Loading...
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);
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);