)
452  */
nfs_release_folio(struct folio * folio,gfp_t gfp)453 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
454 {
455 	dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
456 
457 	/* If the private flag is set, then the folio is not freeable */
458 	if (folio_test_private(folio)) {
459 		if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
460 		    current_is_kswapd())
461 			return false;
462 		if (nfs_wb_folio(folio->mapping->host, folio) < 0)
463 			return false;
464 	}
465 	return nfs_fscache_release_folio(folio, gfp);
466 }
467 
nfs_check_dirty_writeback(struct folio * folio,bool * dirty,bool * writeback)468 static void nfs_check_dirty_writeback(struct folio *folio,
469 				bool *dirty, bool *writeback)
470 {
471 	struct nfs_inode *nfsi;
472 	struct address_space *mapping = folio->mapping;
473 
474 	/*
475 	 * Check if an unstable folio is currently being committed and
476 	 * if so, have the VM treat it as if the folio is under writeback
477 	 * so it will not block due to folios that will shortly be freeable.
478 	 */
479 	nfsi = NFS_I(mapping->host);
480 	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
481 		*writeback = true;
482 		return;
483 	}
484 
485 	/*
486 	 * If the private flag is set, then the folio is not freeable
487 	 * and as the inode is not being committed, it's not going to
488 	 * be cleaned in the near future so treat it as dirty
489 	 */
490 	if (folio_test_private(folio))
491 		*dirty = true;
492 }
493 
494 /*
495  * Attempt to clear the private state associated with a page when an error
496  * occurs that requires the cached contents of an inode to be written back or
497  * destroyed
498  * - Called if either PG_private or fscache is set on the page
499  * - Caller holds page lock
500  * - Return 0 if successful, -error otherwise
501  */
nfs_launder_folio(struct folio * folio)502 static int nfs_launder_folio(struct folio *folio)
503 {
504 	struct inode *inode = folio->mapping->host;
505 	int ret;
506 
507 	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
508 		inode->i_ino, folio_pos(folio));
509 
510 	folio_wait_private_2(folio); /* [DEPRECATED] */
511 	ret = nfs_wb_folio(inode, folio);
512 	trace_nfs_launder_folio_done(inode, folio_pos(folio),
513 			folio_size(folio), ret);
514 	return ret;
515 }
516 
nfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)517 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
518 						sector_t *span)
519 {
520 	unsigned long blocks;
521 	long long isize;
522 	int ret;
523 	struct inode *inode = file_inode(file);
524 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
525 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
526 
527 	spin_lock(&inode->i_lock);
528 	blocks = inode->i_blocks;
529 	isize = inode->i_size;
530 	spin_unlock(&inode->i_lock);
531 	if (blocks*512 < isize) {
532 		pr_warn("swap activate: swapfile has holes\n");
533 		return -EINVAL;
534 	}
535 
536 	ret = rpc_clnt_swap_activate(clnt);
537 	if (ret)
538 		return ret;
539 	ret = add_swap_extent(sis, 0, sis->max, 0);
540 	if (ret < 0) {
541 		rpc_clnt_swap_deactivate(clnt);
542 		return ret;
543 	}
544 
545 	*span = sis->pages;
546 
547 	if (cl->rpc_ops->enable_swap)
548 		cl->rpc_ops->enable_swap(inode);
549 
550 	sis->flags |= SWP_FS_OPS;
551 	return ret;
552 }
553 
nfs_swap_deactivate(struct file * file)554 static void nfs_swap_deactivate(struct file *file)
555 {
556 	struct inode *inode = file_inode(file);
557 	struct rpc_clnt *clnt = NFS_CLIENT(inode);
558 	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
559 
560 	rpc_clnt_swap_deactivate(clnt);
561 	if (cl->rpc_ops->disable_swap)
562 		cl->rpc_ops->disable_swap(file_inode(file));
563 }
564 
565 const struct address_space_operations nfs_file_aops = {
566 	.read_folio = nfs_read_folio,
567 	.readahead = nfs_readahead,
568 	.dirty_folio = filemap_dirty_folio,
569 	.writepages = nfs_writepages,
570 	.write_begin = nfs_write_begin,
571 	.write_end = nfs_write_end,
572 	.invalidate_folio = nfs_invalidate_folio,
573 	.release_folio = nfs_release_folio,
574 	.migrate_folio = nfs_migrate_folio,
575 	.launder_folio = nfs_launder_folio,
576 	.is_dirty_writeback = nfs_check_dirty_writeback,
577 	.error_remove_folio = generic_error_remove_folio,
578 	.swap_activate = nfs_swap_activate,
579 	.swap_deactivate = nfs_swap_deactivate,
580 	.swap_rw = nfs_swap_rw,
581 };
582 
583 /*
584  * Notification that a PTE pointing to an NFS page is about to be made
585  * writable, implying that someone is about to modify the page through a
586  * shared-writable mapping
587  */
nfs_vm_page_mkwrite(struct vm_fault * vmf)588 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
589 {
590 	struct file *filp = vmf->vma->vm_file;
591 	struct inode *inode = file_inode(filp);
592 	unsigned pagelen;
593 	vm_fault_t ret = VM_FAULT_NOPAGE;
594 	struct address_space *mapping;
595 	struct folio *folio = page_folio(vmf->page);
596 
597 	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
598 		 filp, filp->f_mapping->host->i_ino,
599 		 (long long)folio_pos(folio));
600 
601 	sb_start_pagefault(inode->i_sb);
602 
603 	/* make sure the cache has finished storing the page */
604 	if (folio_test_private_2(folio) && /* [DEPRECATED] */
605 	    folio_wait_private_2_killable(folio) < 0) {
606 		ret = VM_FAULT_RETRY;
607 		goto out;
608 	}
609 
610 	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
611 			   nfs_wait_bit_killable,
612 			   TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
613 
614 	folio_lock(folio);
615 	mapping = folio->mapping;
616 	if (mapping != inode->i_mapping)
617 		goto out_unlock;
618 
619 	folio_wait_writeback(folio);
620 
621 	pagelen = nfs_folio_length(folio);
622 	if (pagelen == 0)
623 		goto out_unlock;
624 
625 	ret = VM_FAULT_LOCKED;
626 	if (nfs_flush_incompatible(filp, folio) == 0 &&
627 	    nfs_update_folio(filp, folio, 0, pagelen) == 0)
628 		goto out;
629 
630 	ret = VM_FAULT_SIGBUS;
631 out_unlock:
632 	folio_unlock(folio);
633 out:
634 	sb_end_pagefault(inode->i_sb);
635 	return ret;
636 }
637 
638 static const struct vm_operations_struct nfs_file_vm_ops = {
639 	.fault = filemap_fault,
640 	.map_pages = filemap_map_pages,
641 	.page_mkwrite = nfs_vm_page_mkwrite,
642 };
643 
nfs_file_write(struct kiocb * iocb,struct iov_iter * from)644 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
645 {
646 	struct file *file = iocb->ki_filp;
647 	struct inode *inode = file_inode(file);
648 	unsigned int mntflags = NFS_SERVER(inode)->flags;
649 	ssize_t result, written;
650 	errseq_t since;
651 	int error;
652 
653 	result = nfs_key_timeout_notify(file, inode);
654 	if (result)
655 		return result;
656 
657 	if (iocb->ki_flags & IOCB_DIRECT)
658 		return nfs_file_direct_write(iocb, from, false);
659 
660 	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
661 		file, iov_iter_count(from), (long long) iocb->ki_pos);
662 
663 	if (IS_SWAPFILE(inode))
664 		goto out_swapfile;
665 	/*
666 	 * O_APPEND implies that we must revalidate the file length.
667 	 */
668 	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
669 		result = nfs_revalidate_file_size(inode, file);
670 		if (result)
671 			return result;
672 	}
673 
674 	nfs_clear_invalid_mapping(file->f_mapping);
675 
676 	since = filemap_sample_wb_err(file->f_mapping);
677 	error = nfs_start_io_write(inode);
678 	if (error)
679 		return error;
680 	result = generic_write_checks(iocb, from);
681 	if (result > 0)
682 		result = generic_perform_write(iocb, from);
683 	nfs_end_io_write(inode);
684 	if (result <= 0)
685 		goto out;
686 
687 	written = result;
688 	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
689 
690 	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
691 		result = filemap_fdatawrite_range(file->f_mapping,
692 						  iocb->ki_pos - written,
693 						  iocb->ki_pos - 1);
694 		if (result < 0)
695 			goto out;
696 	}
697 	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
698 		filemap_fdatawait_range(file->f_mapping,
699 					iocb->ki_pos - written,
700 					iocb->ki_pos - 1);
701 	}
702 	result = generic_write_sync(iocb, written);
703 	if (result < 0)
704 		return result;
705 
706 out:
707 	/* Return error values */
708 	error = filemap_check_wb_err(file->f_mapping, since);
709 	switch (error) {
710 	default:
711 		break;
712 	case -EDQUOT:
713 	case -EFBIG:
714 	case -ENOSPC:
715 		nfs_wb_all(inode);
716 		error = file_check_and_advance_wb_err(file);
717 		if (error < 0)
718 			result = error;
719 	}
720 	return result;
721 
722 out_swapfile:
723 	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
724 	return -ETXTBSY;
725 }
726 EXPORT_SYMBOL_GPL(nfs_file_write);
727 
728 static int
do_getlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)729 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
730 {
731 	struct inode *inode = filp->f_mapping->host;
732 	int status = 0;
733 	unsigned int saved_type = fl->c.flc_type;
734 
735 	/* Try local locking first */
736 	posix_test_lock(filp, fl);
737 	if (fl->c.flc_type != F_UNLCK) {
738 		/* found a conflict */
739 		goto out;
740 	}
741 	fl->c.flc_type = saved_type;
742 
743 	if (nfs_have_read_or_write_delegation(inode))
744 		goto out_noconflict;
745 
746 	if (is_local)
747 		goto out_noconflict;
748 
749 	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
750 out:
751 	return status;
752 out_noconflict:
753 	fl->c.flc_type = F_UNLCK;
754 	goto out;
755 }
756 
757 static int
do_unlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)758 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
759 {
760 	struct inode *inode = filp->f_mapping->host;
761 	struct nfs_lock_context *l_ctx;
762 	int status;
763 
764 	/*
765 	 * Flush all pending writes before doing anything
766 	 * with locks..
767 	 */
768 	nfs_wb_all(inode);
769 
770 	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
771 	if (!IS_ERR(l_ctx)) {
772 		status = nfs_iocounter_wait(l_ctx);
773 		nfs_put_lock_context(l_ctx);
774 		/*  NOTE: special case
775 		 * 	If we're signalled while cleaning up locks on process exit, we
776 		 * 	still need to complete the unlock.
777 		 */
778 		if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
779 			return status;
780 	}
781 
782 	/*
783 	 * Use local locking if mounted with "-onolock" or with appropriate
784 	 * "-olocal_lock="
785 	 */
786 	if (!is_local)
787 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
788 	else
789 		status = locks_lock_file_wait(filp, fl);
790 	return status;
791 }
792 
793 static int
do_setlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)794 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
795 {
796 	struct inode *inode = filp->f_mapping->host;
797 	int status;
798 
799 	/*
800 	 * Flush all pending writes before doing anything
801 	 * with locks..
802 	 */
803 	status = nfs_sync_mapping(filp->f_mapping);
804 	if (status != 0)
805 		goto out;
806 
807 	/*
808 	 * Use local locking if mounted with "-onolock" or with appropriate
809 	 * "-olocal_lock="
810 	 */
811 	if (!is_local)
812 		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
813 	else
814 		status = locks_lock_file_wait(filp, fl);
815 	if (status < 0)
816 		goto out;
817 
818 	/*
819 	 * Invalidate cache to prevent missing any changes.  If
820 	 * the file is mapped, clear the page cache as well so
821 	 * those mappings will be loaded.
822 	 *
823 	 * This makes locking act as a cache coherency point.
824 	 */
825 	nfs_sync_mapping(filp->f_mapping);
826 	if (!nfs_have_read_or_write_delegation(inode)) {
827 		nfs_zap_caches(inode);
828 		if (mapping_mapped(filp->f_mapping))
829 			nfs_revalidate_mapping(inode, filp->f_mapping);
830 	}
831 out:
832 	return status;
833 }
834 
835 /*
836  * Lock a (portion of) a file
837  */
nfs_lock(struct file * filp,int cmd,struct file_lock * fl)838 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
839 {
840 	struct inode *inode = filp->f_mapping->host;
841 	int ret = -ENOLCK;
842 	int is_local = 0;
843 
844 	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
845 			filp, fl->c.flc_type, fl->c.flc_flags,
846 			(long long)fl->fl_start, (long long)fl->fl_end);
847 
848 	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
849 
850 	if (fl->c.flc_flags & FL_RECLAIM)
851 		return -ENOGRACE;
852 
853 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
854 		is_local = 1;
855 
856 	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
857 		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
858 		if (ret < 0)
859 			goto out_err;
860 	}
861 
862 	if (IS_GETLK(cmd))
863 		ret = do_getlk(filp, cmd, fl, is_local);
864 	else if (lock_is_unlock(fl))
865 		ret = do_unlk(filp, cmd, fl, is_local);
866 	else
867 		ret = do_setlk(filp, cmd, fl, is_local);
868 out_err:
869 	return ret;
870 }
871 EXPORT_SYMBOL_GPL(nfs_lock);
872 
873 /*
874  * Lock a (portion of) a file
875  */
nfs_flock(struct file * filp,int cmd,struct file_lock * fl)876 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
877 {
878 	struct inode *inode = filp->f_mapping->host;
879 	int is_local = 0;
880 
881 	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
882 			filp, fl->c.flc_type, fl->c.flc_flags);
883 
884 	if (!(fl->c.flc_flags & FL_FLOCK))
885 		return -ENOLCK;
886 
887 	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
888 		is_local = 1;
889 
890 	/* We're simulating flock() locks using posix locks on the server */
891 	if (lock_is_unlock(fl))
892 		return do_unlk(filp, cmd, fl, is_local);
893 	return do_setlk(filp, cmd, fl, is_local);
894 }
895 EXPORT_SYMBOL_GPL(nfs_flock);
896 
897 const struct file_operations nfs_file_operations = {
898 	.llseek		= nfs_file_llseek,
899 	.read_iter	= nfs_file_read,
900 	.write_iter	= nfs_file_write,
901 	.mmap		= nfs_file_mmap,
902 	.open		= nfs_file_open,
903 	.flush		= nfs_file_flush,
904 	.release	= nfs_file_release,
905 	.fsync		= nfs_file_fsync,
906 	.lock		= nfs_lock,
907 	.flock		= nfs_flock,
908 	.splice_read	= nfs_file_splice_read,
909 	.splice_write	= iter_file_splice_write,
910 	.check_flags	= nfs_check_flags,
911 	.setlease	= simple_nosetlease,
912 };
913 EXPORT_SYMBOL_GPL(nfs_file_operations);
914