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/gfp.h>
30#include <linux/swap.h>
31
32#include <asm/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 struct nfs_inode *nfsi = NFS_I(inode);
105
106 if (nfs_have_delegated_attributes(inode))
107 goto out_noreval;
108
109 if (filp->f_flags & O_DIRECT)
110 goto force_reval;
111 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112 goto force_reval;
113 if (nfs_attribute_timeout(inode))
114 goto force_reval;
115out_noreval:
116 return 0;
117force_reval:
118 return __nfs_revalidate_inode(server, inode);
119}
120
121loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
122{
123 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 filp, offset, whence);
125
126 /*
127 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
128 * the cached file length
129 */
130 if (whence != SEEK_SET && whence != SEEK_CUR) {
131 struct inode *inode = filp->f_mapping->host;
132
133 int retval = nfs_revalidate_file_size(inode, filp);
134 if (retval < 0)
135 return (loff_t)retval;
136 }
137
138 return generic_file_llseek(filp, offset, whence);
139}
140EXPORT_SYMBOL_GPL(nfs_file_llseek);
141
142/*
143 * Flush all dirty pages, and check for write errors.
144 */
145static int
146nfs_file_flush(struct file *file, fl_owner_t id)
147{
148 struct inode *inode = file_inode(file);
149
150 dprintk("NFS: flush(%pD2)\n", file);
151
152 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
153 if ((file->f_mode & FMODE_WRITE) == 0)
154 return 0;
155
156 /* Flush writes to the server and return any errors */
157 return vfs_fsync(file, 0);
158}
159
160ssize_t
161nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
162{
163 struct inode *inode = file_inode(iocb->ki_filp);
164 ssize_t result;
165
166 if (iocb->ki_flags & IOCB_DIRECT)
167 return nfs_file_direct_read(iocb, to, iocb->ki_pos);
168
169 dprintk("NFS: read(%pD2, %zu@%lu)\n",
170 iocb->ki_filp,
171 iov_iter_count(to), (unsigned long) iocb->ki_pos);
172
173 result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping);
174 if (!result) {
175 result = generic_file_read_iter(iocb, to);
176 if (result > 0)
177 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
178 }
179 return result;
180}
181EXPORT_SYMBOL_GPL(nfs_file_read);
182
183ssize_t
184nfs_file_splice_read(struct file *filp, loff_t *ppos,
185 struct pipe_inode_info *pipe, size_t count,
186 unsigned int flags)
187{
188 struct inode *inode = file_inode(filp);
189 ssize_t res;
190
191 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
192 filp, (unsigned long) count, (unsigned long long) *ppos);
193
194 res = nfs_revalidate_mapping_protected(inode, filp->f_mapping);
195 if (!res) {
196 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
197 if (res > 0)
198 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
199 }
200 return res;
201}
202EXPORT_SYMBOL_GPL(nfs_file_splice_read);
203
204int
205nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
206{
207 struct inode *inode = file_inode(file);
208 int status;
209
210 dprintk("NFS: mmap(%pD2)\n", file);
211
212 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
213 * so we call that before revalidating the mapping
214 */
215 status = generic_file_mmap(file, vma);
216 if (!status) {
217 vma->vm_ops = &nfs_file_vm_ops;
218 status = nfs_revalidate_mapping(inode, file->f_mapping);
219 }
220 return status;
221}
222EXPORT_SYMBOL_GPL(nfs_file_mmap);
223
224/*
225 * Flush any dirty pages for this process, and check for write errors.
226 * The return status from this call provides a reliable indication of
227 * whether any write errors occurred for this process.
228 *
229 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
230 * disk, but it retrieves and clears ctx->error after synching, despite
231 * the two being set at the same time in nfs_context_set_write_error().
232 * This is because the former is used to notify the _next_ call to
233 * nfs_file_write() that a write error occurred, and hence cause it to
234 * fall back to doing a synchronous write.
235 */
236static int
237nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
238{
239 struct nfs_open_context *ctx = nfs_file_open_context(file);
240 struct inode *inode = file_inode(file);
241 int have_error, do_resend, status;
242 int ret = 0;
243
244 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
245
246 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
247 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
248 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
249 status = nfs_commit_inode(inode, FLUSH_SYNC);
250 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
251 if (have_error) {
252 ret = xchg(&ctx->error, 0);
253 if (ret)
254 goto out;
255 }
256 if (status < 0) {
257 ret = status;
258 goto out;
259 }
260 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
261 if (do_resend)
262 ret = -EAGAIN;
263out:
264 return ret;
265}
266
267int
268nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
269{
270 int ret;
271 struct inode *inode = file_inode(file);
272
273 trace_nfs_fsync_enter(inode);
274
275 inode_dio_wait(inode);
276 do {
277 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
278 if (ret != 0)
279 break;
280 inode_lock(inode);
281 ret = nfs_file_fsync_commit(file, start, end, datasync);
282 if (!ret)
283 ret = pnfs_sync_inode(inode, !!datasync);
284 inode_unlock(inode);
285 /*
286 * If nfs_file_fsync_commit detected a server reboot, then
287 * resend all dirty pages that might have been covered by
288 * the NFS_CONTEXT_RESEND_WRITES flag
289 */
290 start = 0;
291 end = LLONG_MAX;
292 } while (ret == -EAGAIN);
293
294 trace_nfs_fsync_exit(inode, ret);
295 return ret;
296}
297EXPORT_SYMBOL_GPL(nfs_file_fsync);
298
299/*
300 * Decide whether a read/modify/write cycle may be more efficient
301 * then a modify/write/read cycle when writing to a page in the
302 * page cache.
303 *
304 * The modify/write/read cycle may occur if a page is read before
305 * being completely filled by the writer. In this situation, the
306 * page must be completely written to stable storage on the server
307 * before it can be refilled by reading in the page from the server.
308 * This can lead to expensive, small, FILE_SYNC mode writes being
309 * done.
310 *
311 * It may be more efficient to read the page first if the file is
312 * open for reading in addition to writing, the page is not marked
313 * as Uptodate, it is not dirty or waiting to be committed,
314 * indicating that it was previously allocated and then modified,
315 * that there were valid bytes of data in that range of the file,
316 * and that the new data won't completely replace the old data in
317 * that range of the file.
318 */
319static int nfs_want_read_modify_write(struct file *file, struct page *page,
320 loff_t pos, unsigned len)
321{
322 unsigned int pglen = nfs_page_length(page);
323 unsigned int offset = pos & (PAGE_SIZE - 1);
324 unsigned int end = offset + len;
325
326 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
327 if (!PageUptodate(page))
328 return 1;
329 return 0;
330 }
331
332 if ((file->f_mode & FMODE_READ) && /* open for read? */
333 !PageUptodate(page) && /* Uptodate? */
334 !PagePrivate(page) && /* i/o request already? */
335 pglen && /* valid bytes of file? */
336 (end < pglen || offset)) /* replace all valid bytes? */
337 return 1;
338 return 0;
339}
340
341/*
342 * This does the "real" work of the write. We must allocate and lock the
343 * page to be sent back to the generic routine, which then copies the
344 * data from user space.
345 *
346 * If the writer ends up delaying the write, the writer needs to
347 * increment the page use counts until he is done with the page.
348 */
349static int nfs_write_begin(struct file *file, struct address_space *mapping,
350 loff_t pos, unsigned len, unsigned flags,
351 struct page **pagep, void **fsdata)
352{
353 int ret;
354 pgoff_t index = pos >> PAGE_SHIFT;
355 struct page *page;
356 int once_thru = 0;
357
358 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
359 file, mapping->host->i_ino, len, (long long) pos);
360
361start:
362 /*
363 * Prevent starvation issues if someone is doing a consistency
364 * sync-to-disk
365 */
366 ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
367 nfs_wait_bit_killable, TASK_KILLABLE);
368 if (ret)
369 return ret;
370 /*
371 * Wait for O_DIRECT to complete
372 */
373 inode_dio_wait(mapping->host);
374
375 page = grab_cache_page_write_begin(mapping, index, flags);
376 if (!page)
377 return -ENOMEM;
378 *pagep = page;
379
380 ret = nfs_flush_incompatible(file, page);
381 if (ret) {
382 unlock_page(page);
383 put_page(page);
384 } else if (!once_thru &&
385 nfs_want_read_modify_write(file, page, pos, len)) {
386 once_thru = 1;
387 ret = nfs_readpage(file, page);
388 put_page(page);
389 if (!ret)
390 goto start;
391 }
392 return ret;
393}
394
395static int nfs_write_end(struct file *file, struct address_space *mapping,
396 loff_t pos, unsigned len, unsigned copied,
397 struct page *page, void *fsdata)
398{
399 unsigned offset = pos & (PAGE_SIZE - 1);
400 struct nfs_open_context *ctx = nfs_file_open_context(file);
401 int status;
402
403 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
404 file, mapping->host->i_ino, len, (long long) pos);
405
406 /*
407 * Zero any uninitialised parts of the page, and then mark the page
408 * as up to date if it turns out that we're extending the file.
409 */
410 if (!PageUptodate(page)) {
411 unsigned pglen = nfs_page_length(page);
412 unsigned end = offset + len;
413
414 if (pglen == 0) {
415 zero_user_segments(page, 0, offset,
416 end, PAGE_SIZE);
417 SetPageUptodate(page);
418 } else if (end >= pglen) {
419 zero_user_segment(page, end, PAGE_SIZE);
420 if (offset == 0)
421 SetPageUptodate(page);
422 } else
423 zero_user_segment(page, pglen, PAGE_SIZE);
424 }
425
426 status = nfs_updatepage(file, page, offset, copied);
427
428 unlock_page(page);
429 put_page(page);
430
431 if (status < 0)
432 return status;
433 NFS_I(mapping->host)->write_io += copied;
434
435 if (nfs_ctx_key_to_expire(ctx)) {
436 status = nfs_wb_all(mapping->host);
437 if (status < 0)
438 return status;
439 }
440
441 return copied;
442}
443
444/*
445 * Partially or wholly invalidate a page
446 * - Release the private state associated with a page if undergoing complete
447 * page invalidation
448 * - Called if either PG_private or PG_fscache is set on the page
449 * - Caller holds page lock
450 */
451static void nfs_invalidate_page(struct page *page, unsigned int offset,
452 unsigned int length)
453{
454 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
455 page, offset, length);
456
457 if (offset != 0 || length < PAGE_SIZE)
458 return;
459 /* Cancel any unstarted writes on this page */
460 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
461
462 nfs_fscache_invalidate_page(page, page->mapping->host);
463}
464
465/*
466 * Attempt to release the private state associated with a page
467 * - Called if either PG_private or PG_fscache is set on the page
468 * - Caller holds page lock
469 * - Return true (may release page) or false (may not)
470 */
471static int nfs_release_page(struct page *page, gfp_t gfp)
472{
473 struct address_space *mapping = page->mapping;
474
475 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
476
477 /* Always try to initiate a 'commit' if relevant, but only
478 * wait for it if the caller allows blocking. Even then,
479 * only wait 1 second and only if the 'bdi' is not congested.
480 * Waiting indefinitely can cause deadlocks when the NFS
481 * server is on this machine, when a new TCP connection is
482 * needed and in other rare cases. There is no particular
483 * need to wait extensively here. A short wait has the
484 * benefit that someone else can worry about the freezer.
485 */
486 if (mapping) {
487 struct nfs_server *nfss = NFS_SERVER(mapping->host);
488 nfs_commit_inode(mapping->host, 0);
489 if (gfpflags_allow_blocking(gfp) &&
490 !bdi_write_congested(&nfss->backing_dev_info)) {
491 wait_on_page_bit_killable_timeout(page, PG_private,
492 HZ);
493 if (PagePrivate(page))
494 set_bdi_congested(&nfss->backing_dev_info,
495 BLK_RW_ASYNC);
496 }
497 }
498 /* If PagePrivate() is set, then the page is not freeable */
499 if (PagePrivate(page))
500 return 0;
501 return nfs_fscache_release_page(page, gfp);
502}
503
504static void nfs_check_dirty_writeback(struct page *page,
505 bool *dirty, bool *writeback)
506{
507 struct nfs_inode *nfsi;
508 struct address_space *mapping = page_file_mapping(page);
509
510 if (!mapping || PageSwapCache(page))
511 return;
512
513 /*
514 * Check if an unstable page is currently being committed and
515 * if so, have the VM treat it as if the page is under writeback
516 * so it will not block due to pages that will shortly be freeable.
517 */
518 nfsi = NFS_I(mapping->host);
519 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
520 *writeback = true;
521 return;
522 }
523
524 /*
525 * If PagePrivate() is set, then the page is not freeable and as the
526 * inode is not being committed, it's not going to be cleaned in the
527 * near future so treat it as dirty
528 */
529 if (PagePrivate(page))
530 *dirty = true;
531}
532
533/*
534 * Attempt to clear the private state associated with a page when an error
535 * occurs that requires the cached contents of an inode to be written back or
536 * destroyed
537 * - Called if either PG_private or fscache is set on the page
538 * - Caller holds page lock
539 * - Return 0 if successful, -error otherwise
540 */
541static int nfs_launder_page(struct page *page)
542{
543 struct inode *inode = page_file_mapping(page)->host;
544 struct nfs_inode *nfsi = NFS_I(inode);
545
546 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
547 inode->i_ino, (long long)page_offset(page));
548
549 nfs_fscache_wait_on_page_write(nfsi, page);
550 return nfs_wb_launder_page(inode, page);
551}
552
553static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
554 sector_t *span)
555{
556 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
557
558 *span = sis->pages;
559
560 return rpc_clnt_swap_activate(clnt);
561}
562
563static void nfs_swap_deactivate(struct file *file)
564{
565 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
566
567 rpc_clnt_swap_deactivate(clnt);
568}
569
570const struct address_space_operations nfs_file_aops = {
571 .readpage = nfs_readpage,
572 .readpages = nfs_readpages,
573 .set_page_dirty = __set_page_dirty_nobuffers,
574 .writepage = nfs_writepage,
575 .writepages = nfs_writepages,
576 .write_begin = nfs_write_begin,
577 .write_end = nfs_write_end,
578 .invalidatepage = nfs_invalidate_page,
579 .releasepage = nfs_release_page,
580 .direct_IO = nfs_direct_IO,
581 .migratepage = nfs_migrate_page,
582 .launder_page = nfs_launder_page,
583 .is_dirty_writeback = nfs_check_dirty_writeback,
584 .error_remove_page = generic_error_remove_page,
585 .swap_activate = nfs_swap_activate,
586 .swap_deactivate = nfs_swap_deactivate,
587};
588
589/*
590 * Notification that a PTE pointing to an NFS page is about to be made
591 * writable, implying that someone is about to modify the page through a
592 * shared-writable mapping
593 */
594static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
595{
596 struct page *page = vmf->page;
597 struct file *filp = vma->vm_file;
598 struct inode *inode = file_inode(filp);
599 unsigned pagelen;
600 int ret = VM_FAULT_NOPAGE;
601 struct address_space *mapping;
602
603 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
604 filp, filp->f_mapping->host->i_ino,
605 (long long)page_offset(page));
606
607 /* make sure the cache has finished storing the page */
608 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
609
610 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
611 nfs_wait_bit_killable, TASK_KILLABLE);
612
613 lock_page(page);
614 mapping = page_file_mapping(page);
615 if (mapping != inode->i_mapping)
616 goto out_unlock;
617
618 wait_on_page_writeback(page);
619
620 pagelen = nfs_page_length(page);
621 if (pagelen == 0)
622 goto out_unlock;
623
624 ret = VM_FAULT_LOCKED;
625 if (nfs_flush_incompatible(filp, page) == 0 &&
626 nfs_updatepage(filp, page, 0, pagelen) == 0)
627 goto out;
628
629 ret = VM_FAULT_SIGBUS;
630out_unlock:
631 unlock_page(page);
632out:
633 return ret;
634}
635
636static const struct vm_operations_struct nfs_file_vm_ops = {
637 .fault = filemap_fault,
638 .map_pages = filemap_map_pages,
639 .page_mkwrite = nfs_vm_page_mkwrite,
640};
641
642static int nfs_need_check_write(struct file *filp, struct inode *inode)
643{
644 struct nfs_open_context *ctx;
645
646 ctx = nfs_file_open_context(filp);
647 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
648 nfs_ctx_key_to_expire(ctx))
649 return 1;
650 return 0;
651}
652
653ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
654{
655 struct file *file = iocb->ki_filp;
656 struct inode *inode = file_inode(file);
657 unsigned long written = 0;
658 ssize_t result;
659 size_t count = iov_iter_count(from);
660
661 result = nfs_key_timeout_notify(file, inode);
662 if (result)
663 return result;
664
665 if (iocb->ki_flags & IOCB_DIRECT) {
666 result = generic_write_checks(iocb, from);
667 if (result <= 0)
668 return result;
669 return nfs_file_direct_write(iocb, from);
670 }
671
672 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
673 file, count, (long long) iocb->ki_pos);
674
675 result = -EBUSY;
676 if (IS_SWAPFILE(inode))
677 goto out_swapfile;
678 /*
679 * O_APPEND implies that we must revalidate the file length.
680 */
681 if (iocb->ki_flags & IOCB_APPEND) {
682 result = nfs_revalidate_file_size(inode, file);
683 if (result)
684 goto out;
685 }
686
687 result = count;
688 if (!count)
689 goto out;
690
691 result = generic_file_write_iter(iocb, from);
692 if (result > 0)
693 written = result;
694
695 /* Return error values */
696 if (result >= 0 && nfs_need_check_write(file, inode)) {
697 int err = vfs_fsync(file, 0);
698 if (err < 0)
699 result = err;
700 }
701 if (result > 0)
702 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
703out:
704 return result;
705
706out_swapfile:
707 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
708 goto out;
709}
710EXPORT_SYMBOL_GPL(nfs_file_write);
711
712static int
713do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
714{
715 struct inode *inode = filp->f_mapping->host;
716 int status = 0;
717 unsigned int saved_type = fl->fl_type;
718
719 /* Try local locking first */
720 posix_test_lock(filp, fl);
721 if (fl->fl_type != F_UNLCK) {
722 /* found a conflict */
723 goto out;
724 }
725 fl->fl_type = saved_type;
726
727 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
728 goto out_noconflict;
729
730 if (is_local)
731 goto out_noconflict;
732
733 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
734out:
735 return status;
736out_noconflict:
737 fl->fl_type = F_UNLCK;
738 goto out;
739}
740
741static int do_vfs_lock(struct file *file, struct file_lock *fl)
742{
743 return locks_lock_file_wait(file, fl);
744}
745
746static int
747do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
748{
749 struct inode *inode = filp->f_mapping->host;
750 struct nfs_lock_context *l_ctx;
751 int status;
752
753 /*
754 * Flush all pending writes before doing anything
755 * with locks..
756 */
757 vfs_fsync(filp, 0);
758
759 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
760 if (!IS_ERR(l_ctx)) {
761 status = nfs_iocounter_wait(l_ctx);
762 nfs_put_lock_context(l_ctx);
763 if (status < 0)
764 return status;
765 }
766
767 /* NOTE: special case
768 * If we're signalled while cleaning up locks on process exit, we
769 * still need to complete the unlock.
770 */
771 /*
772 * Use local locking if mounted with "-onolock" or with appropriate
773 * "-olocal_lock="
774 */
775 if (!is_local)
776 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
777 else
778 status = do_vfs_lock(filp, fl);
779 return status;
780}
781
782static int
783is_time_granular(struct timespec *ts) {
784 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
785}
786
787static int
788do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
789{
790 struct inode *inode = filp->f_mapping->host;
791 int status;
792
793 /*
794 * Flush all pending writes before doing anything
795 * with locks..
796 */
797 status = nfs_sync_mapping(filp->f_mapping);
798 if (status != 0)
799 goto out;
800
801 /*
802 * Use local locking if mounted with "-onolock" or with appropriate
803 * "-olocal_lock="
804 */
805 if (!is_local)
806 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
807 else
808 status = do_vfs_lock(filp, fl);
809 if (status < 0)
810 goto out;
811
812 /*
813 * Revalidate the cache if the server has time stamps granular
814 * enough to detect subsecond changes. Otherwise, clear the
815 * cache to prevent missing any changes.
816 *
817 * This makes locking act as a cache coherency point.
818 */
819 nfs_sync_mapping(filp->f_mapping);
820 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
821 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
822 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
823 else
824 nfs_zap_caches(inode);
825 }
826out:
827 return status;
828}
829
830/*
831 * Lock a (portion of) a file
832 */
833int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
834{
835 struct inode *inode = filp->f_mapping->host;
836 int ret = -ENOLCK;
837 int is_local = 0;
838
839 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
840 filp, fl->fl_type, fl->fl_flags,
841 (long long)fl->fl_start, (long long)fl->fl_end);
842
843 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
844
845 /* No mandatory locks over NFS */
846 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
847 goto out_err;
848
849 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
850 is_local = 1;
851
852 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
853 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
854 if (ret < 0)
855 goto out_err;
856 }
857
858 if (IS_GETLK(cmd))
859 ret = do_getlk(filp, cmd, fl, is_local);
860 else if (fl->fl_type == F_UNLCK)
861 ret = do_unlk(filp, cmd, fl, is_local);
862 else
863 ret = do_setlk(filp, cmd, fl, is_local);
864out_err:
865 return ret;
866}
867EXPORT_SYMBOL_GPL(nfs_lock);
868
869/*
870 * Lock a (portion of) a file
871 */
872int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
873{
874 struct inode *inode = filp->f_mapping->host;
875 int is_local = 0;
876
877 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
878 filp, fl->fl_type, fl->fl_flags);
879
880 if (!(fl->fl_flags & FL_FLOCK))
881 return -ENOLCK;
882
883 /*
884 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
885 * any standard. In principle we might be able to support LOCK_MAND
886 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
887 * NFS code is not set up for it.
888 */
889 if (fl->fl_type & LOCK_MAND)
890 return -EINVAL;
891
892 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
893 is_local = 1;
894
895 /* We're simulating flock() locks using posix locks on the server */
896 if (fl->fl_type == F_UNLCK)
897 return do_unlk(filp, cmd, fl, is_local);
898 return do_setlk(filp, cmd, fl, is_local);
899}
900EXPORT_SYMBOL_GPL(nfs_flock);
901
902const struct file_operations nfs_file_operations = {
903 .llseek = nfs_file_llseek,
904 .read_iter = nfs_file_read,
905 .write_iter = nfs_file_write,
906 .mmap = nfs_file_mmap,
907 .open = nfs_file_open,
908 .flush = nfs_file_flush,
909 .release = nfs_file_release,
910 .fsync = nfs_file_fsync,
911 .lock = nfs_lock,
912 .flock = nfs_flock,
913 .splice_read = nfs_file_splice_read,
914 .splice_write = iter_file_splice_write,
915 .check_flags = nfs_check_flags,
916 .setlease = simple_nosetlease,
917};
918EXPORT_SYMBOL_GPL(nfs_file_operations);