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/gfp.h>
30#include <linux/swap.h>
31
32#include <linux/uaccess.h>
33
34#include "delegation.h"
35#include "internal.h"
36#include "iostat.h"
37#include "fscache.h"
38#include "pnfs.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 nfs_file_clear_open_context(filp);
86 return 0;
87}
88EXPORT_SYMBOL_GPL(nfs_file_release);
89
90/**
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - 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_REVAL_PAGECACHE))
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
143 dprintk("NFS: flush(%pD2)\n", file);
144
145 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
146 if ((file->f_mode & FMODE_WRITE) == 0)
147 return 0;
148
149 /* Flush writes to the server and return any errors */
150 return vfs_fsync(file, 0);
151}
152
153ssize_t
154nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
155{
156 struct inode *inode = file_inode(iocb->ki_filp);
157 ssize_t result;
158
159 if (iocb->ki_flags & IOCB_DIRECT)
160 return nfs_file_direct_read(iocb, to);
161
162 dprintk("NFS: read(%pD2, %zu@%lu)\n",
163 iocb->ki_filp,
164 iov_iter_count(to), (unsigned long) iocb->ki_pos);
165
166 nfs_start_io_read(inode);
167 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
168 if (!result) {
169 result = generic_file_read_iter(iocb, to);
170 if (result > 0)
171 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
172 }
173 nfs_end_io_read(inode);
174 return result;
175}
176EXPORT_SYMBOL_GPL(nfs_file_read);
177
178int
179nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
180{
181 struct inode *inode = file_inode(file);
182 int status;
183
184 dprintk("NFS: mmap(%pD2)\n", file);
185
186 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
187 * so we call that before revalidating the mapping
188 */
189 status = generic_file_mmap(file, vma);
190 if (!status) {
191 vma->vm_ops = &nfs_file_vm_ops;
192 status = nfs_revalidate_mapping(inode, file->f_mapping);
193 }
194 return status;
195}
196EXPORT_SYMBOL_GPL(nfs_file_mmap);
197
198/*
199 * Flush any dirty pages for this process, and check for write errors.
200 * The return status from this call provides a reliable indication of
201 * whether any write errors occurred for this process.
202 *
203 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
204 * disk, but it retrieves and clears ctx->error after synching, despite
205 * the two being set at the same time in nfs_context_set_write_error().
206 * This is because the former is used to notify the _next_ call to
207 * nfs_file_write() that a write error occurred, and hence cause it to
208 * fall back to doing a synchronous write.
209 */
210static int
211nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
212{
213 struct nfs_open_context *ctx = nfs_file_open_context(file);
214 struct inode *inode = file_inode(file);
215 int have_error, do_resend, status;
216 int ret = 0;
217
218 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
219
220 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
221 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
222 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
223 status = nfs_commit_inode(inode, FLUSH_SYNC);
224 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
225 if (have_error) {
226 ret = xchg(&ctx->error, 0);
227 if (ret)
228 goto out;
229 }
230 if (status < 0) {
231 ret = status;
232 goto out;
233 }
234 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
235 if (do_resend)
236 ret = -EAGAIN;
237out:
238 return ret;
239}
240
241int
242nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
243{
244 int ret;
245 struct inode *inode = file_inode(file);
246
247 trace_nfs_fsync_enter(inode);
248
249 do {
250 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
251 if (ret != 0)
252 break;
253 ret = nfs_file_fsync_commit(file, start, end, datasync);
254 if (!ret)
255 ret = pnfs_sync_inode(inode, !!datasync);
256 /*
257 * If nfs_file_fsync_commit detected a server reboot, then
258 * resend all dirty pages that might have been covered by
259 * the NFS_CONTEXT_RESEND_WRITES flag
260 */
261 start = 0;
262 end = LLONG_MAX;
263 } while (ret == -EAGAIN);
264
265 trace_nfs_fsync_exit(inode, ret);
266 return ret;
267}
268EXPORT_SYMBOL_GPL(nfs_file_fsync);
269
270/*
271 * Decide whether a read/modify/write cycle may be more efficient
272 * then a modify/write/read cycle when writing to a page in the
273 * page cache.
274 *
275 * The modify/write/read cycle may occur if a page is read before
276 * being completely filled by the writer. In this situation, the
277 * page must be completely written to stable storage on the server
278 * before it can be refilled by reading in the page from the server.
279 * This can lead to expensive, small, FILE_SYNC mode writes being
280 * done.
281 *
282 * It may be more efficient to read the page first if the file is
283 * open for reading in addition to writing, the page is not marked
284 * as Uptodate, it is not dirty or waiting to be committed,
285 * indicating that it was previously allocated and then modified,
286 * that there were valid bytes of data in that range of the file,
287 * and that the new data won't completely replace the old data in
288 * that range of the file.
289 */
290static int nfs_want_read_modify_write(struct file *file, struct page *page,
291 loff_t pos, unsigned len)
292{
293 unsigned int pglen = nfs_page_length(page);
294 unsigned int offset = pos & (PAGE_SIZE - 1);
295 unsigned int end = offset + len;
296
297 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
298 if (!PageUptodate(page))
299 return 1;
300 return 0;
301 }
302
303 if ((file->f_mode & FMODE_READ) && /* open for read? */
304 !PageUptodate(page) && /* Uptodate? */
305 !PagePrivate(page) && /* i/o request already? */
306 pglen && /* valid bytes of file? */
307 (end < pglen || offset)) /* replace all valid bytes? */
308 return 1;
309 return 0;
310}
311
312/*
313 * This does the "real" work of the write. We must allocate and lock the
314 * page to be sent back to the generic routine, which then copies the
315 * data from user space.
316 *
317 * If the writer ends up delaying the write, the writer needs to
318 * increment the page use counts until he is done with the page.
319 */
320static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned flags,
322 struct page **pagep, void **fsdata)
323{
324 int ret;
325 pgoff_t index = pos >> PAGE_SHIFT;
326 struct page *page;
327 int once_thru = 0;
328
329 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
330 file, mapping->host->i_ino, len, (long long) pos);
331
332start:
333 page = grab_cache_page_write_begin(mapping, index, flags);
334 if (!page)
335 return -ENOMEM;
336 *pagep = page;
337
338 ret = nfs_flush_incompatible(file, page);
339 if (ret) {
340 unlock_page(page);
341 put_page(page);
342 } else if (!once_thru &&
343 nfs_want_read_modify_write(file, page, pos, len)) {
344 once_thru = 1;
345 ret = nfs_readpage(file, page);
346 put_page(page);
347 if (!ret)
348 goto start;
349 }
350 return ret;
351}
352
353static int nfs_write_end(struct file *file, struct address_space *mapping,
354 loff_t pos, unsigned len, unsigned copied,
355 struct page *page, void *fsdata)
356{
357 unsigned offset = pos & (PAGE_SIZE - 1);
358 struct nfs_open_context *ctx = nfs_file_open_context(file);
359 int status;
360
361 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
362 file, mapping->host->i_ino, len, (long long) pos);
363
364 /*
365 * Zero any uninitialised parts of the page, and then mark the page
366 * as up to date if it turns out that we're extending the file.
367 */
368 if (!PageUptodate(page)) {
369 unsigned pglen = nfs_page_length(page);
370 unsigned end = offset + copied;
371
372 if (pglen == 0) {
373 zero_user_segments(page, 0, offset,
374 end, PAGE_SIZE);
375 SetPageUptodate(page);
376 } else if (end >= pglen) {
377 zero_user_segment(page, end, PAGE_SIZE);
378 if (offset == 0)
379 SetPageUptodate(page);
380 } else
381 zero_user_segment(page, pglen, PAGE_SIZE);
382 }
383
384 status = nfs_updatepage(file, page, offset, copied);
385
386 unlock_page(page);
387 put_page(page);
388
389 if (status < 0)
390 return status;
391 NFS_I(mapping->host)->write_io += copied;
392
393 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
394 status = nfs_wb_all(mapping->host);
395 if (status < 0)
396 return status;
397 }
398
399 return copied;
400}
401
402/*
403 * Partially or wholly invalidate a page
404 * - Release the private state associated with a page if undergoing complete
405 * page invalidation
406 * - Called if either PG_private or PG_fscache is set on the page
407 * - Caller holds page lock
408 */
409static void nfs_invalidate_page(struct page *page, unsigned int offset,
410 unsigned int length)
411{
412 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
413 page, offset, length);
414
415 if (offset != 0 || length < PAGE_SIZE)
416 return;
417 /* Cancel any unstarted writes on this page */
418 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
419
420 nfs_fscache_invalidate_page(page, page->mapping->host);
421}
422
423/*
424 * Attempt to release the private state associated with a page
425 * - Called if either PG_private or PG_fscache is set on the page
426 * - Caller holds page lock
427 * - Return true (may release page) or false (may not)
428 */
429static int nfs_release_page(struct page *page, gfp_t gfp)
430{
431 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
432
433 /* If PagePrivate() is set, then the page is not freeable */
434 if (PagePrivate(page))
435 return 0;
436 return nfs_fscache_release_page(page, gfp);
437}
438
439static void nfs_check_dirty_writeback(struct page *page,
440 bool *dirty, bool *writeback)
441{
442 struct nfs_inode *nfsi;
443 struct address_space *mapping = page_file_mapping(page);
444
445 if (!mapping || PageSwapCache(page))
446 return;
447
448 /*
449 * Check if an unstable page is currently being committed and
450 * if so, have the VM treat it as if the page is under writeback
451 * so it will not block due to pages that will shortly be freeable.
452 */
453 nfsi = NFS_I(mapping->host);
454 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
455 *writeback = true;
456 return;
457 }
458
459 /*
460 * If PagePrivate() is set, then the page is not freeable and as the
461 * inode is not being committed, it's not going to be cleaned in the
462 * near future so treat it as dirty
463 */
464 if (PagePrivate(page))
465 *dirty = true;
466}
467
468/*
469 * Attempt to clear the private state associated with a page when an error
470 * occurs that requires the cached contents of an inode to be written back or
471 * destroyed
472 * - Called if either PG_private or fscache is set on the page
473 * - Caller holds page lock
474 * - Return 0 if successful, -error otherwise
475 */
476static int nfs_launder_page(struct page *page)
477{
478 struct inode *inode = page_file_mapping(page)->host;
479 struct nfs_inode *nfsi = NFS_I(inode);
480
481 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
482 inode->i_ino, (long long)page_offset(page));
483
484 nfs_fscache_wait_on_page_write(nfsi, page);
485 return nfs_wb_launder_page(inode, page);
486}
487
488static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
489 sector_t *span)
490{
491 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
492
493 *span = sis->pages;
494
495 return rpc_clnt_swap_activate(clnt);
496}
497
498static void nfs_swap_deactivate(struct file *file)
499{
500 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
501
502 rpc_clnt_swap_deactivate(clnt);
503}
504
505const struct address_space_operations nfs_file_aops = {
506 .readpage = nfs_readpage,
507 .readpages = nfs_readpages,
508 .set_page_dirty = __set_page_dirty_nobuffers,
509 .writepage = nfs_writepage,
510 .writepages = nfs_writepages,
511 .write_begin = nfs_write_begin,
512 .write_end = nfs_write_end,
513 .invalidatepage = nfs_invalidate_page,
514 .releasepage = nfs_release_page,
515 .direct_IO = nfs_direct_IO,
516#ifdef CONFIG_MIGRATION
517 .migratepage = nfs_migrate_page,
518#endif
519 .launder_page = nfs_launder_page,
520 .is_dirty_writeback = nfs_check_dirty_writeback,
521 .error_remove_page = generic_error_remove_page,
522 .swap_activate = nfs_swap_activate,
523 .swap_deactivate = nfs_swap_deactivate,
524};
525
526/*
527 * Notification that a PTE pointing to an NFS page is about to be made
528 * writable, implying that someone is about to modify the page through a
529 * shared-writable mapping
530 */
531static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
532{
533 struct page *page = vmf->page;
534 struct file *filp = vma->vm_file;
535 struct inode *inode = file_inode(filp);
536 unsigned pagelen;
537 int ret = VM_FAULT_NOPAGE;
538 struct address_space *mapping;
539
540 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
541 filp, filp->f_mapping->host->i_ino,
542 (long long)page_offset(page));
543
544 sb_start_pagefault(inode->i_sb);
545
546 /* make sure the cache has finished storing the page */
547 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
548
549 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
550 nfs_wait_bit_killable, TASK_KILLABLE);
551
552 lock_page(page);
553 mapping = page_file_mapping(page);
554 if (mapping != inode->i_mapping)
555 goto out_unlock;
556
557 wait_on_page_writeback(page);
558
559 pagelen = nfs_page_length(page);
560 if (pagelen == 0)
561 goto out_unlock;
562
563 ret = VM_FAULT_LOCKED;
564 if (nfs_flush_incompatible(filp, page) == 0 &&
565 nfs_updatepage(filp, page, 0, pagelen) == 0)
566 goto out;
567
568 ret = VM_FAULT_SIGBUS;
569out_unlock:
570 unlock_page(page);
571out:
572 sb_end_pagefault(inode->i_sb);
573 return ret;
574}
575
576static const struct vm_operations_struct nfs_file_vm_ops = {
577 .fault = filemap_fault,
578 .map_pages = filemap_map_pages,
579 .page_mkwrite = nfs_vm_page_mkwrite,
580};
581
582static int nfs_need_check_write(struct file *filp, struct inode *inode)
583{
584 struct nfs_open_context *ctx;
585
586 ctx = nfs_file_open_context(filp);
587 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
588 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
621 nfs_start_io_write(inode);
622 result = generic_write_checks(iocb, from);
623 if (result > 0) {
624 current->backing_dev_info = inode_to_bdi(inode);
625 result = generic_perform_write(file, from, iocb->ki_pos);
626 current->backing_dev_info = NULL;
627 }
628 nfs_end_io_write(inode);
629 if (result <= 0)
630 goto out;
631
632 result = generic_write_sync(iocb, result);
633 if (result < 0)
634 goto out;
635 written = result;
636 iocb->ki_pos += written;
637
638 /* Return error values */
639 if (nfs_need_check_write(file, inode)) {
640 int err = vfs_fsync(file, 0);
641 if (err < 0)
642 result = err;
643 }
644 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
645out:
646 return result;
647
648out_swapfile:
649 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
650 return -EBUSY;
651}
652EXPORT_SYMBOL_GPL(nfs_file_write);
653
654static int
655do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
656{
657 struct inode *inode = filp->f_mapping->host;
658 int status = 0;
659 unsigned int saved_type = fl->fl_type;
660
661 /* Try local locking first */
662 posix_test_lock(filp, fl);
663 if (fl->fl_type != F_UNLCK) {
664 /* found a conflict */
665 goto out;
666 }
667 fl->fl_type = saved_type;
668
669 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
670 goto out_noconflict;
671
672 if (is_local)
673 goto out_noconflict;
674
675 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
676out:
677 return status;
678out_noconflict:
679 fl->fl_type = F_UNLCK;
680 goto out;
681}
682
683static int
684do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
685{
686 struct inode *inode = filp->f_mapping->host;
687 struct nfs_lock_context *l_ctx;
688 int status;
689
690 /*
691 * Flush all pending writes before doing anything
692 * with locks..
693 */
694 vfs_fsync(filp, 0);
695
696 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
697 if (!IS_ERR(l_ctx)) {
698 status = nfs_iocounter_wait(l_ctx);
699 nfs_put_lock_context(l_ctx);
700 if (status < 0)
701 return status;
702 }
703
704 /* NOTE: special case
705 * If we're signalled while cleaning up locks on process exit, we
706 * still need to complete the unlock.
707 */
708 /*
709 * Use local locking if mounted with "-onolock" or with appropriate
710 * "-olocal_lock="
711 */
712 if (!is_local)
713 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
714 else
715 status = locks_lock_file_wait(filp, fl);
716 return status;
717}
718
719static int
720do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
721{
722 struct inode *inode = filp->f_mapping->host;
723 int status;
724
725 /*
726 * Flush all pending writes before doing anything
727 * with locks..
728 */
729 status = nfs_sync_mapping(filp->f_mapping);
730 if (status != 0)
731 goto out;
732
733 /*
734 * Use local locking if mounted with "-onolock" or with appropriate
735 * "-olocal_lock="
736 */
737 if (!is_local)
738 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
739 else
740 status = locks_lock_file_wait(filp, fl);
741 if (status < 0)
742 goto out;
743
744 /*
745 * Revalidate the cache if the server has time stamps granular
746 * enough to detect subsecond changes. Otherwise, clear the
747 * cache to prevent missing any changes.
748 *
749 * This makes locking act as a cache coherency point.
750 */
751 nfs_sync_mapping(filp->f_mapping);
752 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
753 nfs_zap_mapping(inode, filp->f_mapping);
754out:
755 return status;
756}
757
758/*
759 * Lock a (portion of) a file
760 */
761int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
762{
763 struct inode *inode = filp->f_mapping->host;
764 int ret = -ENOLCK;
765 int is_local = 0;
766
767 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
768 filp, fl->fl_type, fl->fl_flags,
769 (long long)fl->fl_start, (long long)fl->fl_end);
770
771 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
772
773 /* No mandatory locks over NFS */
774 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
775 goto out_err;
776
777 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
778 is_local = 1;
779
780 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
781 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
782 if (ret < 0)
783 goto out_err;
784 }
785
786 if (IS_GETLK(cmd))
787 ret = do_getlk(filp, cmd, fl, is_local);
788 else if (fl->fl_type == F_UNLCK)
789 ret = do_unlk(filp, cmd, fl, is_local);
790 else
791 ret = do_setlk(filp, cmd, fl, is_local);
792out_err:
793 return ret;
794}
795EXPORT_SYMBOL_GPL(nfs_lock);
796
797/*
798 * Lock a (portion of) a file
799 */
800int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
801{
802 struct inode *inode = filp->f_mapping->host;
803 int is_local = 0;
804
805 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
806 filp, fl->fl_type, fl->fl_flags);
807
808 if (!(fl->fl_flags & FL_FLOCK))
809 return -ENOLCK;
810
811 /*
812 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
813 * any standard. In principle we might be able to support LOCK_MAND
814 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
815 * NFS code is not set up for it.
816 */
817 if (fl->fl_type & LOCK_MAND)
818 return -EINVAL;
819
820 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
821 is_local = 1;
822
823 /* We're simulating flock() locks using posix locks on the server */
824 if (fl->fl_type == F_UNLCK)
825 return do_unlk(filp, cmd, fl, is_local);
826 return do_setlk(filp, cmd, fl, is_local);
827}
828EXPORT_SYMBOL_GPL(nfs_flock);
829
830const struct file_operations nfs_file_operations = {
831 .llseek = nfs_file_llseek,
832 .read_iter = nfs_file_read,
833 .write_iter = nfs_file_write,
834 .mmap = nfs_file_mmap,
835 .open = nfs_file_open,
836 .flush = nfs_file_flush,
837 .release = nfs_file_release,
838 .fsync = nfs_file_fsync,
839 .lock = nfs_lock,
840 .flock = nfs_flock,
841 .splice_read = generic_file_splice_read,
842 .splice_write = iter_file_splice_write,
843 .check_flags = nfs_check_flags,
844 .setlease = simple_nosetlease,
845};
846EXPORT_SYMBOL_GPL(nfs_file_operations);