Linux Audio

Check our new training course

Loading...
v6.9.4
  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
 46static const struct vm_operations_struct nfs_file_vm_ops;
 47
 48int 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}
 55EXPORT_SYMBOL_GPL(nfs_check_flags);
 56
 57/*
 58 * Open file
 59 */
 60static int
 61nfs_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
 78int
 79nfs_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}
 88EXPORT_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 */
101static 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;
110force_reval:
111	return __nfs_revalidate_inode(server, inode);
112}
113
114loff_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}
133EXPORT_SYMBOL_GPL(nfs_file_llseek);
134
135/*
136 * Flush all dirty pages, and check for write errors.
137 */
138static int
139nfs_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
156ssize_t
157nfs_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	nfs_start_io_read(inode);
 
 
 
170	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
171	if (!result) {
172		result = generic_file_read_iter(iocb, to);
173		if (result > 0)
174			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
175	}
176	nfs_end_io_read(inode);
177	return result;
178}
179EXPORT_SYMBOL_GPL(nfs_file_read);
180
181ssize_t
182nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
183		     size_t len, unsigned int flags)
184{
185	struct inode *inode = file_inode(in);
186	ssize_t result;
187
188	dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
189
190	nfs_start_io_read(inode);
 
 
 
191	result = nfs_revalidate_mapping(inode, in->f_mapping);
192	if (!result) {
193		result = filemap_splice_read(in, ppos, pipe, len, flags);
194		if (result > 0)
195			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
196	}
197	nfs_end_io_read(inode);
198	return result;
199}
200EXPORT_SYMBOL_GPL(nfs_file_splice_read);
201
202int
203nfs_file_mmap(struct file *file, struct vm_area_struct *vma)
204{
205	struct inode *inode = file_inode(file);
206	int	status;
207
208	dprintk("NFS: mmap(%pD2)\n", file);
209
210	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
211	 *       so we call that before revalidating the mapping
212	 */
213	status = generic_file_mmap(file, vma);
214	if (!status) {
215		vma->vm_ops = &nfs_file_vm_ops;
216		status = nfs_revalidate_mapping(inode, file->f_mapping);
217	}
218	return status;
219}
220EXPORT_SYMBOL_GPL(nfs_file_mmap);
221
222/*
223 * Flush any dirty pages for this process, and check for write errors.
224 * The return status from this call provides a reliable indication of
225 * whether any write errors occurred for this process.
226 */
227static int
228nfs_file_fsync_commit(struct file *file, int datasync)
229{
230	struct inode *inode = file_inode(file);
231	int ret, ret2;
232
233	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
234
235	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
236	ret = nfs_commit_inode(inode, FLUSH_SYNC);
237	ret2 = file_check_and_advance_wb_err(file);
238	if (ret2 < 0)
239		return ret2;
240	return ret;
241}
242
243int
244nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
245{
246	struct inode *inode = file_inode(file);
247	struct nfs_inode *nfsi = NFS_I(inode);
248	long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
249	long nredirtied;
250	int ret;
251
252	trace_nfs_fsync_enter(inode);
253
254	for (;;) {
255		ret = file_write_and_wait_range(file, start, end);
256		if (ret != 0)
257			break;
258		ret = nfs_file_fsync_commit(file, datasync);
259		if (ret != 0)
260			break;
261		ret = pnfs_sync_inode(inode, !!datasync);
262		if (ret != 0)
263			break;
264		nredirtied = atomic_long_read(&nfsi->redirtied_pages);
265		if (nredirtied == save_nredirtied)
266			break;
267		save_nredirtied = nredirtied;
268	}
269
270	trace_nfs_fsync_exit(inode, ret);
271	return ret;
272}
273EXPORT_SYMBOL_GPL(nfs_file_fsync);
274
275/*
276 * Decide whether a read/modify/write cycle may be more efficient
277 * then a modify/write/read cycle when writing to a page in the
278 * page cache.
279 *
280 * Some pNFS layout drivers can only read/write at a certain block
281 * granularity like all block devices and therefore we must perform
282 * read/modify/write whenever a page hasn't read yet and the data
283 * to be written there is not aligned to a block boundary and/or
284 * smaller than the block size.
285 *
286 * The modify/write/read cycle may occur if a page is read before
287 * being completely filled by the writer.  In this situation, the
288 * page must be completely written to stable storage on the server
289 * before it can be refilled by reading in the page from the server.
290 * This can lead to expensive, small, FILE_SYNC mode writes being
291 * done.
292 *
293 * It may be more efficient to read the page first if the file is
294 * open for reading in addition to writing, the page is not marked
295 * as Uptodate, it is not dirty or waiting to be committed,
296 * indicating that it was previously allocated and then modified,
297 * that there were valid bytes of data in that range of the file,
298 * and that the new data won't completely replace the old data in
299 * that range of the file.
300 */
301static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
302				    unsigned int len)
303{
304	unsigned int pglen = nfs_folio_length(folio);
305	unsigned int offset = offset_in_folio(folio, pos);
306	unsigned int end = offset + len;
307
308	return !pglen || (end >= pglen && !offset);
309}
310
311static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
312				       loff_t pos, unsigned int len)
313{
314	/*
315	 * Up-to-date pages, those with ongoing or full-page write
316	 * don't need read/modify/write
317	 */
318	if (folio_test_uptodate(folio) || folio_test_private(folio) ||
319	    nfs_folio_is_full_write(folio, pos, len))
320		return false;
321
322	if (pnfs_ld_read_whole_page(file_inode(file)))
323		return true;
324	/* Open for reading too? */
325	if (file->f_mode & FMODE_READ)
326		return true;
327	return false;
328}
329
330/*
331 * This does the "real" work of the write. We must allocate and lock the
332 * page to be sent back to the generic routine, which then copies the
333 * data from user space.
334 *
335 * If the writer ends up delaying the write, the writer needs to
336 * increment the page use counts until he is done with the page.
337 */
338static int nfs_write_begin(struct file *file, struct address_space *mapping,
339			   loff_t pos, unsigned len, struct page **pagep,
340			   void **fsdata)
341{
 
342	struct folio *folio;
343	int once_thru = 0;
344	int ret;
345
346	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
347		file, mapping->host->i_ino, len, (long long) pos);
348
 
349start:
350	folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, FGP_WRITEBEGIN,
351				    mapping_gfp_mask(mapping));
352	if (IS_ERR(folio))
353		return PTR_ERR(folio);
354	*pagep = &folio->page;
355
356	ret = nfs_flush_incompatible(file, folio);
357	if (ret) {
358		folio_unlock(folio);
359		folio_put(folio);
360	} else if (!once_thru &&
361		   nfs_want_read_modify_write(file, folio, pos, len)) {
362		once_thru = 1;
363		ret = nfs_read_folio(file, folio);
364		folio_put(folio);
365		if (!ret)
366			goto start;
367	}
368	return ret;
369}
370
371static int nfs_write_end(struct file *file, struct address_space *mapping,
372			 loff_t pos, unsigned len, unsigned copied,
373			 struct page *page, void *fsdata)
374{
375	struct nfs_open_context *ctx = nfs_file_open_context(file);
376	struct folio *folio = page_folio(page);
377	unsigned offset = offset_in_folio(folio, pos);
378	int status;
379
380	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
381		file, mapping->host->i_ino, len, (long long) pos);
382
383	/*
384	 * Zero any uninitialised parts of the page, and then mark the page
385	 * as up to date if it turns out that we're extending the file.
386	 */
387	if (!folio_test_uptodate(folio)) {
388		size_t fsize = folio_size(folio);
389		unsigned pglen = nfs_folio_length(folio);
390		unsigned end = offset + copied;
391
392		if (pglen == 0) {
393			folio_zero_segments(folio, 0, offset, end, fsize);
394			folio_mark_uptodate(folio);
395		} else if (end >= pglen) {
396			folio_zero_segment(folio, end, fsize);
397			if (offset == 0)
398				folio_mark_uptodate(folio);
399		} else
400			folio_zero_segment(folio, pglen, fsize);
401	}
402
403	status = nfs_update_folio(file, folio, offset, copied);
404
405	folio_unlock(folio);
406	folio_put(folio);
407
408	if (status < 0)
409		return status;
410	NFS_I(mapping->host)->write_io += copied;
411
412	if (nfs_ctx_key_to_expire(ctx, mapping->host))
413		nfs_wb_all(mapping->host);
414
415	return copied;
416}
417
418/*
419 * Partially or wholly invalidate a page
420 * - Release the private state associated with a page if undergoing complete
421 *   page invalidation
422 * - Called if either PG_private or PG_fscache is set on the page
423 * - Caller holds page lock
424 */
425static void nfs_invalidate_folio(struct folio *folio, size_t offset,
426				size_t length)
427{
428	struct inode *inode = folio_file_mapping(folio)->host;
429	dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
430		 folio->index, offset, length);
431
432	if (offset != 0 || length < folio_size(folio))
433		return;
434	/* Cancel any unstarted writes on this page */
435	nfs_wb_folio_cancel(inode, folio);
436	folio_wait_fscache(folio);
437	trace_nfs_invalidate_folio(inode, folio);
438}
439
440/*
441 * Attempt to release the private state associated with a folio
442 * - Called if either private or fscache flags are set on the folio
443 * - Caller holds folio lock
444 * - Return true (may release folio) or false (may not)
445 */
446static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
447{
448	dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
449
450	/* If the private flag is set, then the folio is not freeable */
451	if (folio_test_private(folio)) {
452		if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
453		    current_is_kswapd())
454			return false;
455		if (nfs_wb_folio(folio_file_mapping(folio)->host, folio) < 0)
456			return false;
457	}
458	return nfs_fscache_release_folio(folio, gfp);
459}
460
461static void nfs_check_dirty_writeback(struct folio *folio,
462				bool *dirty, bool *writeback)
463{
464	struct nfs_inode *nfsi;
465	struct address_space *mapping = folio->mapping;
466
467	/*
468	 * Check if an unstable folio is currently being committed and
469	 * if so, have the VM treat it as if the folio is under writeback
470	 * so it will not block due to folios that will shortly be freeable.
471	 */
472	nfsi = NFS_I(mapping->host);
473	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
474		*writeback = true;
475		return;
476	}
477
478	/*
479	 * If the private flag is set, then the folio is not freeable
480	 * and as the inode is not being committed, it's not going to
481	 * be cleaned in the near future so treat it as dirty
482	 */
483	if (folio_test_private(folio))
484		*dirty = true;
485}
486
487/*
488 * Attempt to clear the private state associated with a page when an error
489 * occurs that requires the cached contents of an inode to be written back or
490 * destroyed
491 * - Called if either PG_private or fscache is set on the page
492 * - Caller holds page lock
493 * - Return 0 if successful, -error otherwise
494 */
495static int nfs_launder_folio(struct folio *folio)
496{
497	struct inode *inode = folio->mapping->host;
498	int ret;
499
500	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
501		inode->i_ino, folio_pos(folio));
502
503	folio_wait_fscache(folio);
504	ret = nfs_wb_folio(inode, folio);
505	trace_nfs_launder_folio_done(inode, folio, ret);
 
506	return ret;
507}
508
509static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
510						sector_t *span)
511{
512	unsigned long blocks;
513	long long isize;
514	int ret;
515	struct inode *inode = file_inode(file);
516	struct rpc_clnt *clnt = NFS_CLIENT(inode);
517	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
518
519	spin_lock(&inode->i_lock);
520	blocks = inode->i_blocks;
521	isize = inode->i_size;
522	spin_unlock(&inode->i_lock);
523	if (blocks*512 < isize) {
524		pr_warn("swap activate: swapfile has holes\n");
525		return -EINVAL;
526	}
527
528	ret = rpc_clnt_swap_activate(clnt);
529	if (ret)
530		return ret;
531	ret = add_swap_extent(sis, 0, sis->max, 0);
532	if (ret < 0) {
533		rpc_clnt_swap_deactivate(clnt);
534		return ret;
535	}
536
537	*span = sis->pages;
538
539	if (cl->rpc_ops->enable_swap)
540		cl->rpc_ops->enable_swap(inode);
541
542	sis->flags |= SWP_FS_OPS;
543	return ret;
544}
545
546static void nfs_swap_deactivate(struct file *file)
547{
548	struct inode *inode = file_inode(file);
549	struct rpc_clnt *clnt = NFS_CLIENT(inode);
550	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
551
552	rpc_clnt_swap_deactivate(clnt);
553	if (cl->rpc_ops->disable_swap)
554		cl->rpc_ops->disable_swap(file_inode(file));
555}
556
557const struct address_space_operations nfs_file_aops = {
558	.read_folio = nfs_read_folio,
559	.readahead = nfs_readahead,
560	.dirty_folio = filemap_dirty_folio,
561	.writepages = nfs_writepages,
562	.write_begin = nfs_write_begin,
563	.write_end = nfs_write_end,
564	.invalidate_folio = nfs_invalidate_folio,
565	.release_folio = nfs_release_folio,
566	.migrate_folio = nfs_migrate_folio,
567	.launder_folio = nfs_launder_folio,
568	.is_dirty_writeback = nfs_check_dirty_writeback,
569	.error_remove_folio = generic_error_remove_folio,
570	.swap_activate = nfs_swap_activate,
571	.swap_deactivate = nfs_swap_deactivate,
572	.swap_rw = nfs_swap_rw,
573};
574
575/*
576 * Notification that a PTE pointing to an NFS page is about to be made
577 * writable, implying that someone is about to modify the page through a
578 * shared-writable mapping
579 */
580static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
581{
582	struct file *filp = vmf->vma->vm_file;
583	struct inode *inode = file_inode(filp);
584	unsigned pagelen;
585	vm_fault_t ret = VM_FAULT_NOPAGE;
586	struct address_space *mapping;
587	struct folio *folio = page_folio(vmf->page);
588
589	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
590		 filp, filp->f_mapping->host->i_ino,
591		 (long long)folio_file_pos(folio));
592
593	sb_start_pagefault(inode->i_sb);
594
595	/* make sure the cache has finished storing the page */
596	if (folio_test_fscache(folio) &&
597	    folio_wait_fscache_killable(folio) < 0) {
598		ret = VM_FAULT_RETRY;
599		goto out;
600	}
601
602	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
603			   nfs_wait_bit_killable,
604			   TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
605
606	folio_lock(folio);
607	mapping = folio_file_mapping(folio);
608	if (mapping != inode->i_mapping)
609		goto out_unlock;
610
611	folio_wait_writeback(folio);
612
613	pagelen = nfs_folio_length(folio);
614	if (pagelen == 0)
615		goto out_unlock;
616
617	ret = VM_FAULT_LOCKED;
618	if (nfs_flush_incompatible(filp, folio) == 0 &&
619	    nfs_update_folio(filp, folio, 0, pagelen) == 0)
620		goto out;
621
622	ret = VM_FAULT_SIGBUS;
623out_unlock:
624	folio_unlock(folio);
625out:
626	sb_end_pagefault(inode->i_sb);
627	return ret;
628}
629
630static const struct vm_operations_struct nfs_file_vm_ops = {
631	.fault = filemap_fault,
632	.map_pages = filemap_map_pages,
633	.page_mkwrite = nfs_vm_page_mkwrite,
634};
635
636ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
637{
638	struct file *file = iocb->ki_filp;
639	struct inode *inode = file_inode(file);
640	unsigned int mntflags = NFS_SERVER(inode)->flags;
641	ssize_t result, written;
642	errseq_t since;
643	int error;
644
645	result = nfs_key_timeout_notify(file, inode);
646	if (result)
647		return result;
648
649	if (iocb->ki_flags & IOCB_DIRECT)
650		return nfs_file_direct_write(iocb, from, false);
651
652	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
653		file, iov_iter_count(from), (long long) iocb->ki_pos);
654
655	if (IS_SWAPFILE(inode))
656		goto out_swapfile;
657	/*
658	 * O_APPEND implies that we must revalidate the file length.
659	 */
660	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
661		result = nfs_revalidate_file_size(inode, file);
662		if (result)
663			return result;
664	}
665
666	nfs_clear_invalid_mapping(file->f_mapping);
667
668	since = filemap_sample_wb_err(file->f_mapping);
669	nfs_start_io_write(inode);
 
 
670	result = generic_write_checks(iocb, from);
671	if (result > 0)
672		result = generic_perform_write(iocb, from);
673	nfs_end_io_write(inode);
674	if (result <= 0)
675		goto out;
676
677	written = result;
678	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
679
680	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
681		result = filemap_fdatawrite_range(file->f_mapping,
682						  iocb->ki_pos - written,
683						  iocb->ki_pos - 1);
684		if (result < 0)
685			goto out;
686	}
687	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
688		filemap_fdatawait_range(file->f_mapping,
689					iocb->ki_pos - written,
690					iocb->ki_pos - 1);
691	}
692	result = generic_write_sync(iocb, written);
693	if (result < 0)
694		return result;
695
696out:
697	/* Return error values */
698	error = filemap_check_wb_err(file->f_mapping, since);
699	switch (error) {
700	default:
701		break;
702	case -EDQUOT:
703	case -EFBIG:
704	case -ENOSPC:
705		nfs_wb_all(inode);
706		error = file_check_and_advance_wb_err(file);
707		if (error < 0)
708			result = error;
709	}
710	return result;
711
712out_swapfile:
713	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
714	return -ETXTBSY;
715}
716EXPORT_SYMBOL_GPL(nfs_file_write);
717
718static int
719do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
720{
721	struct inode *inode = filp->f_mapping->host;
722	int status = 0;
723	unsigned int saved_type = fl->c.flc_type;
724
725	/* Try local locking first */
726	posix_test_lock(filp, fl);
727	if (fl->c.flc_type != F_UNLCK) {
728		/* found a conflict */
729		goto out;
730	}
731	fl->c.flc_type = saved_type;
732
733	if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
734		goto out_noconflict;
735
736	if (is_local)
737		goto out_noconflict;
738
739	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
740out:
741	return status;
742out_noconflict:
743	fl->c.flc_type = F_UNLCK;
744	goto out;
745}
746
747static int
748do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
749{
750	struct inode *inode = filp->f_mapping->host;
751	struct nfs_lock_context *l_ctx;
752	int status;
753
754	/*
755	 * Flush all pending writes before doing anything
756	 * with locks..
757	 */
758	nfs_wb_all(inode);
759
760	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
761	if (!IS_ERR(l_ctx)) {
762		status = nfs_iocounter_wait(l_ctx);
763		nfs_put_lock_context(l_ctx);
764		/*  NOTE: special case
765		 * 	If we're signalled while cleaning up locks on process exit, we
766		 * 	still need to complete the unlock.
767		 */
768		if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
769			return status;
770	}
771
772	/*
773	 * Use local locking if mounted with "-onolock" or with appropriate
774	 * "-olocal_lock="
775	 */
776	if (!is_local)
777		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
778	else
779		status = locks_lock_file_wait(filp, fl);
780	return status;
781}
782
783static int
784do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
785{
786	struct inode *inode = filp->f_mapping->host;
787	int status;
788
789	/*
790	 * Flush all pending writes before doing anything
791	 * with locks..
792	 */
793	status = nfs_sync_mapping(filp->f_mapping);
794	if (status != 0)
795		goto out;
796
797	/*
798	 * Use local locking if mounted with "-onolock" or with appropriate
799	 * "-olocal_lock="
800	 */
801	if (!is_local)
802		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
803	else
804		status = locks_lock_file_wait(filp, fl);
805	if (status < 0)
806		goto out;
807
808	/*
809	 * Invalidate cache to prevent missing any changes.  If
810	 * the file is mapped, clear the page cache as well so
811	 * those mappings will be loaded.
812	 *
813	 * This makes locking act as a cache coherency point.
814	 */
815	nfs_sync_mapping(filp->f_mapping);
816	if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
817		nfs_zap_caches(inode);
818		if (mapping_mapped(filp->f_mapping))
819			nfs_revalidate_mapping(inode, filp->f_mapping);
820	}
821out:
822	return status;
823}
824
825/*
826 * Lock a (portion of) a file
827 */
828int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
829{
830	struct inode *inode = filp->f_mapping->host;
831	int ret = -ENOLCK;
832	int is_local = 0;
833
834	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
835			filp, fl->c.flc_type, fl->c.flc_flags,
836			(long long)fl->fl_start, (long long)fl->fl_end);
837
838	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
839
840	if (fl->c.flc_flags & FL_RECLAIM)
841		return -ENOGRACE;
842
843	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
844		is_local = 1;
845
846	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
847		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
848		if (ret < 0)
849			goto out_err;
850	}
851
852	if (IS_GETLK(cmd))
853		ret = do_getlk(filp, cmd, fl, is_local);
854	else if (lock_is_unlock(fl))
855		ret = do_unlk(filp, cmd, fl, is_local);
856	else
857		ret = do_setlk(filp, cmd, fl, is_local);
858out_err:
859	return ret;
860}
861EXPORT_SYMBOL_GPL(nfs_lock);
862
863/*
864 * Lock a (portion of) a file
865 */
866int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
867{
868	struct inode *inode = filp->f_mapping->host;
869	int is_local = 0;
870
871	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
872			filp, fl->c.flc_type, fl->c.flc_flags);
873
874	if (!(fl->c.flc_flags & FL_FLOCK))
875		return -ENOLCK;
876
877	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
878		is_local = 1;
879
880	/* We're simulating flock() locks using posix locks on the server */
881	if (lock_is_unlock(fl))
882		return do_unlk(filp, cmd, fl, is_local);
883	return do_setlk(filp, cmd, fl, is_local);
884}
885EXPORT_SYMBOL_GPL(nfs_flock);
886
887const struct file_operations nfs_file_operations = {
888	.llseek		= nfs_file_llseek,
889	.read_iter	= nfs_file_read,
890	.write_iter	= nfs_file_write,
891	.mmap		= nfs_file_mmap,
892	.open		= nfs_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	= iter_file_splice_write,
900	.check_flags	= nfs_check_flags,
901	.setlease	= simple_nosetlease,
902};
903EXPORT_SYMBOL_GPL(nfs_file_operations);
v6.13.7
  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#include <linux/compaction.h>
 33
 34#include <linux/uaccess.h>
 35#include <linux/filelock.h>
 36
 37#include "delegation.h"
 38#include "internal.h"
 39#include "iostat.h"
 40#include "fscache.h"
 41#include "pnfs.h"
 42
 43#include "nfstrace.h"
 44
 45#define NFSDBG_FACILITY		NFSDBG_FILE
 46
 47static const struct vm_operations_struct nfs_file_vm_ops;
 48
 49int nfs_check_flags(int flags)
 50{
 51	if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 52		return -EINVAL;
 53
 54	return 0;
 55}
 56EXPORT_SYMBOL_GPL(nfs_check_flags);
 57
 58/*
 59 * Open file
 60 */
 61static int
 62nfs_file_open(struct inode *inode, struct file *filp)
 63{
 64	int res;
 65
 66	dprintk("NFS: open file(%pD2)\n", filp);
 67
 68	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 69	res = nfs_check_flags(filp->f_flags);
 70	if (res)
 71		return res;
 72
 73	res = nfs_open(inode, filp);
 74	if (res == 0)
 75		filp->f_mode |= FMODE_CAN_ODIRECT;
 76	return res;
 77}
 78
 79int
 80nfs_file_release(struct inode *inode, struct file *filp)
 81{
 82	dprintk("NFS: release(%pD2)\n", filp);
 83
 84	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 85	nfs_file_clear_open_context(filp);
 86	nfs_fscache_release_file(inode, filp);
 87	return 0;
 88}
 89EXPORT_SYMBOL_GPL(nfs_file_release);
 90
 91/**
 92 * nfs_revalidate_file_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 */
102static 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_INVALID_SIZE))
109		goto force_reval;
110	return 0;
111force_reval:
112	return __nfs_revalidate_inode(server, inode);
113}
114
115loff_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}
134EXPORT_SYMBOL_GPL(nfs_file_llseek);
135
136/*
137 * Flush all dirty pages, and check for write errors.
138 */
139static int
140nfs_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
157ssize_t
158nfs_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	result = nfs_start_io_read(inode);
171	if (result)
172		return result;
173
174	result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
175	if (!result) {
176		result = generic_file_read_iter(iocb, to);
177		if (result > 0)
178			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
179	}
180	nfs_end_io_read(inode);
181	return result;
182}
183EXPORT_SYMBOL_GPL(nfs_file_read);
184
185ssize_t
186nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
187		     size_t len, unsigned int flags)
188{
189	struct inode *inode = file_inode(in);
190	ssize_t result;
191
192	dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
193
194	result = nfs_start_io_read(inode);
195	if (result)
196		return result;
197
198	result = nfs_revalidate_mapping(inode, in->f_mapping);
199	if (!result) {
200		result = filemap_splice_read(in, ppos, pipe, len, flags);
201		if (result > 0)
202			nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
203	}
204	nfs_end_io_read(inode);
205	return result;
206}
207EXPORT_SYMBOL_GPL(nfs_file_splice_read);
208
209int
210nfs_file_mmap(struct file *file, struct vm_area_struct *vma)
211{
212	struct inode *inode = file_inode(file);
213	int	status;
214
215	dprintk("NFS: mmap(%pD2)\n", file);
216
217	/* Note: generic_file_mmap() returns ENOSYS on nommu systems
218	 *       so we call that before revalidating the mapping
219	 */
220	status = generic_file_mmap(file, vma);
221	if (!status) {
222		vma->vm_ops = &nfs_file_vm_ops;
223		status = nfs_revalidate_mapping(inode, file->f_mapping);
224	}
225	return status;
226}
227EXPORT_SYMBOL_GPL(nfs_file_mmap);
228
229/*
230 * Flush any dirty pages for this process, and check for write errors.
231 * The return status from this call provides a reliable indication of
232 * whether any write errors occurred for this process.
233 */
234static int
235nfs_file_fsync_commit(struct file *file, int datasync)
236{
237	struct inode *inode = file_inode(file);
238	int ret, ret2;
239
240	dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
241
242	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
243	ret = nfs_commit_inode(inode, FLUSH_SYNC);
244	ret2 = file_check_and_advance_wb_err(file);
245	if (ret2 < 0)
246		return ret2;
247	return ret;
248}
249
250int
251nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
252{
253	struct inode *inode = file_inode(file);
254	struct nfs_inode *nfsi = NFS_I(inode);
255	long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
256	long nredirtied;
257	int ret;
258
259	trace_nfs_fsync_enter(inode);
260
261	for (;;) {
262		ret = file_write_and_wait_range(file, start, end);
263		if (ret != 0)
264			break;
265		ret = nfs_file_fsync_commit(file, datasync);
266		if (ret != 0)
267			break;
268		ret = pnfs_sync_inode(inode, !!datasync);
269		if (ret != 0)
270			break;
271		nredirtied = atomic_long_read(&nfsi->redirtied_pages);
272		if (nredirtied == save_nredirtied)
273			break;
274		save_nredirtied = nredirtied;
275	}
276
277	trace_nfs_fsync_exit(inode, ret);
278	return ret;
279}
280EXPORT_SYMBOL_GPL(nfs_file_fsync);
281
282/*
283 * Decide whether a read/modify/write cycle may be more efficient
284 * then a modify/write/read cycle when writing to a page in the
285 * page cache.
286 *
287 * Some pNFS layout drivers can only read/write at a certain block
288 * granularity like all block devices and therefore we must perform
289 * read/modify/write whenever a page hasn't read yet and the data
290 * to be written there is not aligned to a block boundary and/or
291 * smaller than the block size.
292 *
293 * The modify/write/read cycle may occur if a page is read before
294 * being completely filled by the writer.  In this situation, the
295 * page must be completely written to stable storage on the server
296 * before it can be refilled by reading in the page from the server.
297 * This can lead to expensive, small, FILE_SYNC mode writes being
298 * done.
299 *
300 * It may be more efficient to read the page first if the file is
301 * open for reading in addition to writing, the page is not marked
302 * as Uptodate, it is not dirty or waiting to be committed,
303 * indicating that it was previously allocated and then modified,
304 * that there were valid bytes of data in that range of the file,
305 * and that the new data won't completely replace the old data in
306 * that range of the file.
307 */
308static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
309				    unsigned int len)
310{
311	unsigned int pglen = nfs_folio_length(folio);
312	unsigned int offset = offset_in_folio(folio, pos);
313	unsigned int end = offset + len;
314
315	return !pglen || (end >= pglen && !offset);
316}
317
318static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
319				       loff_t pos, unsigned int len)
320{
321	/*
322	 * Up-to-date pages, those with ongoing or full-page write
323	 * don't need read/modify/write
324	 */
325	if (folio_test_uptodate(folio) || folio_test_private(folio) ||
326	    nfs_folio_is_full_write(folio, pos, len))
327		return false;
328
329	if (pnfs_ld_read_whole_page(file_inode(file)))
330		return true;
331	/* Open for reading too? */
332	if (file->f_mode & FMODE_READ)
333		return true;
334	return false;
335}
336
337/*
338 * This does the "real" work of the write. We must allocate and lock the
339 * page to be sent back to the generic routine, which then copies the
340 * data from user space.
341 *
342 * If the writer ends up delaying the write, the writer needs to
343 * increment the page use counts until he is done with the page.
344 */
345static int nfs_write_begin(struct file *file, struct address_space *mapping,
346			   loff_t pos, unsigned len, struct folio **foliop,
347			   void **fsdata)
348{
349	fgf_t fgp = FGP_WRITEBEGIN;
350	struct folio *folio;
351	int once_thru = 0;
352	int ret;
353
354	dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
355		file, mapping->host->i_ino, len, (long long) pos);
356
357	fgp |= fgf_set_order(len);
358start:
359	folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp,
360				    mapping_gfp_mask(mapping));
361	if (IS_ERR(folio))
362		return PTR_ERR(folio);
363	*foliop = folio;
364
365	ret = nfs_flush_incompatible(file, folio);
366	if (ret) {
367		folio_unlock(folio);
368		folio_put(folio);
369	} else if (!once_thru &&
370		   nfs_want_read_modify_write(file, folio, pos, len)) {
371		once_thru = 1;
372		ret = nfs_read_folio(file, folio);
373		folio_put(folio);
374		if (!ret)
375			goto start;
376	}
377	return ret;
378}
379
380static int nfs_write_end(struct file *file, struct address_space *mapping,
381			 loff_t pos, unsigned len, unsigned copied,
382			 struct folio *folio, void *fsdata)
383{
384	struct nfs_open_context *ctx = nfs_file_open_context(file);
 
385	unsigned offset = offset_in_folio(folio, pos);
386	int status;
387
388	dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
389		file, mapping->host->i_ino, len, (long long) pos);
390
391	/*
392	 * Zero any uninitialised parts of the page, and then mark the page
393	 * as up to date if it turns out that we're extending the file.
394	 */
395	if (!folio_test_uptodate(folio)) {
396		size_t fsize = folio_size(folio);
397		unsigned pglen = nfs_folio_length(folio);
398		unsigned end = offset + copied;
399
400		if (pglen == 0) {
401			folio_zero_segments(folio, 0, offset, end, fsize);
402			folio_mark_uptodate(folio);
403		} else if (end >= pglen) {
404			folio_zero_segment(folio, end, fsize);
405			if (offset == 0)
406				folio_mark_uptodate(folio);
407		} else
408			folio_zero_segment(folio, pglen, fsize);
409	}
410
411	status = nfs_update_folio(file, folio, offset, copied);
412
413	folio_unlock(folio);
414	folio_put(folio);
415
416	if (status < 0)
417		return status;
418	NFS_I(mapping->host)->write_io += copied;
419
420	if (nfs_ctx_key_to_expire(ctx, mapping->host))
421		nfs_wb_all(mapping->host);
422
423	return copied;
424}
425
426/*
427 * Partially or wholly invalidate a page
428 * - Release the private state associated with a page if undergoing complete
429 *   page invalidation
430 * - Called if either PG_private or PG_fscache is set on the page
431 * - Caller holds page lock
432 */
433static void nfs_invalidate_folio(struct folio *folio, size_t offset,
434				size_t length)
435{
436	struct inode *inode = folio->mapping->host;
437	dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
438		 folio->index, offset, length);
439
440	if (offset != 0 || length < folio_size(folio))
441		return;
442	/* Cancel any unstarted writes on this page */
443	nfs_wb_folio_cancel(inode, folio);
444	folio_wait_private_2(folio); /* [DEPRECATED] */
445	trace_nfs_invalidate_folio(inode, folio_pos(folio) + offset, length);
446}
447
448/*
449 * Attempt to release the private state associated with a folio
450 * - Called if either private or fscache flags are set on the folio
451 * - Caller holds folio lock
452 * - Return true (may release folio) or false (may not)
453 */
454static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
455{
456	dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
457
458	/* If the private flag is set, then the folio is not freeable */
459	if (folio_test_private(folio)) {
460		if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
461		    current_is_kswapd() || current_is_kcompactd())
462			return false;
463		if (nfs_wb_folio(folio->mapping->host, folio) < 0)
464			return false;
465	}
466	return nfs_fscache_release_folio(folio, gfp);
467}
468
469static void nfs_check_dirty_writeback(struct folio *folio,
470				bool *dirty, bool *writeback)
471{
472	struct nfs_inode *nfsi;
473	struct address_space *mapping = folio->mapping;
474
475	/*
476	 * Check if an unstable folio is currently being committed and
477	 * if so, have the VM treat it as if the folio is under writeback
478	 * so it will not block due to folios that will shortly be freeable.
479	 */
480	nfsi = NFS_I(mapping->host);
481	if (atomic_read(&nfsi->commit_info.rpcs_out)) {
482		*writeback = true;
483		return;
484	}
485
486	/*
487	 * If the private flag is set, then the folio is not freeable
488	 * and as the inode is not being committed, it's not going to
489	 * be cleaned in the near future so treat it as dirty
490	 */
491	if (folio_test_private(folio))
492		*dirty = true;
493}
494
495/*
496 * Attempt to clear the private state associated with a page when an error
497 * occurs that requires the cached contents of an inode to be written back or
498 * destroyed
499 * - Called if either PG_private or fscache is set on the page
500 * - Caller holds page lock
501 * - Return 0 if successful, -error otherwise
502 */
503static int nfs_launder_folio(struct folio *folio)
504{
505	struct inode *inode = folio->mapping->host;
506	int ret;
507
508	dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
509		inode->i_ino, folio_pos(folio));
510
511	folio_wait_private_2(folio); /* [DEPRECATED] */
512	ret = nfs_wb_folio(inode, folio);
513	trace_nfs_launder_folio_done(inode, folio_pos(folio),
514			folio_size(folio), ret);
515	return ret;
516}
517
518static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
519						sector_t *span)
520{
521	unsigned long blocks;
522	long long isize;
523	int ret;
524	struct inode *inode = file_inode(file);
525	struct rpc_clnt *clnt = NFS_CLIENT(inode);
526	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
527
528	spin_lock(&inode->i_lock);
529	blocks = inode->i_blocks;
530	isize = inode->i_size;
531	spin_unlock(&inode->i_lock);
532	if (blocks*512 < isize) {
533		pr_warn("swap activate: swapfile has holes\n");
534		return -EINVAL;
535	}
536
537	ret = rpc_clnt_swap_activate(clnt);
538	if (ret)
539		return ret;
540	ret = add_swap_extent(sis, 0, sis->max, 0);
541	if (ret < 0) {
542		rpc_clnt_swap_deactivate(clnt);
543		return ret;
544	}
545
546	*span = sis->pages;
547
548	if (cl->rpc_ops->enable_swap)
549		cl->rpc_ops->enable_swap(inode);
550
551	sis->flags |= SWP_FS_OPS;
552	return ret;
553}
554
555static void nfs_swap_deactivate(struct file *file)
556{
557	struct inode *inode = file_inode(file);
558	struct rpc_clnt *clnt = NFS_CLIENT(inode);
559	struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
560
561	rpc_clnt_swap_deactivate(clnt);
562	if (cl->rpc_ops->disable_swap)
563		cl->rpc_ops->disable_swap(file_inode(file));
564}
565
566const struct address_space_operations nfs_file_aops = {
567	.read_folio = nfs_read_folio,
568	.readahead = nfs_readahead,
569	.dirty_folio = filemap_dirty_folio,
570	.writepages = nfs_writepages,
571	.write_begin = nfs_write_begin,
572	.write_end = nfs_write_end,
573	.invalidate_folio = nfs_invalidate_folio,
574	.release_folio = nfs_release_folio,
575	.migrate_folio = nfs_migrate_folio,
576	.launder_folio = nfs_launder_folio,
577	.is_dirty_writeback = nfs_check_dirty_writeback,
578	.error_remove_folio = generic_error_remove_folio,
579	.swap_activate = nfs_swap_activate,
580	.swap_deactivate = nfs_swap_deactivate,
581	.swap_rw = nfs_swap_rw,
582};
583
584/*
585 * Notification that a PTE pointing to an NFS page is about to be made
586 * writable, implying that someone is about to modify the page through a
587 * shared-writable mapping
588 */
589static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
590{
591	struct file *filp = vmf->vma->vm_file;
592	struct inode *inode = file_inode(filp);
593	unsigned pagelen;
594	vm_fault_t ret = VM_FAULT_NOPAGE;
595	struct address_space *mapping;
596	struct folio *folio = page_folio(vmf->page);
597
598	dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
599		 filp, filp->f_mapping->host->i_ino,
600		 (long long)folio_pos(folio));
601
602	sb_start_pagefault(inode->i_sb);
603
604	/* make sure the cache has finished storing the page */
605	if (folio_test_private_2(folio) && /* [DEPRECATED] */
606	    folio_wait_private_2_killable(folio) < 0) {
607		ret = VM_FAULT_RETRY;
608		goto out;
609	}
610
611	wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
612			   nfs_wait_bit_killable,
613			   TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
614
615	folio_lock(folio);
616	mapping = folio->mapping;
617	if (mapping != inode->i_mapping)
618		goto out_unlock;
619
620	folio_wait_writeback(folio);
621
622	pagelen = nfs_folio_length(folio);
623	if (pagelen == 0)
624		goto out_unlock;
625
626	ret = VM_FAULT_LOCKED;
627	if (nfs_flush_incompatible(filp, folio) == 0 &&
628	    nfs_update_folio(filp, folio, 0, pagelen) == 0)
629		goto out;
630
631	ret = VM_FAULT_SIGBUS;
632out_unlock:
633	folio_unlock(folio);
634out:
635	sb_end_pagefault(inode->i_sb);
636	return ret;
637}
638
639static const struct vm_operations_struct nfs_file_vm_ops = {
640	.fault = filemap_fault,
641	.map_pages = filemap_map_pages,
642	.page_mkwrite = nfs_vm_page_mkwrite,
643};
644
645ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
646{
647	struct file *file = iocb->ki_filp;
648	struct inode *inode = file_inode(file);
649	unsigned int mntflags = NFS_SERVER(inode)->flags;
650	ssize_t result, written;
651	errseq_t since;
652	int error;
653
654	result = nfs_key_timeout_notify(file, inode);
655	if (result)
656		return result;
657
658	if (iocb->ki_flags & IOCB_DIRECT)
659		return nfs_file_direct_write(iocb, from, false);
660
661	dprintk("NFS: write(%pD2, %zu@%Ld)\n",
662		file, iov_iter_count(from), (long long) iocb->ki_pos);
663
664	if (IS_SWAPFILE(inode))
665		goto out_swapfile;
666	/*
667	 * O_APPEND implies that we must revalidate the file length.
668	 */
669	if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
670		result = nfs_revalidate_file_size(inode, file);
671		if (result)
672			return result;
673	}
674
675	nfs_clear_invalid_mapping(file->f_mapping);
676
677	since = filemap_sample_wb_err(file->f_mapping);
678	error = nfs_start_io_write(inode);
679	if (error)
680		return error;
681	result = generic_write_checks(iocb, from);
682	if (result > 0)
683		result = generic_perform_write(iocb, from);
684	nfs_end_io_write(inode);
685	if (result <= 0)
686		goto out;
687
688	written = result;
689	nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
690
691	if (mntflags & NFS_MOUNT_WRITE_EAGER) {
692		result = filemap_fdatawrite_range(file->f_mapping,
693						  iocb->ki_pos - written,
694						  iocb->ki_pos - 1);
695		if (result < 0)
696			goto out;
697	}
698	if (mntflags & NFS_MOUNT_WRITE_WAIT) {
699		filemap_fdatawait_range(file->f_mapping,
700					iocb->ki_pos - written,
701					iocb->ki_pos - 1);
702	}
703	result = generic_write_sync(iocb, written);
704	if (result < 0)
705		return result;
706
707out:
708	/* Return error values */
709	error = filemap_check_wb_err(file->f_mapping, since);
710	switch (error) {
711	default:
712		break;
713	case -EDQUOT:
714	case -EFBIG:
715	case -ENOSPC:
716		nfs_wb_all(inode);
717		error = file_check_and_advance_wb_err(file);
718		if (error < 0)
719			result = error;
720	}
721	return result;
722
723out_swapfile:
724	printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
725	return -ETXTBSY;
726}
727EXPORT_SYMBOL_GPL(nfs_file_write);
728
729static int
730do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
731{
732	struct inode *inode = filp->f_mapping->host;
733	int status = 0;
734	unsigned int saved_type = fl->c.flc_type;
735
736	/* Try local locking first */
737	posix_test_lock(filp, fl);
738	if (fl->c.flc_type != F_UNLCK) {
739		/* found a conflict */
740		goto out;
741	}
742	fl->c.flc_type = saved_type;
743
744	if (nfs_have_read_or_write_delegation(inode))
745		goto out_noconflict;
746
747	if (is_local)
748		goto out_noconflict;
749
750	status = NFS_PROTO(inode)->lock(filp, cmd, fl);
751out:
752	return status;
753out_noconflict:
754	fl->c.flc_type = F_UNLCK;
755	goto out;
756}
757
758static int
759do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
760{
761	struct inode *inode = filp->f_mapping->host;
762	struct nfs_lock_context *l_ctx;
763	int status;
764
765	/*
766	 * Flush all pending writes before doing anything
767	 * with locks..
768	 */
769	nfs_wb_all(inode);
770
771	l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
772	if (!IS_ERR(l_ctx)) {
773		status = nfs_iocounter_wait(l_ctx);
774		nfs_put_lock_context(l_ctx);
775		/*  NOTE: special case
776		 * 	If we're signalled while cleaning up locks on process exit, we
777		 * 	still need to complete the unlock.
778		 */
779		if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
780			return status;
781	}
782
783	/*
784	 * Use local locking if mounted with "-onolock" or with appropriate
785	 * "-olocal_lock="
786	 */
787	if (!is_local)
788		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
789	else
790		status = locks_lock_file_wait(filp, fl);
791	return status;
792}
793
794static int
795do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
796{
797	struct inode *inode = filp->f_mapping->host;
798	int status;
799
800	/*
801	 * Flush all pending writes before doing anything
802	 * with locks..
803	 */
804	status = nfs_sync_mapping(filp->f_mapping);
805	if (status != 0)
806		goto out;
807
808	/*
809	 * Use local locking if mounted with "-onolock" or with appropriate
810	 * "-olocal_lock="
811	 */
812	if (!is_local)
813		status = NFS_PROTO(inode)->lock(filp, cmd, fl);
814	else
815		status = locks_lock_file_wait(filp, fl);
816	if (status < 0)
817		goto out;
818
819	/*
820	 * Invalidate cache to prevent missing any changes.  If
821	 * the file is mapped, clear the page cache as well so
822	 * those mappings will be loaded.
823	 *
824	 * This makes locking act as a cache coherency point.
825	 */
826	nfs_sync_mapping(filp->f_mapping);
827	if (!nfs_have_read_or_write_delegation(inode)) {
828		nfs_zap_caches(inode);
829		if (mapping_mapped(filp->f_mapping))
830			nfs_revalidate_mapping(inode, filp->f_mapping);
831	}
832out:
833	return status;
834}
835
836/*
837 * Lock a (portion of) a file
838 */
839int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
840{
841	struct inode *inode = filp->f_mapping->host;
842	int ret = -ENOLCK;
843	int is_local = 0;
844
845	dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
846			filp, fl->c.flc_type, fl->c.flc_flags,
847			(long long)fl->fl_start, (long long)fl->fl_end);
848
849	nfs_inc_stats(inode, NFSIOS_VFSLOCK);
850
851	if (fl->c.flc_flags & FL_RECLAIM)
852		return -ENOGRACE;
853
854	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
855		is_local = 1;
856
857	if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
858		ret = NFS_PROTO(inode)->lock_check_bounds(fl);
859		if (ret < 0)
860			goto out_err;
861	}
862
863	if (IS_GETLK(cmd))
864		ret = do_getlk(filp, cmd, fl, is_local);
865	else if (lock_is_unlock(fl))
866		ret = do_unlk(filp, cmd, fl, is_local);
867	else
868		ret = do_setlk(filp, cmd, fl, is_local);
869out_err:
870	return ret;
871}
872EXPORT_SYMBOL_GPL(nfs_lock);
873
874/*
875 * Lock a (portion of) a file
876 */
877int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
878{
879	struct inode *inode = filp->f_mapping->host;
880	int is_local = 0;
881
882	dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
883			filp, fl->c.flc_type, fl->c.flc_flags);
884
885	if (!(fl->c.flc_flags & FL_FLOCK))
886		return -ENOLCK;
887
888	if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
889		is_local = 1;
890
891	/* We're simulating flock() locks using posix locks on the server */
892	if (lock_is_unlock(fl))
893		return do_unlk(filp, cmd, fl, is_local);
894	return do_setlk(filp, cmd, fl, is_local);
895}
896EXPORT_SYMBOL_GPL(nfs_flock);
897
898const struct file_operations nfs_file_operations = {
899	.llseek		= nfs_file_llseek,
900	.read_iter	= nfs_file_read,
901	.write_iter	= nfs_file_write,
902	.mmap		= nfs_file_mmap,
903	.open		= nfs_file_open,
904	.flush		= nfs_file_flush,
905	.release	= nfs_file_release,
906	.fsync		= nfs_file_fsync,
907	.lock		= nfs_lock,
908	.flock		= nfs_flock,
909	.splice_read	= nfs_file_splice_read,
910	.splice_write	= iter_file_splice_write,
911	.check_flags	= nfs_check_flags,
912	.setlease	= simple_nosetlease,
913};
914EXPORT_SYMBOL_GPL(nfs_file_operations);