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/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
19#include <linux/module.h>
20#include <linux/time.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/fcntl.h>
24#include <linux/stat.h>
25#include <linux/nfs_fs.h>
26#include <linux/nfs_mount.h>
27#include <linux/mm.h>
28#include <linux/pagemap.h>
29#include <linux/aio.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32
33#include <asm/uaccess.h>
34
35#include "delegation.h"
36#include "internal.h"
37#include "iostat.h"
38#include "fscache.h"
39
40#include "nfstrace.h"
41
42#define NFSDBG_FACILITY NFSDBG_FILE
43
44static const struct vm_operations_struct nfs_file_vm_ops;
45
46/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
51int nfs_check_flags(int flags)
52{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
58EXPORT_SYMBOL_GPL(nfs_check_flags);
59
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
66 int res;
67
68 dprintk("NFS: open file(%pD2)\n", filp);
69
70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
75 res = nfs_open(inode, filp);
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 return nfs_release(inode, filp);
86}
87EXPORT_SYMBOL_GPL(nfs_file_release);
88
89/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
121{
122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
124
125 /*
126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
127 * the cached file length
128 */
129 if (whence != SEEK_SET && whence != SEEK_CUR) {
130 struct inode *inode = filp->f_mapping->host;
131
132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
135 }
136
137 return generic_file_llseek(filp, offset, whence);
138}
139EXPORT_SYMBOL_GPL(nfs_file_llseek);
140
141/*
142 * Flush all dirty pages, and check for write errors.
143 */
144int
145nfs_file_flush(struct file *file, fl_owner_t id)
146{
147 struct inode *inode = file_inode(file);
148
149 dprintk("NFS: flush(%pD2)\n", file);
150
151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
154
155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
160 return filemap_fdatawrite(file->f_mapping);
161
162 /* Flush writes to the server and return any errors */
163 return vfs_fsync(file, 0);
164}
165EXPORT_SYMBOL_GPL(nfs_file_flush);
166
167ssize_t
168nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
170{
171 struct inode *inode = file_inode(iocb->ki_filp);
172 ssize_t result;
173
174 if (iocb->ki_filp->f_flags & O_DIRECT)
175 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
176
177 dprintk("NFS: read(%pD2, %lu@%lu)\n",
178 iocb->ki_filp,
179 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
180
181 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
182 if (!result) {
183 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
184 if (result > 0)
185 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 }
187 return result;
188}
189EXPORT_SYMBOL_GPL(nfs_file_read);
190
191ssize_t
192nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 struct pipe_inode_info *pipe, size_t count,
194 unsigned int flags)
195{
196 struct inode *inode = file_inode(filp);
197 ssize_t res;
198
199 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 filp, (unsigned long) count, (unsigned long long) *ppos);
201
202 res = nfs_revalidate_mapping(inode, filp->f_mapping);
203 if (!res) {
204 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
205 if (res > 0)
206 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 }
208 return res;
209}
210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
211
212int
213nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214{
215 struct inode *inode = file_inode(file);
216 int status;
217
218 dprintk("NFS: mmap(%pD2)\n", file);
219
220 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 * so we call that before revalidating the mapping
222 */
223 status = generic_file_mmap(file, vma);
224 if (!status) {
225 vma->vm_ops = &nfs_file_vm_ops;
226 status = nfs_revalidate_mapping(inode, file->f_mapping);
227 }
228 return status;
229}
230EXPORT_SYMBOL_GPL(nfs_file_mmap);
231
232/*
233 * Flush any dirty pages for this process, and check for write errors.
234 * The return status from this call provides a reliable indication of
235 * whether any write errors occurred for this process.
236 *
237 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238 * disk, but it retrieves and clears ctx->error after synching, despite
239 * the two being set at the same time in nfs_context_set_write_error().
240 * This is because the former is used to notify the _next_ call to
241 * nfs_file_write() that a write error occurred, and hence cause it to
242 * fall back to doing a synchronous write.
243 */
244int
245nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
246{
247 struct nfs_open_context *ctx = nfs_file_open_context(file);
248 struct inode *inode = file_inode(file);
249 int have_error, do_resend, status;
250 int ret = 0;
251
252 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
253
254 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
255 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
256 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 status = nfs_commit_inode(inode, FLUSH_SYNC);
258 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
259 if (have_error) {
260 ret = xchg(&ctx->error, 0);
261 if (ret)
262 goto out;
263 }
264 if (status < 0) {
265 ret = status;
266 goto out;
267 }
268 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 if (do_resend)
270 ret = -EAGAIN;
271out:
272 return ret;
273}
274EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
275
276static int
277nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278{
279 int ret;
280 struct inode *inode = file_inode(file);
281
282 trace_nfs_fsync_enter(inode);
283
284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
298 } while (ret == -EAGAIN);
299
300 trace_nfs_fsync_exit(inode, ret);
301 return ret;
302}
303
304/*
305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
331 if ((file->f_mode & FMODE_READ) && /* open for read? */
332 !PageUptodate(page) && /* Uptodate? */
333 !PagePrivate(page) && /* i/o request already? */
334 pglen && /* valid bytes of file? */
335 (end < pglen || offset)) /* replace all valid bytes? */
336 return 1;
337 return 0;
338}
339
340/*
341 * This does the "real" work of the write. We must allocate and lock the
342 * page to be sent back to the generic routine, which then copies the
343 * data from user space.
344 *
345 * If the writer ends up delaying the write, the writer needs to
346 * increment the page use counts until he is done with the page.
347 */
348static int nfs_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
351{
352 int ret;
353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
354 struct page *page;
355 int once_thru = 0;
356
357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
358 file, mapping->host->i_ino, len, (long long) pos);
359
360start:
361 /*
362 * Prevent starvation issues if someone is doing a consistency
363 * sync-to-disk
364 */
365 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
366 nfs_wait_bit_killable, TASK_KILLABLE);
367 if (ret)
368 return ret;
369
370 page = grab_cache_page_write_begin(mapping, index, flags);
371 if (!page)
372 return -ENOMEM;
373 *pagep = page;
374
375 ret = nfs_flush_incompatible(file, page);
376 if (ret) {
377 unlock_page(page);
378 page_cache_release(page);
379 } else if (!once_thru &&
380 nfs_want_read_modify_write(file, page, pos, len)) {
381 once_thru = 1;
382 ret = nfs_readpage(file, page);
383 page_cache_release(page);
384 if (!ret)
385 goto start;
386 }
387 return ret;
388}
389
390static int nfs_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
393{
394 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
395 struct nfs_open_context *ctx = nfs_file_open_context(file);
396 int status;
397
398 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
399 file, mapping->host->i_ino, len, (long long) pos);
400
401 /*
402 * Zero any uninitialised parts of the page, and then mark the page
403 * as up to date if it turns out that we're extending the file.
404 */
405 if (!PageUptodate(page)) {
406 unsigned pglen = nfs_page_length(page);
407 unsigned end = offset + len;
408
409 if (pglen == 0) {
410 zero_user_segments(page, 0, offset,
411 end, PAGE_CACHE_SIZE);
412 SetPageUptodate(page);
413 } else if (end >= pglen) {
414 zero_user_segment(page, end, PAGE_CACHE_SIZE);
415 if (offset == 0)
416 SetPageUptodate(page);
417 } else
418 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
419 }
420
421 status = nfs_updatepage(file, page, offset, copied);
422
423 unlock_page(page);
424 page_cache_release(page);
425
426 if (status < 0)
427 return status;
428 NFS_I(mapping->host)->write_io += copied;
429
430 if (nfs_ctx_key_to_expire(ctx)) {
431 status = nfs_wb_all(mapping->host);
432 if (status < 0)
433 return status;
434 }
435
436 return copied;
437}
438
439/*
440 * Partially or wholly invalidate a page
441 * - Release the private state associated with a page if undergoing complete
442 * page invalidation
443 * - Called if either PG_private or PG_fscache is set on the page
444 * - Caller holds page lock
445 */
446static void nfs_invalidate_page(struct page *page, unsigned int offset,
447 unsigned int length)
448{
449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
450 page, offset, length);
451
452 if (offset != 0 || length < PAGE_CACHE_SIZE)
453 return;
454 /* Cancel any unstarted writes on this page */
455 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
456
457 nfs_fscache_invalidate_page(page, page->mapping->host);
458}
459
460/*
461 * Attempt to release the private state associated with a page
462 * - Called if either PG_private or PG_fscache is set on the page
463 * - Caller holds page lock
464 * - Return true (may release page) or false (may not)
465 */
466static int nfs_release_page(struct page *page, gfp_t gfp)
467{
468 struct address_space *mapping = page->mapping;
469
470 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
471
472 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
473 * doing this memory reclaim for a fs-related allocation.
474 */
475 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
476 !(current->flags & PF_FSTRANS)) {
477 int how = FLUSH_SYNC;
478
479 /* Don't let kswapd deadlock waiting for OOM RPC calls */
480 if (current_is_kswapd())
481 how = 0;
482 nfs_commit_inode(mapping->host, how);
483 }
484 /* If PagePrivate() is set, then the page is not freeable */
485 if (PagePrivate(page))
486 return 0;
487 return nfs_fscache_release_page(page, gfp);
488}
489
490static void nfs_check_dirty_writeback(struct page *page,
491 bool *dirty, bool *writeback)
492{
493 struct nfs_inode *nfsi;
494 struct address_space *mapping = page_file_mapping(page);
495
496 if (!mapping || PageSwapCache(page))
497 return;
498
499 /*
500 * Check if an unstable page is currently being committed and
501 * if so, have the VM treat it as if the page is under writeback
502 * so it will not block due to pages that will shortly be freeable.
503 */
504 nfsi = NFS_I(mapping->host);
505 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
506 *writeback = true;
507 return;
508 }
509
510 /*
511 * If PagePrivate() is set, then the page is not freeable and as the
512 * inode is not being committed, it's not going to be cleaned in the
513 * near future so treat it as dirty
514 */
515 if (PagePrivate(page))
516 *dirty = true;
517}
518
519/*
520 * Attempt to clear the private state associated with a page when an error
521 * occurs that requires the cached contents of an inode to be written back or
522 * destroyed
523 * - Called if either PG_private or fscache is set on the page
524 * - Caller holds page lock
525 * - Return 0 if successful, -error otherwise
526 */
527static int nfs_launder_page(struct page *page)
528{
529 struct inode *inode = page_file_mapping(page)->host;
530 struct nfs_inode *nfsi = NFS_I(inode);
531
532 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
533 inode->i_ino, (long long)page_offset(page));
534
535 nfs_fscache_wait_on_page_write(nfsi, page);
536 return nfs_wb_page(inode, page);
537}
538
539#ifdef CONFIG_NFS_SWAP
540static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
541 sector_t *span)
542{
543 *span = sis->pages;
544 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
545}
546
547static void nfs_swap_deactivate(struct file *file)
548{
549 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
550}
551#endif
552
553const struct address_space_operations nfs_file_aops = {
554 .readpage = nfs_readpage,
555 .readpages = nfs_readpages,
556 .set_page_dirty = __set_page_dirty_nobuffers,
557 .writepage = nfs_writepage,
558 .writepages = nfs_writepages,
559 .write_begin = nfs_write_begin,
560 .write_end = nfs_write_end,
561 .invalidatepage = nfs_invalidate_page,
562 .releasepage = nfs_release_page,
563 .direct_IO = nfs_direct_IO,
564 .migratepage = nfs_migrate_page,
565 .launder_page = nfs_launder_page,
566 .is_dirty_writeback = nfs_check_dirty_writeback,
567 .error_remove_page = generic_error_remove_page,
568#ifdef CONFIG_NFS_SWAP
569 .swap_activate = nfs_swap_activate,
570 .swap_deactivate = nfs_swap_deactivate,
571#endif
572};
573
574/*
575 * Notification that a PTE pointing to an NFS page is about to be made
576 * writable, implying that someone is about to modify the page through a
577 * shared-writable mapping
578 */
579static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
580{
581 struct page *page = vmf->page;
582 struct file *filp = vma->vm_file;
583 struct inode *inode = file_inode(filp);
584 unsigned pagelen;
585 int ret = VM_FAULT_NOPAGE;
586 struct address_space *mapping;
587
588 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
589 filp, filp->f_mapping->host->i_ino,
590 (long long)page_offset(page));
591
592 /* make sure the cache has finished storing the page */
593 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
594
595 lock_page(page);
596 mapping = page_file_mapping(page);
597 if (mapping != inode->i_mapping)
598 goto out_unlock;
599
600 wait_on_page_writeback(page);
601
602 pagelen = nfs_page_length(page);
603 if (pagelen == 0)
604 goto out_unlock;
605
606 ret = VM_FAULT_LOCKED;
607 if (nfs_flush_incompatible(filp, page) == 0 &&
608 nfs_updatepage(filp, page, 0, pagelen) == 0)
609 goto out;
610
611 ret = VM_FAULT_SIGBUS;
612out_unlock:
613 unlock_page(page);
614out:
615 return ret;
616}
617
618static const struct vm_operations_struct nfs_file_vm_ops = {
619 .fault = filemap_fault,
620 .map_pages = filemap_map_pages,
621 .page_mkwrite = nfs_vm_page_mkwrite,
622 .remap_pages = generic_file_remap_pages,
623};
624
625static int nfs_need_sync_write(struct file *filp, struct inode *inode)
626{
627 struct nfs_open_context *ctx;
628
629 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
630 return 1;
631 ctx = nfs_file_open_context(filp);
632 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
633 nfs_ctx_key_to_expire(ctx))
634 return 1;
635 return 0;
636}
637
638ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
639 unsigned long nr_segs, loff_t pos)
640{
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
643 unsigned long written = 0;
644 ssize_t result;
645 size_t count = iov_length(iov, nr_segs);
646
647 result = nfs_key_timeout_notify(file, inode);
648 if (result)
649 return result;
650
651 if (file->f_flags & O_DIRECT)
652 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
653
654 dprintk("NFS: write(%pD2, %lu@%Ld)\n",
655 file, (unsigned long) count, (long long) pos);
656
657 result = -EBUSY;
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
663 if (file->f_flags & O_APPEND) {
664 result = nfs_revalidate_file_size(inode, file);
665 if (result)
666 goto out;
667 }
668
669 result = count;
670 if (!count)
671 goto out;
672
673 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
674 if (result > 0)
675 written = result;
676
677 /* Return error values for O_DSYNC and IS_SYNC() */
678 if (result >= 0 && nfs_need_sync_write(file, inode)) {
679 int err = vfs_fsync(file, 0);
680 if (err < 0)
681 result = err;
682 }
683 if (result > 0)
684 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
685out:
686 return result;
687
688out_swapfile:
689 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 goto out;
691}
692EXPORT_SYMBOL_GPL(nfs_file_write);
693
694ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
695 struct file *filp, loff_t *ppos,
696 size_t count, unsigned int flags)
697{
698 struct inode *inode = file_inode(filp);
699 unsigned long written = 0;
700 ssize_t ret;
701
702 dprintk("NFS splice_write(%pD2, %lu@%llu)\n",
703 filp, (unsigned long) count, (unsigned long long) *ppos);
704
705 /*
706 * The combination of splice and an O_APPEND destination is disallowed.
707 */
708
709 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
710 if (ret > 0)
711 written = ret;
712
713 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
714 int err = vfs_fsync(filp, 0);
715 if (err < 0)
716 ret = err;
717 }
718 if (ret > 0)
719 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
720 return ret;
721}
722EXPORT_SYMBOL_GPL(nfs_file_splice_write);
723
724static int
725do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
726{
727 struct inode *inode = filp->f_mapping->host;
728 int status = 0;
729 unsigned int saved_type = fl->fl_type;
730
731 /* Try local locking first */
732 posix_test_lock(filp, fl);
733 if (fl->fl_type != F_UNLCK) {
734 /* found a conflict */
735 goto out;
736 }
737 fl->fl_type = saved_type;
738
739 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
740 goto out_noconflict;
741
742 if (is_local)
743 goto out_noconflict;
744
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746out:
747 return status;
748out_noconflict:
749 fl->fl_type = F_UNLCK;
750 goto out;
751}
752
753static int do_vfs_lock(struct file *file, struct file_lock *fl)
754{
755 int res = 0;
756 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
757 case FL_POSIX:
758 res = posix_lock_file_wait(file, fl);
759 break;
760 case FL_FLOCK:
761 res = flock_lock_file_wait(file, fl);
762 break;
763 default:
764 BUG();
765 }
766 return res;
767}
768
769static int
770do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
771{
772 struct inode *inode = filp->f_mapping->host;
773 struct nfs_lock_context *l_ctx;
774 int status;
775
776 /*
777 * Flush all pending writes before doing anything
778 * with locks..
779 */
780 nfs_sync_mapping(filp->f_mapping);
781
782 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
783 if (!IS_ERR(l_ctx)) {
784 status = nfs_iocounter_wait(&l_ctx->io_count);
785 nfs_put_lock_context(l_ctx);
786 if (status < 0)
787 return status;
788 }
789
790 /* NOTE: special case
791 * If we're signalled while cleaning up locks on process exit, we
792 * still need to complete the unlock.
793 */
794 /*
795 * Use local locking if mounted with "-onolock" or with appropriate
796 * "-olocal_lock="
797 */
798 if (!is_local)
799 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
800 else
801 status = do_vfs_lock(filp, fl);
802 return status;
803}
804
805static int
806is_time_granular(struct timespec *ts) {
807 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
808}
809
810static int
811do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
812{
813 struct inode *inode = filp->f_mapping->host;
814 int status;
815
816 /*
817 * Flush all pending writes before doing anything
818 * with locks..
819 */
820 status = nfs_sync_mapping(filp->f_mapping);
821 if (status != 0)
822 goto out;
823
824 /*
825 * Use local locking if mounted with "-onolock" or with appropriate
826 * "-olocal_lock="
827 */
828 if (!is_local)
829 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
830 else
831 status = do_vfs_lock(filp, fl);
832 if (status < 0)
833 goto out;
834
835 /*
836 * Revalidate the cache if the server has time stamps granular
837 * enough to detect subsecond changes. Otherwise, clear the
838 * cache to prevent missing any changes.
839 *
840 * This makes locking act as a cache coherency point.
841 */
842 nfs_sync_mapping(filp->f_mapping);
843 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
844 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
845 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
846 else
847 nfs_zap_caches(inode);
848 }
849out:
850 return status;
851}
852
853/*
854 * Lock a (portion of) a file
855 */
856int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
857{
858 struct inode *inode = filp->f_mapping->host;
859 int ret = -ENOLCK;
860 int is_local = 0;
861
862 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
863 filp, fl->fl_type, fl->fl_flags,
864 (long long)fl->fl_start, (long long)fl->fl_end);
865
866 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
867
868 /* No mandatory locks over NFS */
869 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
870 goto out_err;
871
872 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
873 is_local = 1;
874
875 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
876 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
877 if (ret < 0)
878 goto out_err;
879 }
880
881 if (IS_GETLK(cmd))
882 ret = do_getlk(filp, cmd, fl, is_local);
883 else if (fl->fl_type == F_UNLCK)
884 ret = do_unlk(filp, cmd, fl, is_local);
885 else
886 ret = do_setlk(filp, cmd, fl, is_local);
887out_err:
888 return ret;
889}
890EXPORT_SYMBOL_GPL(nfs_lock);
891
892/*
893 * Lock a (portion of) a file
894 */
895int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
896{
897 struct inode *inode = filp->f_mapping->host;
898 int is_local = 0;
899
900 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
901 filp, fl->fl_type, fl->fl_flags);
902
903 if (!(fl->fl_flags & FL_FLOCK))
904 return -ENOLCK;
905
906 /*
907 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
908 * any standard. In principle we might be able to support LOCK_MAND
909 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
910 * NFS code is not set up for it.
911 */
912 if (fl->fl_type & LOCK_MAND)
913 return -EINVAL;
914
915 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
916 is_local = 1;
917
918 /* We're simulating flock() locks using posix locks on the server */
919 fl->fl_owner = (fl_owner_t)filp;
920 fl->fl_start = 0;
921 fl->fl_end = OFFSET_MAX;
922
923 if (fl->fl_type == F_UNLCK)
924 return do_unlk(filp, cmd, fl, is_local);
925 return do_setlk(filp, cmd, fl, is_local);
926}
927EXPORT_SYMBOL_GPL(nfs_flock);
928
929/*
930 * There is no protocol support for leases, so we have no way to implement
931 * them correctly in the face of opens by other clients.
932 */
933int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
934{
935 dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg);
936 return -EINVAL;
937}
938EXPORT_SYMBOL_GPL(nfs_setlease);
939
940const struct file_operations nfs_file_operations = {
941 .llseek = nfs_file_llseek,
942 .read = do_sync_read,
943 .write = do_sync_write,
944 .aio_read = nfs_file_read,
945 .aio_write = nfs_file_write,
946 .mmap = nfs_file_mmap,
947 .open = nfs_file_open,
948 .flush = nfs_file_flush,
949 .release = nfs_file_release,
950 .fsync = nfs_file_fsync,
951 .lock = nfs_lock,
952 .flock = nfs_flock,
953 .splice_read = nfs_file_splice_read,
954 .splice_write = nfs_file_splice_write,
955 .check_flags = nfs_check_flags,
956 .setlease = nfs_setlease,
957};
958EXPORT_SYMBOL_GPL(nfs_file_operations);