Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * Changes Copyright (C) 1994 by Florian La Roche
8 * - Do not copy data too often around in the kernel.
9 * - In nfs_file_read the return value of kmalloc wasn't checked.
10 * - Put in a better version of read look-ahead buffering. Original idea
11 * and implementation by Wai S Kok elekokws@ee.nus.sg.
12 *
13 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 *
15 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 *
17 * nfs regular file handling functions
18 */
19
20#include <linux/module.h>
21#include <linux/time.h>
22#include <linux/kernel.h>
23#include <linux/errno.h>
24#include <linux/fcntl.h>
25#include <linux/stat.h>
26#include <linux/nfs_fs.h>
27#include <linux/nfs_mount.h>
28#include <linux/mm.h>
29#include <linux/pagemap.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32
33#include <linux/uaccess.h>
34
35#include "delegation.h"
36#include "internal.h"
37#include "iostat.h"
38#include "fscache.h"
39#include "pnfs.h"
40
41#include "nfstrace.h"
42
43#define NFSDBG_FACILITY NFSDBG_FILE
44
45static const struct vm_operations_struct nfs_file_vm_ops;
46
47/* Hack for future NFS swap support */
48#ifndef IS_SWAPFILE
49# define IS_SWAPFILE(inode) (0)
50#endif
51
52int nfs_check_flags(int flags)
53{
54 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
55 return -EINVAL;
56
57 return 0;
58}
59EXPORT_SYMBOL_GPL(nfs_check_flags);
60
61/*
62 * Open file
63 */
64static int
65nfs_file_open(struct inode *inode, struct file *filp)
66{
67 int res;
68
69 dprintk("NFS: open file(%pD2)\n", filp);
70
71 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
72 res = nfs_check_flags(filp->f_flags);
73 if (res)
74 return res;
75
76 res = nfs_open(inode, filp);
77 return res;
78}
79
80int
81nfs_file_release(struct inode *inode, struct file *filp)
82{
83 dprintk("NFS: release(%pD2)\n", filp);
84
85 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
86 nfs_file_clear_open_context(filp);
87 return 0;
88}
89EXPORT_SYMBOL_GPL(nfs_file_release);
90
91/**
92 * nfs_revalidate_size - Revalidate the file size
93 * @inode: pointer to inode struct
94 * @filp: pointer to struct file
95 *
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
101 */
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_REVAL_PAGECACHE))
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
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 return nfs_wb_all(inode);
152}
153
154ssize_t
155nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
156{
157 struct inode *inode = file_inode(iocb->ki_filp);
158 ssize_t result;
159
160 if (iocb->ki_flags & IOCB_DIRECT)
161 return nfs_file_direct_read(iocb, to);
162
163 dprintk("NFS: read(%pD2, %zu@%lu)\n",
164 iocb->ki_filp,
165 iov_iter_count(to), (unsigned long) iocb->ki_pos);
166
167 nfs_start_io_read(inode);
168 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
169 if (!result) {
170 result = generic_file_read_iter(iocb, to);
171 if (result > 0)
172 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
173 }
174 nfs_end_io_read(inode);
175 return result;
176}
177EXPORT_SYMBOL_GPL(nfs_file_read);
178
179int
180nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
181{
182 struct inode *inode = file_inode(file);
183 int status;
184
185 dprintk("NFS: mmap(%pD2)\n", file);
186
187 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
188 * so we call that before revalidating the mapping
189 */
190 status = generic_file_mmap(file, vma);
191 if (!status) {
192 vma->vm_ops = &nfs_file_vm_ops;
193 status = nfs_revalidate_mapping(inode, file->f_mapping);
194 }
195 return status;
196}
197EXPORT_SYMBOL_GPL(nfs_file_mmap);
198
199/*
200 * Flush any dirty pages for this process, and check for write errors.
201 * The return status from this call provides a reliable indication of
202 * whether any write errors occurred for this process.
203 */
204static int
205nfs_file_fsync_commit(struct file *file, int datasync)
206{
207 struct nfs_open_context *ctx = nfs_file_open_context(file);
208 struct inode *inode = file_inode(file);
209 int do_resend, status;
210 int ret = 0;
211
212 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
213
214 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
215 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
216 status = nfs_commit_inode(inode, FLUSH_SYNC);
217 if (status == 0)
218 status = file_check_and_advance_wb_err(file);
219 if (status < 0) {
220 ret = status;
221 goto out;
222 }
223 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
224 if (do_resend)
225 ret = -EAGAIN;
226out:
227 return ret;
228}
229
230int
231nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
232{
233 int ret;
234 struct inode *inode = file_inode(file);
235
236 trace_nfs_fsync_enter(inode);
237
238 do {
239 ret = file_write_and_wait_range(file, start, end);
240 if (ret != 0)
241 break;
242 ret = nfs_file_fsync_commit(file, datasync);
243 if (!ret)
244 ret = pnfs_sync_inode(inode, !!datasync);
245 /*
246 * If nfs_file_fsync_commit detected a server reboot, then
247 * resend all dirty pages that might have been covered by
248 * the NFS_CONTEXT_RESEND_WRITES flag
249 */
250 start = 0;
251 end = LLONG_MAX;
252 } while (ret == -EAGAIN);
253
254 trace_nfs_fsync_exit(inode, ret);
255 return ret;
256}
257EXPORT_SYMBOL_GPL(nfs_file_fsync);
258
259/*
260 * Decide whether a read/modify/write cycle may be more efficient
261 * then a modify/write/read cycle when writing to a page in the
262 * page cache.
263 *
264 * Some pNFS layout drivers can only read/write at a certain block
265 * granularity like all block devices and therefore we must perform
266 * read/modify/write whenever a page hasn't read yet and the data
267 * to be written there is not aligned to a block boundary and/or
268 * smaller than the block size.
269 *
270 * The modify/write/read cycle may occur if a page is read before
271 * being completely filled by the writer. In this situation, the
272 * page must be completely written to stable storage on the server
273 * before it can be refilled by reading in the page from the server.
274 * This can lead to expensive, small, FILE_SYNC mode writes being
275 * done.
276 *
277 * It may be more efficient to read the page first if the file is
278 * open for reading in addition to writing, the page is not marked
279 * as Uptodate, it is not dirty or waiting to be committed,
280 * indicating that it was previously allocated and then modified,
281 * that there were valid bytes of data in that range of the file,
282 * and that the new data won't completely replace the old data in
283 * that range of the file.
284 */
285static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
286{
287 unsigned int pglen = nfs_page_length(page);
288 unsigned int offset = pos & (PAGE_SIZE - 1);
289 unsigned int end = offset + len;
290
291 return !pglen || (end >= pglen && !offset);
292}
293
294static bool nfs_want_read_modify_write(struct file *file, struct page *page,
295 loff_t pos, unsigned int len)
296{
297 /*
298 * Up-to-date pages, those with ongoing or full-page write
299 * don't need read/modify/write
300 */
301 if (PageUptodate(page) || PagePrivate(page) ||
302 nfs_full_page_write(page, pos, len))
303 return false;
304
305 if (pnfs_ld_read_whole_page(file->f_mapping->host))
306 return true;
307 /* Open for reading too? */
308 if (file->f_mode & FMODE_READ)
309 return true;
310 return false;
311}
312
313/*
314 * This does the "real" work of the write. We must allocate and lock the
315 * page to be sent back to the generic routine, which then copies the
316 * data from user space.
317 *
318 * If the writer ends up delaying the write, the writer needs to
319 * increment the page use counts until he is done with the page.
320 */
321static int nfs_write_begin(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned flags,
323 struct page **pagep, void **fsdata)
324{
325 int ret;
326 pgoff_t index = pos >> PAGE_SHIFT;
327 struct page *page;
328 int once_thru = 0;
329
330 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
331 file, mapping->host->i_ino, len, (long long) pos);
332
333start:
334 page = grab_cache_page_write_begin(mapping, index, flags);
335 if (!page)
336 return -ENOMEM;
337 *pagep = page;
338
339 ret = nfs_flush_incompatible(file, page);
340 if (ret) {
341 unlock_page(page);
342 put_page(page);
343 } else if (!once_thru &&
344 nfs_want_read_modify_write(file, page, pos, len)) {
345 once_thru = 1;
346 ret = nfs_readpage(file, page);
347 put_page(page);
348 if (!ret)
349 goto start;
350 }
351 return ret;
352}
353
354static int nfs_write_end(struct file *file, struct address_space *mapping,
355 loff_t pos, unsigned len, unsigned copied,
356 struct page *page, void *fsdata)
357{
358 unsigned offset = pos & (PAGE_SIZE - 1);
359 struct nfs_open_context *ctx = nfs_file_open_context(file);
360 int status;
361
362 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
363 file, mapping->host->i_ino, len, (long long) pos);
364
365 /*
366 * Zero any uninitialised parts of the page, and then mark the page
367 * as up to date if it turns out that we're extending the file.
368 */
369 if (!PageUptodate(page)) {
370 unsigned pglen = nfs_page_length(page);
371 unsigned end = offset + copied;
372
373 if (pglen == 0) {
374 zero_user_segments(page, 0, offset,
375 end, PAGE_SIZE);
376 SetPageUptodate(page);
377 } else if (end >= pglen) {
378 zero_user_segment(page, end, PAGE_SIZE);
379 if (offset == 0)
380 SetPageUptodate(page);
381 } else
382 zero_user_segment(page, pglen, PAGE_SIZE);
383 }
384
385 status = nfs_updatepage(file, page, offset, copied);
386
387 unlock_page(page);
388 put_page(page);
389
390 if (status < 0)
391 return status;
392 NFS_I(mapping->host)->write_io += copied;
393
394 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
395 status = nfs_wb_all(mapping->host);
396 if (status < 0)
397 return status;
398 }
399
400 return copied;
401}
402
403/*
404 * Partially or wholly invalidate a page
405 * - Release the private state associated with a page if undergoing complete
406 * page invalidation
407 * - Called if either PG_private or PG_fscache is set on the page
408 * - Caller holds page lock
409 */
410static void nfs_invalidate_page(struct page *page, unsigned int offset,
411 unsigned int length)
412{
413 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
414 page, offset, length);
415
416 if (offset != 0 || length < PAGE_SIZE)
417 return;
418 /* Cancel any unstarted writes on this page */
419 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
420
421 nfs_fscache_invalidate_page(page, page->mapping->host);
422}
423
424/*
425 * Attempt to release the private state associated with a page
426 * - Called if either PG_private or PG_fscache is set on the page
427 * - Caller holds page lock
428 * - Return true (may release page) or false (may not)
429 */
430static int nfs_release_page(struct page *page, gfp_t gfp)
431{
432 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
433
434 /* If PagePrivate() is set, then the page is not freeable */
435 if (PagePrivate(page))
436 return 0;
437 return nfs_fscache_release_page(page, gfp);
438}
439
440static void nfs_check_dirty_writeback(struct page *page,
441 bool *dirty, bool *writeback)
442{
443 struct nfs_inode *nfsi;
444 struct address_space *mapping = page_file_mapping(page);
445
446 if (!mapping || PageSwapCache(page))
447 return;
448
449 /*
450 * Check if an unstable page is currently being committed and
451 * if so, have the VM treat it as if the page is under writeback
452 * so it will not block due to pages that will shortly be freeable.
453 */
454 nfsi = NFS_I(mapping->host);
455 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
456 *writeback = true;
457 return;
458 }
459
460 /*
461 * If PagePrivate() is set, then the page is not freeable and as the
462 * inode is not being committed, it's not going to be cleaned in the
463 * near future so treat it as dirty
464 */
465 if (PagePrivate(page))
466 *dirty = true;
467}
468
469/*
470 * Attempt to clear the private state associated with a page when an error
471 * occurs that requires the cached contents of an inode to be written back or
472 * destroyed
473 * - Called if either PG_private or fscache is set on the page
474 * - Caller holds page lock
475 * - Return 0 if successful, -error otherwise
476 */
477static int nfs_launder_page(struct page *page)
478{
479 struct inode *inode = page_file_mapping(page)->host;
480 struct nfs_inode *nfsi = NFS_I(inode);
481
482 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
483 inode->i_ino, (long long)page_offset(page));
484
485 nfs_fscache_wait_on_page_write(nfsi, page);
486 return nfs_wb_page(inode, page);
487}
488
489static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
490 sector_t *span)
491{
492 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
493
494 *span = sis->pages;
495
496 return rpc_clnt_swap_activate(clnt);
497}
498
499static void nfs_swap_deactivate(struct file *file)
500{
501 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
502
503 rpc_clnt_swap_deactivate(clnt);
504}
505
506const struct address_space_operations nfs_file_aops = {
507 .readpage = nfs_readpage,
508 .readpages = nfs_readpages,
509 .set_page_dirty = __set_page_dirty_nobuffers,
510 .writepage = nfs_writepage,
511 .writepages = nfs_writepages,
512 .write_begin = nfs_write_begin,
513 .write_end = nfs_write_end,
514 .invalidatepage = nfs_invalidate_page,
515 .releasepage = nfs_release_page,
516 .direct_IO = nfs_direct_IO,
517#ifdef CONFIG_MIGRATION
518 .migratepage = nfs_migrate_page,
519#endif
520 .launder_page = nfs_launder_page,
521 .is_dirty_writeback = nfs_check_dirty_writeback,
522 .error_remove_page = generic_error_remove_page,
523 .swap_activate = nfs_swap_activate,
524 .swap_deactivate = nfs_swap_deactivate,
525};
526
527/*
528 * Notification that a PTE pointing to an NFS page is about to be made
529 * writable, implying that someone is about to modify the page through a
530 * shared-writable mapping
531 */
532static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
533{
534 struct page *page = vmf->page;
535 struct file *filp = vmf->vma->vm_file;
536 struct inode *inode = file_inode(filp);
537 unsigned pagelen;
538 vm_fault_t ret = VM_FAULT_NOPAGE;
539 struct address_space *mapping;
540
541 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
542 filp, filp->f_mapping->host->i_ino,
543 (long long)page_offset(page));
544
545 sb_start_pagefault(inode->i_sb);
546
547 /* make sure the cache has finished storing the page */
548 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
549
550 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
551 nfs_wait_bit_killable, TASK_KILLABLE);
552
553 lock_page(page);
554 mapping = page_file_mapping(page);
555 if (mapping != inode->i_mapping)
556 goto out_unlock;
557
558 wait_on_page_writeback(page);
559
560 pagelen = nfs_page_length(page);
561 if (pagelen == 0)
562 goto out_unlock;
563
564 ret = VM_FAULT_LOCKED;
565 if (nfs_flush_incompatible(filp, page) == 0 &&
566 nfs_updatepage(filp, page, 0, pagelen) == 0)
567 goto out;
568
569 ret = VM_FAULT_SIGBUS;
570out_unlock:
571 unlock_page(page);
572out:
573 sb_end_pagefault(inode->i_sb);
574 return ret;
575}
576
577static const struct vm_operations_struct nfs_file_vm_ops = {
578 .fault = filemap_fault,
579 .map_pages = filemap_map_pages,
580 .page_mkwrite = nfs_vm_page_mkwrite,
581};
582
583static int nfs_need_check_write(struct file *filp, struct inode *inode)
584{
585 struct nfs_open_context *ctx;
586
587 ctx = nfs_file_open_context(filp);
588 if (nfs_ctx_key_to_expire(ctx, inode))
589 return 1;
590 return 0;
591}
592
593ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
594{
595 struct file *file = iocb->ki_filp;
596 struct inode *inode = file_inode(file);
597 unsigned long written = 0;
598 ssize_t result;
599
600 result = nfs_key_timeout_notify(file, inode);
601 if (result)
602 return result;
603
604 if (iocb->ki_flags & IOCB_DIRECT)
605 return nfs_file_direct_write(iocb, from);
606
607 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
608 file, iov_iter_count(from), (long long) iocb->ki_pos);
609
610 if (IS_SWAPFILE(inode))
611 goto out_swapfile;
612 /*
613 * O_APPEND implies that we must revalidate the file length.
614 */
615 if (iocb->ki_flags & IOCB_APPEND) {
616 result = nfs_revalidate_file_size(inode, file);
617 if (result)
618 goto out;
619 }
620 if (iocb->ki_pos > i_size_read(inode))
621 nfs_revalidate_mapping(inode, file->f_mapping);
622
623 nfs_start_io_write(inode);
624 result = generic_write_checks(iocb, from);
625 if (result > 0) {
626 current->backing_dev_info = inode_to_bdi(inode);
627 result = generic_perform_write(file, from, iocb->ki_pos);
628 current->backing_dev_info = NULL;
629 }
630 nfs_end_io_write(inode);
631 if (result <= 0)
632 goto out;
633
634 written = result;
635 iocb->ki_pos += written;
636 result = generic_write_sync(iocb, written);
637 if (result < 0)
638 goto out;
639
640 /* Return error values */
641 if (nfs_need_check_write(file, inode)) {
642 int err = nfs_wb_all(inode);
643 if (err < 0)
644 result = err;
645 }
646 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
647out:
648 return result;
649
650out_swapfile:
651 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
652 return -EBUSY;
653}
654EXPORT_SYMBOL_GPL(nfs_file_write);
655
656static int
657do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
658{
659 struct inode *inode = filp->f_mapping->host;
660 int status = 0;
661 unsigned int saved_type = fl->fl_type;
662
663 /* Try local locking first */
664 posix_test_lock(filp, fl);
665 if (fl->fl_type != F_UNLCK) {
666 /* found a conflict */
667 goto out;
668 }
669 fl->fl_type = saved_type;
670
671 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
672 goto out_noconflict;
673
674 if (is_local)
675 goto out_noconflict;
676
677 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
678out:
679 return status;
680out_noconflict:
681 fl->fl_type = F_UNLCK;
682 goto out;
683}
684
685static int
686do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
687{
688 struct inode *inode = filp->f_mapping->host;
689 struct nfs_lock_context *l_ctx;
690 int status;
691
692 /*
693 * Flush all pending writes before doing anything
694 * with locks..
695 */
696 nfs_wb_all(inode);
697
698 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
699 if (!IS_ERR(l_ctx)) {
700 status = nfs_iocounter_wait(l_ctx);
701 nfs_put_lock_context(l_ctx);
702 /* NOTE: special case
703 * If we're signalled while cleaning up locks on process exit, we
704 * still need to complete the unlock.
705 */
706 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
707 return status;
708 }
709
710 /*
711 * Use local locking if mounted with "-onolock" or with appropriate
712 * "-olocal_lock="
713 */
714 if (!is_local)
715 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
716 else
717 status = locks_lock_file_wait(filp, fl);
718 return status;
719}
720
721static int
722do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
723{
724 struct inode *inode = filp->f_mapping->host;
725 int status;
726
727 /*
728 * Flush all pending writes before doing anything
729 * with locks..
730 */
731 status = nfs_sync_mapping(filp->f_mapping);
732 if (status != 0)
733 goto out;
734
735 /*
736 * Use local locking if mounted with "-onolock" or with appropriate
737 * "-olocal_lock="
738 */
739 if (!is_local)
740 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
741 else
742 status = locks_lock_file_wait(filp, fl);
743 if (status < 0)
744 goto out;
745
746 /*
747 * Invalidate cache to prevent missing any changes. If
748 * the file is mapped, clear the page cache as well so
749 * those mappings will be loaded.
750 *
751 * This makes locking act as a cache coherency point.
752 */
753 nfs_sync_mapping(filp->f_mapping);
754 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
755 nfs_zap_caches(inode);
756 if (mapping_mapped(filp->f_mapping))
757 nfs_revalidate_mapping(inode, filp->f_mapping);
758 }
759out:
760 return status;
761}
762
763/*
764 * Lock a (portion of) a file
765 */
766int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
767{
768 struct inode *inode = filp->f_mapping->host;
769 int ret = -ENOLCK;
770 int is_local = 0;
771
772 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
773 filp, fl->fl_type, fl->fl_flags,
774 (long long)fl->fl_start, (long long)fl->fl_end);
775
776 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
777
778 /* No mandatory locks over NFS */
779 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
780 goto out_err;
781
782 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
783 is_local = 1;
784
785 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
786 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
787 if (ret < 0)
788 goto out_err;
789 }
790
791 if (IS_GETLK(cmd))
792 ret = do_getlk(filp, cmd, fl, is_local);
793 else if (fl->fl_type == F_UNLCK)
794 ret = do_unlk(filp, cmd, fl, is_local);
795 else
796 ret = do_setlk(filp, cmd, fl, is_local);
797out_err:
798 return ret;
799}
800EXPORT_SYMBOL_GPL(nfs_lock);
801
802/*
803 * Lock a (portion of) a file
804 */
805int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
806{
807 struct inode *inode = filp->f_mapping->host;
808 int is_local = 0;
809
810 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
811 filp, fl->fl_type, fl->fl_flags);
812
813 if (!(fl->fl_flags & FL_FLOCK))
814 return -ENOLCK;
815
816 /*
817 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
818 * any standard. In principle we might be able to support LOCK_MAND
819 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
820 * NFS code is not set up for it.
821 */
822 if (fl->fl_type & LOCK_MAND)
823 return -EINVAL;
824
825 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
826 is_local = 1;
827
828 /* We're simulating flock() locks using posix locks on the server */
829 if (fl->fl_type == F_UNLCK)
830 return do_unlk(filp, cmd, fl, is_local);
831 return do_setlk(filp, cmd, fl, is_local);
832}
833EXPORT_SYMBOL_GPL(nfs_flock);
834
835const struct file_operations nfs_file_operations = {
836 .llseek = nfs_file_llseek,
837 .read_iter = nfs_file_read,
838 .write_iter = nfs_file_write,
839 .mmap = nfs_file_mmap,
840 .open = nfs_file_open,
841 .flush = nfs_file_flush,
842 .release = nfs_file_release,
843 .fsync = nfs_file_fsync,
844 .lock = nfs_lock,
845 .flock = nfs_flock,
846 .splice_read = generic_file_splice_read,
847 .splice_write = iter_file_splice_write,
848 .check_flags = nfs_check_flags,
849 .setlease = simple_nosetlease,
850};
851EXPORT_SYMBOL_GPL(nfs_file_operations);
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->fl_type;
724
725 /* Try local locking first */
726 posix_test_lock(filp, fl);
727 if (fl->fl_type != F_UNLCK) {
728 /* found a conflict */
729 goto out;
730 }
731 fl->fl_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->fl_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->fl_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->fl_type, fl->fl_flags,
836 (long long)fl->fl_start, (long long)fl->fl_end);
837
838 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
839
840 if (fl->fl_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 (fl->fl_type == F_UNLCK)
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->fl_type, fl->fl_flags);
873
874 if (!(fl->fl_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 (fl->fl_type == F_UNLCK)
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);