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/rmap.h>
32#include <linux/swap.h>
33#include <linux/compaction.h>
34
35#include <linux/uaccess.h>
36#include <linux/filelock.h>
37
38#include "delegation.h"
39#include "internal.h"
40#include "iostat.h"
41#include "fscache.h"
42#include "pnfs.h"
43
44#include "nfstrace.h"
45
46#define NFSDBG_FACILITY NFSDBG_FILE
47
48static const struct vm_operations_struct nfs_file_vm_ops;
49
50int nfs_check_flags(int flags)
51{
52 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
53 return -EINVAL;
54
55 return 0;
56}
57EXPORT_SYMBOL_GPL(nfs_check_flags);
58
59/*
60 * Open file
61 */
62static int
63nfs_file_open(struct inode *inode, struct file *filp)
64{
65 int res;
66
67 dprintk("NFS: open file(%pD2)\n", filp);
68
69 nfs_inc_stats(inode, stat: NFSIOS_VFSOPEN);
70 res = nfs_check_flags(filp->f_flags);
71 if (res)
72 return res;
73
74 res = nfs_open(inode, filp);
75 if (res == 0)
76 filp->f_mode |= FMODE_CAN_ODIRECT;
77 return res;
78}
79
80int
81nfs_file_release(struct inode *inode, struct file *filp)
82{
83 dprintk("NFS: release(%pD2)\n", filp);
84
85 nfs_inc_stats(inode, stat: NFSIOS_VFSRELEASE);
86 nfs_file_clear_open_context(flip: filp);
87 nfs_fscache_release_file(inode, file: filp);
88 return 0;
89}
90EXPORT_SYMBOL_GPL(nfs_file_release);
91
92/**
93 * nfs_revalidate_file_size - Revalidate the file size
94 * @inode: pointer to inode struct
95 * @filp: pointer to struct file
96 *
97 * Revalidates the file length. This is basically a wrapper around
98 * nfs_revalidate_inode() that takes into account the fact that we may
99 * have cached writes (in which case we don't care about the server's
100 * idea of what the file length is), or O_DIRECT (in which case we
101 * shouldn't trust the cache).
102 */
103static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
104{
105 struct nfs_server *server = NFS_SERVER(inode);
106
107 if (filp->f_flags & O_DIRECT)
108 goto force_reval;
109 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
110 goto force_reval;
111 return 0;
112force_reval:
113 return __nfs_revalidate_inode(server, inode);
114}
115
116loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
117{
118 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
119 filp, offset, whence);
120
121 /*
122 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
123 * the cached file length
124 */
125 if (whence != SEEK_SET && whence != SEEK_CUR) {
126 struct inode *inode = filp->f_mapping->host;
127
128 int retval = nfs_revalidate_file_size(inode, filp);
129 if (retval < 0)
130 return (loff_t)retval;
131 }
132
133 return generic_file_llseek(file: filp, offset, whence);
134}
135EXPORT_SYMBOL_GPL(nfs_file_llseek);
136
137/*
138 * Flush all dirty pages, and check for write errors.
139 */
140static int
141nfs_file_flush(struct file *file, fl_owner_t id)
142{
143 struct inode *inode = file_inode(f: file);
144 errseq_t since;
145
146 dprintk("NFS: flush(%pD2)\n", file);
147
148 nfs_inc_stats(inode, stat: NFSIOS_VFSFLUSH);
149 if ((file->f_mode & FMODE_WRITE) == 0)
150 return 0;
151
152 /* Flush writes to the server and return any errors */
153 since = filemap_sample_wb_err(mapping: file->f_mapping);
154 nfs_wb_all(inode);
155 return filemap_check_wb_err(mapping: file->f_mapping, since);
156}
157
158ssize_t
159nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
160{
161 struct inode *inode = file_inode(f: iocb->ki_filp);
162 ssize_t result;
163
164 trace_nfs_file_read(iocb, iter: to);
165
166 if (iocb->ki_flags & IOCB_DIRECT)
167 return nfs_file_direct_read(iocb, iter: to, swap: false);
168
169 dprintk("NFS: read(%pD2, %zu@%lu)\n",
170 iocb->ki_filp,
171 iov_iter_count(to), (unsigned long) iocb->ki_pos);
172
173 result = nfs_start_io_read(inode);
174 if (result)
175 return result;
176
177 result = nfs_revalidate_mapping(inode, mapping: iocb->ki_filp->f_mapping);
178 if (!result) {
179 result = generic_file_read_iter(iocb, to);
180 if (result > 0)
181 nfs_add_stats(inode, stat: NFSIOS_NORMALREADBYTES, addend: result);
182 }
183 nfs_end_io_read(inode);
184 return result;
185}
186EXPORT_SYMBOL_GPL(nfs_file_read);
187
188ssize_t
189nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
190 size_t len, unsigned int flags)
191{
192 struct inode *inode = file_inode(f: in);
193 ssize_t result;
194
195 dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
196
197 result = nfs_start_io_read(inode);
198 if (result)
199 return result;
200
201 result = nfs_revalidate_mapping(inode, mapping: in->f_mapping);
202 if (!result) {
203 result = filemap_splice_read(in, ppos, pipe, len, flags);
204 if (result > 0)
205 nfs_add_stats(inode, stat: NFSIOS_NORMALREADBYTES, addend: result);
206 }
207 nfs_end_io_read(inode);
208 return result;
209}
210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
211
212int
213nfs_file_mmap_prepare(struct vm_area_desc *desc)
214{
215 struct file *file = desc->file;
216 struct inode *inode = file_inode(f: file);
217 int status;
218
219 dprintk("NFS: mmap(%pD2)\n", file);
220
221 /* Note: generic_file_mmap_prepare() returns ENOSYS on nommu systems
222 * so we call that before revalidating the mapping
223 */
224 status = generic_file_mmap_prepare(desc);
225 if (!status) {
226 desc->vm_ops = &nfs_file_vm_ops;
227 status = nfs_revalidate_mapping(inode, mapping: file->f_mapping);
228 }
229 return status;
230}
231EXPORT_SYMBOL_GPL(nfs_file_mmap_prepare);
232
233/*
234 * Flush any dirty pages for this process, and check for write errors.
235 * The return status from this call provides a reliable indication of
236 * whether any write errors occurred for this process.
237 */
238static int
239nfs_file_fsync_commit(struct file *file, int datasync)
240{
241 struct inode *inode = file_inode(f: file);
242 int ret, ret2;
243
244 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
245
246 nfs_inc_stats(inode, stat: NFSIOS_VFSFSYNC);
247 ret = nfs_commit_inode(inode, FLUSH_SYNC);
248 ret2 = file_check_and_advance_wb_err(file);
249 if (ret2 < 0)
250 return ret2;
251 return ret;
252}
253
254int
255nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
256{
257 struct inode *inode = file_inode(f: file);
258 struct nfs_inode *nfsi = NFS_I(inode);
259 long save_nredirtied = atomic_long_read(v: &nfsi->redirtied_pages);
260 long nredirtied;
261 int ret;
262
263 trace_nfs_fsync_enter(inode);
264
265 for (;;) {
266 ret = file_write_and_wait_range(file, start, end);
267 if (ret != 0)
268 break;
269 ret = nfs_file_fsync_commit(file, datasync);
270 if (ret != 0)
271 break;
272 ret = pnfs_sync_inode(inode, datasync: !!datasync);
273 if (ret != 0)
274 break;
275 nredirtied = atomic_long_read(v: &nfsi->redirtied_pages);
276 if (nredirtied == save_nredirtied)
277 break;
278 save_nredirtied = nredirtied;
279 }
280
281 trace_nfs_fsync_exit(inode, error: ret);
282 return ret;
283}
284EXPORT_SYMBOL_GPL(nfs_file_fsync);
285
286void nfs_truncate_last_folio(struct address_space *mapping, loff_t from,
287 loff_t to)
288{
289 struct folio *folio;
290
291 if (from >= to)
292 return;
293
294 folio = filemap_lock_folio(mapping, index: from >> PAGE_SHIFT);
295 if (IS_ERR(ptr: folio))
296 return;
297
298 if (folio_mkclean(folio))
299 folio_mark_dirty(folio);
300
301 if (folio_test_uptodate(folio)) {
302 loff_t fpos = folio_pos(folio);
303 size_t offset = from - fpos;
304 size_t end = folio_size(folio);
305
306 if (to - fpos < end)
307 end = to - fpos;
308 folio_zero_segment(folio, start: offset, xend: end);
309 trace_nfs_size_truncate_folio(inode: mapping->host, new_size: to);
310 }
311
312 folio_unlock(folio);
313 folio_put(folio);
314}
315EXPORT_SYMBOL_GPL(nfs_truncate_last_folio);
316
317/*
318 * Decide whether a read/modify/write cycle may be more efficient
319 * then a modify/write/read cycle when writing to a page in the
320 * page cache.
321 *
322 * Some pNFS layout drivers can only read/write at a certain block
323 * granularity like all block devices and therefore we must perform
324 * read/modify/write whenever a page hasn't read yet and the data
325 * to be written there is not aligned to a block boundary and/or
326 * smaller than the block size.
327 *
328 * The modify/write/read cycle may occur if a page is read before
329 * being completely filled by the writer. In this situation, the
330 * page must be completely written to stable storage on the server
331 * before it can be refilled by reading in the page from the server.
332 * This can lead to expensive, small, FILE_SYNC mode writes being
333 * done.
334 *
335 * It may be more efficient to read the page first if the file is
336 * open for reading in addition to writing, the page is not marked
337 * as Uptodate, it is not dirty or waiting to be committed,
338 * indicating that it was previously allocated and then modified,
339 * that there were valid bytes of data in that range of the file,
340 * and that the new data won't completely replace the old data in
341 * that range of the file.
342 */
343static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
344 unsigned int len)
345{
346 unsigned int pglen = nfs_folio_length(folio);
347 unsigned int offset = offset_in_folio(folio, pos);
348 unsigned int end = offset + len;
349
350 return !pglen || (end >= pglen && !offset);
351}
352
353static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
354 loff_t pos, unsigned int len)
355{
356 /*
357 * Up-to-date pages, those with ongoing or full-page write
358 * don't need read/modify/write
359 */
360 if (folio_test_uptodate(folio) || folio_test_private(folio) ||
361 nfs_folio_is_full_write(folio, pos, len))
362 return false;
363
364 if (pnfs_ld_read_whole_page(inode: file_inode(f: file)))
365 return true;
366 if (folio_test_dropbehind(folio))
367 return false;
368 /* Open for reading too? */
369 if (file->f_mode & FMODE_READ)
370 return true;
371 return false;
372}
373
374/*
375 * This does the "real" work of the write. We must allocate and lock the
376 * page to be sent back to the generic routine, which then copies the
377 * data from user space.
378 *
379 * If the writer ends up delaying the write, the writer needs to
380 * increment the page use counts until he is done with the page.
381 */
382static int nfs_write_begin(const struct kiocb *iocb,
383 struct address_space *mapping,
384 loff_t pos, unsigned len, struct folio **foliop,
385 void **fsdata)
386{
387 struct folio *folio;
388 struct file *file = iocb->ki_filp;
389 int once_thru = 0;
390 int ret;
391
392 trace_nfs_write_begin(inode: file_inode(f: file), offset: pos, count: len);
393
394 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
395 file, mapping->host->i_ino, len, (long long) pos);
396 nfs_truncate_last_folio(mapping, i_size_read(inode: mapping->host), pos);
397
398start:
399 folio = write_begin_get_folio(iocb, mapping, index: pos >> PAGE_SHIFT, len);
400 if (IS_ERR(ptr: folio)) {
401 ret = PTR_ERR(ptr: folio);
402 goto out;
403 }
404 *foliop = folio;
405
406 ret = nfs_flush_incompatible(file, folio);
407 if (ret) {
408 folio_unlock(folio);
409 folio_put(folio);
410 } else if (!once_thru &&
411 nfs_want_read_modify_write(file, folio, pos, len)) {
412 once_thru = 1;
413 folio_clear_dropbehind(folio);
414 ret = nfs_read_folio(file, folio);
415 folio_put(folio);
416 if (!ret)
417 goto start;
418 }
419out:
420 trace_nfs_write_begin_done(inode: file_inode(f: file), offset: pos, count: len, ret);
421 return ret;
422}
423
424static int nfs_write_end(const struct kiocb *iocb,
425 struct address_space *mapping,
426 loff_t pos, unsigned len, unsigned copied,
427 struct folio *folio, void *fsdata)
428{
429 struct file *file = iocb->ki_filp;
430 struct nfs_open_context *ctx = nfs_file_open_context(filp: file);
431 unsigned offset = offset_in_folio(folio, pos);
432 int status;
433
434 trace_nfs_write_end(inode: file_inode(f: file), offset: pos, count: len);
435 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
436 file, mapping->host->i_ino, len, (long long) pos);
437
438 /*
439 * Zero any uninitialised parts of the page, and then mark the page
440 * as up to date if it turns out that we're extending the file.
441 */
442 if (!folio_test_uptodate(folio)) {
443 size_t fsize = folio_size(folio);
444 unsigned pglen = nfs_folio_length(folio);
445 unsigned end = offset + copied;
446
447 if (pglen == 0) {
448 folio_zero_segments(folio, start1: 0, xend1: offset, start2: end, xend2: fsize);
449 folio_mark_uptodate(folio);
450 } else if (end >= pglen) {
451 folio_zero_segment(folio, start: end, xend: fsize);
452 if (offset == 0)
453 folio_mark_uptodate(folio);
454 } else
455 folio_zero_segment(folio, start: pglen, xend: fsize);
456 }
457
458 status = nfs_update_folio(file, folio, offset, count: copied);
459
460 folio_unlock(folio);
461 folio_put(folio);
462
463 if (status < 0) {
464 trace_nfs_write_end_done(inode: file_inode(f: file), offset: pos, count: len, ret: status);
465 return status;
466 }
467 NFS_I(inode: mapping->host)->write_io += copied;
468
469 if (nfs_ctx_key_to_expire(ctx, inode: mapping->host))
470 nfs_wb_all(inode: mapping->host);
471
472 trace_nfs_write_end_done(inode: file_inode(f: file), offset: pos, count: len, ret: copied);
473 return copied;
474}
475
476/*
477 * Partially or wholly invalidate a page
478 * - Release the private state associated with a page if undergoing complete
479 * page invalidation
480 * - Called if either PG_private or PG_fscache is set on the page
481 * - Caller holds page lock
482 */
483static void nfs_invalidate_folio(struct folio *folio, size_t offset,
484 size_t length)
485{
486 struct inode *inode = folio->mapping->host;
487 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
488 folio->index, offset, length);
489
490 /* Cancel any unstarted writes on this page */
491 if (offset != 0 || length < folio_size(folio))
492 nfs_wb_folio(inode, folio);
493 else
494 nfs_wb_folio_cancel(inode, folio);
495 folio_wait_private_2(folio); /* [DEPRECATED] */
496 trace_nfs_invalidate_folio(inode, offset: folio_pos(folio) + offset, count: length);
497}
498
499/*
500 * Attempt to release the private state associated with a folio
501 * - Called if either private or fscache flags are set on the folio
502 * - Caller holds folio lock
503 * - Return true (may release folio) or false (may not)
504 */
505static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
506{
507 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
508
509 /* If the private flag is set, then the folio is not freeable */
510 if (folio_test_private(folio)) {
511 if ((current_gfp_context(flags: gfp) & GFP_KERNEL) != GFP_KERNEL ||
512 current_is_kswapd() || current_is_kcompactd())
513 return false;
514 if (nfs_wb_folio(inode: folio->mapping->host, folio) < 0)
515 return false;
516 }
517 return nfs_fscache_release_folio(folio, gfp);
518}
519
520static void nfs_check_dirty_writeback(struct folio *folio,
521 bool *dirty, bool *writeback)
522{
523 struct nfs_inode *nfsi;
524 struct address_space *mapping = folio->mapping;
525
526 /*
527 * Check if an unstable folio is currently being committed and
528 * if so, have the VM treat it as if the folio is under writeback
529 * so it will not block due to folios that will shortly be freeable.
530 */
531 nfsi = NFS_I(inode: mapping->host);
532 if (atomic_read(v: &nfsi->commit_info.rpcs_out)) {
533 *writeback = true;
534 return;
535 }
536
537 /*
538 * If the private flag is set, then the folio is not freeable
539 * and as the inode is not being committed, it's not going to
540 * be cleaned in the near future so treat it as dirty
541 */
542 if (folio_test_private(folio))
543 *dirty = true;
544}
545
546/*
547 * Attempt to clear the private state associated with a page when an error
548 * occurs that requires the cached contents of an inode to be written back or
549 * destroyed
550 * - Called if either PG_private or fscache is set on the page
551 * - Caller holds page lock
552 * - Return 0 if successful, -error otherwise
553 */
554static int nfs_launder_folio(struct folio *folio)
555{
556 struct inode *inode = folio->mapping->host;
557 int ret;
558
559 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
560 inode->i_ino, folio_pos(folio));
561
562 folio_wait_private_2(folio); /* [DEPRECATED] */
563 ret = nfs_wb_folio(inode, folio);
564 trace_nfs_launder_folio_done(inode, offset: folio_pos(folio),
565 count: folio_size(folio), ret);
566 return ret;
567}
568
569static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
570 sector_t *span)
571{
572 unsigned long blocks;
573 long long isize;
574 int ret;
575 struct inode *inode = file_inode(f: file);
576 struct rpc_clnt *clnt = NFS_CLIENT(inode);
577 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
578
579 spin_lock(lock: &inode->i_lock);
580 blocks = inode->i_blocks;
581 isize = inode->i_size;
582 spin_unlock(lock: &inode->i_lock);
583 if (blocks*512 < isize) {
584 pr_warn("swap activate: swapfile has holes\n");
585 return -EINVAL;
586 }
587
588 ret = rpc_clnt_swap_activate(clnt);
589 if (ret)
590 return ret;
591 ret = add_swap_extent(sis, start_page: 0, nr_pages: sis->max, start_block: 0);
592 if (ret < 0) {
593 rpc_clnt_swap_deactivate(clnt);
594 return ret;
595 }
596
597 *span = sis->pages;
598
599 if (cl->rpc_ops->enable_swap)
600 cl->rpc_ops->enable_swap(inode);
601
602 sis->flags |= SWP_FS_OPS;
603 return ret;
604}
605
606static void nfs_swap_deactivate(struct file *file)
607{
608 struct inode *inode = file_inode(f: file);
609 struct rpc_clnt *clnt = NFS_CLIENT(inode);
610 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
611
612 rpc_clnt_swap_deactivate(clnt);
613 if (cl->rpc_ops->disable_swap)
614 cl->rpc_ops->disable_swap(file_inode(f: file));
615}
616
617const struct address_space_operations nfs_file_aops = {
618 .read_folio = nfs_read_folio,
619 .readahead = nfs_readahead,
620 .dirty_folio = filemap_dirty_folio,
621 .writepages = nfs_writepages,
622 .write_begin = nfs_write_begin,
623 .write_end = nfs_write_end,
624 .invalidate_folio = nfs_invalidate_folio,
625 .release_folio = nfs_release_folio,
626 .migrate_folio = nfs_migrate_folio,
627 .launder_folio = nfs_launder_folio,
628 .is_dirty_writeback = nfs_check_dirty_writeback,
629 .error_remove_folio = generic_error_remove_folio,
630 .swap_activate = nfs_swap_activate,
631 .swap_deactivate = nfs_swap_deactivate,
632 .swap_rw = nfs_swap_rw,
633};
634
635/*
636 * Notification that a PTE pointing to an NFS page is about to be made
637 * writable, implying that someone is about to modify the page through a
638 * shared-writable mapping
639 */
640static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
641{
642 struct file *filp = vmf->vma->vm_file;
643 struct inode *inode = file_inode(f: filp);
644 unsigned pagelen;
645 vm_fault_t ret = VM_FAULT_NOPAGE;
646 struct address_space *mapping;
647 struct folio *folio = page_folio(vmf->page);
648
649 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
650 filp, filp->f_mapping->host->i_ino,
651 (long long)folio_pos(folio));
652
653 sb_start_pagefault(sb: inode->i_sb);
654
655 /* make sure the cache has finished storing the page */
656 if (folio_test_private_2(folio) && /* [DEPRECATED] */
657 folio_wait_private_2_killable(folio) < 0) {
658 ret = VM_FAULT_RETRY;
659 goto out;
660 }
661
662 wait_on_bit_action(word: &NFS_I(inode)->flags, NFS_INO_INVALIDATING,
663 action: nfs_wait_bit_killable,
664 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
665
666 folio_lock(folio);
667 mapping = folio->mapping;
668 if (mapping != inode->i_mapping)
669 goto out_unlock;
670
671 folio_wait_writeback(folio);
672
673 pagelen = nfs_folio_length(folio);
674 if (pagelen == 0)
675 goto out_unlock;
676
677 ret = VM_FAULT_LOCKED;
678 if (nfs_flush_incompatible(file: filp, folio) == 0 &&
679 nfs_update_folio(file: filp, folio, offset: 0, count: pagelen) == 0)
680 goto out;
681
682 ret = VM_FAULT_SIGBUS;
683out_unlock:
684 folio_unlock(folio);
685out:
686 sb_end_pagefault(sb: inode->i_sb);
687 return ret;
688}
689
690static const struct vm_operations_struct nfs_file_vm_ops = {
691 .fault = filemap_fault,
692 .map_pages = filemap_map_pages,
693 .page_mkwrite = nfs_vm_page_mkwrite,
694};
695
696ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
697{
698 struct file *file = iocb->ki_filp;
699 struct inode *inode = file_inode(f: file);
700 unsigned int mntflags = NFS_SERVER(inode)->flags;
701 ssize_t result, written;
702 errseq_t since;
703 int error;
704
705 trace_nfs_file_write(iocb, iter: from);
706
707 result = nfs_key_timeout_notify(filp: file, inode);
708 if (result)
709 return result;
710
711 if (iocb->ki_flags & IOCB_DIRECT)
712 return nfs_file_direct_write(iocb, iter: from, swap: false);
713
714 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
715 file, iov_iter_count(from), (long long) iocb->ki_pos);
716
717 if (IS_SWAPFILE(inode))
718 goto out_swapfile;
719 /*
720 * O_APPEND implies that we must revalidate the file length.
721 */
722 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
723 result = nfs_revalidate_file_size(inode, filp: file);
724 if (result)
725 return result;
726 }
727
728 nfs_clear_invalid_mapping(mapping: file->f_mapping);
729
730 since = filemap_sample_wb_err(mapping: file->f_mapping);
731 error = nfs_start_io_write(inode);
732 if (error)
733 return error;
734 result = generic_write_checks(iocb, from);
735 if (result > 0)
736 result = generic_perform_write(iocb, from);
737 nfs_end_io_write(inode);
738 if (result <= 0)
739 goto out;
740
741 written = result;
742 nfs_add_stats(inode, stat: NFSIOS_NORMALWRITTENBYTES, addend: written);
743
744 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
745 result = filemap_fdatawrite_range(mapping: file->f_mapping,
746 start: iocb->ki_pos - written,
747 end: iocb->ki_pos - 1);
748 if (result < 0)
749 goto out;
750 }
751 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
752 filemap_fdatawait_range(file->f_mapping,
753 lstart: iocb->ki_pos - written,
754 lend: iocb->ki_pos - 1);
755 }
756 result = generic_write_sync(iocb, count: written);
757 if (result < 0)
758 return result;
759
760out:
761 /* Return error values */
762 error = filemap_check_wb_err(mapping: file->f_mapping, since);
763 switch (error) {
764 default:
765 break;
766 case -EDQUOT:
767 case -EFBIG:
768 case -ENOSPC:
769 nfs_wb_all(inode);
770 error = file_check_and_advance_wb_err(file);
771 if (error < 0)
772 result = error;
773 }
774 return result;
775
776out_swapfile:
777 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
778 return -ETXTBSY;
779}
780EXPORT_SYMBOL_GPL(nfs_file_write);
781
782static int
783do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
784{
785 struct inode *inode = filp->f_mapping->host;
786 int status = 0;
787 unsigned int saved_type = fl->c.flc_type;
788
789 /* Try local locking first */
790 posix_test_lock(filp, fl);
791 if (fl->c.flc_type != F_UNLCK) {
792 /* found a conflict */
793 goto out;
794 }
795 fl->c.flc_type = saved_type;
796
797 if (nfs_have_read_or_write_delegation(inode))
798 goto out_noconflict;
799
800 if (is_local)
801 goto out_noconflict;
802
803 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
804out:
805 return status;
806out_noconflict:
807 fl->c.flc_type = F_UNLCK;
808 goto out;
809}
810
811static int
812do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
813{
814 struct inode *inode = filp->f_mapping->host;
815 struct nfs_lock_context *l_ctx;
816 int status;
817
818 /*
819 * Flush all pending writes before doing anything
820 * with locks..
821 */
822 nfs_wb_all(inode);
823
824 l_ctx = nfs_get_lock_context(ctx: nfs_file_open_context(filp));
825 if (!IS_ERR(ptr: l_ctx)) {
826 status = nfs_iocounter_wait(l_ctx);
827 nfs_put_lock_context(l_ctx);
828 /* NOTE: special case
829 * If we're signalled while cleaning up locks on process exit, we
830 * still need to complete the unlock.
831 */
832 if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
833 return status;
834 }
835
836 /*
837 * Use local locking if mounted with "-onolock" or with appropriate
838 * "-olocal_lock="
839 */
840 if (!is_local)
841 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
842 else
843 status = locks_lock_file_wait(filp, fl);
844 return status;
845}
846
847static int
848do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
849{
850 struct inode *inode = filp->f_mapping->host;
851 int status;
852
853 /*
854 * Flush all pending writes before doing anything
855 * with locks..
856 */
857 status = nfs_sync_mapping(mapping: filp->f_mapping);
858 if (status != 0)
859 goto out;
860
861 /*
862 * Use local locking if mounted with "-onolock" or with appropriate
863 * "-olocal_lock="
864 */
865 if (!is_local)
866 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
867 else
868 status = locks_lock_file_wait(filp, fl);
869 if (status < 0)
870 goto out;
871
872 /*
873 * Invalidate cache to prevent missing any changes. If
874 * the file is mapped, clear the page cache as well so
875 * those mappings will be loaded.
876 *
877 * This makes locking act as a cache coherency point.
878 */
879 nfs_sync_mapping(mapping: filp->f_mapping);
880 if (!nfs_have_read_or_write_delegation(inode)) {
881 nfs_zap_caches(inode);
882 if (mapping_mapped(mapping: filp->f_mapping))
883 nfs_revalidate_mapping(inode, mapping: filp->f_mapping);
884 }
885out:
886 return status;
887}
888
889/*
890 * Lock a (portion of) a file
891 */
892int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
893{
894 struct inode *inode = filp->f_mapping->host;
895 int ret = -ENOLCK;
896 int is_local = 0;
897
898 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
899 filp, fl->c.flc_type, fl->c.flc_flags,
900 (long long)fl->fl_start, (long long)fl->fl_end);
901
902 nfs_inc_stats(inode, stat: NFSIOS_VFSLOCK);
903
904 if (fl->c.flc_flags & FL_RECLAIM)
905 return -ENOGRACE;
906
907 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
908 is_local = 1;
909
910 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
911 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
912 if (ret < 0)
913 goto out_err;
914 }
915
916 if (IS_GETLK(cmd))
917 ret = do_getlk(filp, cmd, fl, is_local);
918 else if (lock_is_unlock(fl))
919 ret = do_unlk(filp, cmd, fl, is_local);
920 else
921 ret = do_setlk(filp, cmd, fl, is_local);
922out_err:
923 return ret;
924}
925EXPORT_SYMBOL_GPL(nfs_lock);
926
927/*
928 * Lock a (portion of) a file
929 */
930int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
931{
932 struct inode *inode = filp->f_mapping->host;
933 int is_local = 0;
934
935 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
936 filp, fl->c.flc_type, fl->c.flc_flags);
937
938 if (!(fl->c.flc_flags & FL_FLOCK))
939 return -ENOLCK;
940
941 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
942 is_local = 1;
943
944 /* We're simulating flock() locks using posix locks on the server */
945 if (lock_is_unlock(fl))
946 return do_unlk(filp, cmd, fl, is_local);
947 return do_setlk(filp, cmd, fl, is_local);
948}
949EXPORT_SYMBOL_GPL(nfs_flock);
950
951const struct file_operations nfs_file_operations = {
952 .llseek = nfs_file_llseek,
953 .read_iter = nfs_file_read,
954 .write_iter = nfs_file_write,
955 .mmap_prepare = nfs_file_mmap_prepare,
956 .open = nfs_file_open,
957 .flush = nfs_file_flush,
958 .release = nfs_file_release,
959 .fsync = nfs_file_fsync,
960 .lock = nfs_lock,
961 .flock = nfs_flock,
962 .splice_read = nfs_file_splice_read,
963 .splice_write = iter_file_splice_write,
964 .check_flags = nfs_check_flags,
965 .setlease = simple_nosetlease,
966 .fop_flags = FOP_DONTCACHE,
967};
968EXPORT_SYMBOL_GPL(nfs_file_operations);
969