Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * Changes Copyright (C) 1994 by Florian La Roche
8 * - Do not copy data too often around in the kernel.
9 * - In nfs_file_read the return value of kmalloc wasn't checked.
10 * - Put in a better version of read look-ahead buffering. Original idea
11 * and implementation by Wai S Kok elekokws@ee.nus.sg.
12 *
13 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 *
15 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 *
17 * nfs regular file handling functions
18 */
19
20#include <linux/module.h>
21#include <linux/time.h>
22#include <linux/kernel.h>
23#include <linux/errno.h>
24#include <linux/fcntl.h>
25#include <linux/stat.h>
26#include <linux/nfs_fs.h>
27#include <linux/nfs_mount.h>
28#include <linux/mm.h>
29#include <linux/pagemap.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32#include <linux/compaction.h>
33
34#include <linux/uaccess.h>
35#include <linux/filelock.h>
36
37#include "delegation.h"
38#include "internal.h"
39#include "iostat.h"
40#include "fscache.h"
41#include "pnfs.h"
42
43#include "nfstrace.h"
44
45#define NFSDBG_FACILITY NFSDBG_FILE
46
47static const struct vm_operations_struct nfs_file_vm_ops;
48
49int nfs_check_flags(int flags)
50{
51 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
52 return -EINVAL;
53
54 return 0;
55}
56EXPORT_SYMBOL_GPL(nfs_check_flags);
57
58/*
59 * Open file
60 */
61static int
62nfs_file_open(struct inode *inode, struct file *filp)
63{
64 int res;
65
66 dprintk("NFS: open file(%pD2)\n", filp);
67
68 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
69 res = nfs_check_flags(filp->f_flags);
70 if (res)
71 return res;
72
73 res = nfs_open(inode, filp);
74 if (res == 0)
75 filp->f_mode |= FMODE_CAN_ODIRECT;
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 nfs_fscache_release_file(inode, filp);
87 return 0;
88}
89EXPORT_SYMBOL_GPL(nfs_file_release);
90
91/**
92 * nfs_revalidate_file_size - Revalidate the file size
93 * @inode: pointer to inode struct
94 * @filp: pointer to struct file
95 *
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
101 */
102static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103{
104 struct nfs_server *server = NFS_SERVER(inode);
105
106 if (filp->f_flags & O_DIRECT)
107 goto force_reval;
108 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_SIZE))
109 goto force_reval;
110 return 0;
111force_reval:
112 return __nfs_revalidate_inode(server, inode);
113}
114
115loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
116{
117 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
118 filp, offset, whence);
119
120 /*
121 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
122 * the cached file length
123 */
124 if (whence != SEEK_SET && whence != SEEK_CUR) {
125 struct inode *inode = filp->f_mapping->host;
126
127 int retval = nfs_revalidate_file_size(inode, filp);
128 if (retval < 0)
129 return (loff_t)retval;
130 }
131
132 return generic_file_llseek(filp, offset, whence);
133}
134EXPORT_SYMBOL_GPL(nfs_file_llseek);
135
136/*
137 * Flush all dirty pages, and check for write errors.
138 */
139static int
140nfs_file_flush(struct file *file, fl_owner_t id)
141{
142 struct inode *inode = file_inode(file);
143 errseq_t since;
144
145 dprintk("NFS: flush(%pD2)\n", file);
146
147 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
148 if ((file->f_mode & FMODE_WRITE) == 0)
149 return 0;
150
151 /* Flush writes to the server and return any errors */
152 since = filemap_sample_wb_err(file->f_mapping);
153 nfs_wb_all(inode);
154 return filemap_check_wb_err(file->f_mapping, since);
155}
156
157ssize_t
158nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
159{
160 struct inode *inode = file_inode(iocb->ki_filp);
161 ssize_t result;
162
163 if (iocb->ki_flags & IOCB_DIRECT)
164 return nfs_file_direct_read(iocb, to, false);
165
166 dprintk("NFS: read(%pD2, %zu@%lu)\n",
167 iocb->ki_filp,
168 iov_iter_count(to), (unsigned long) iocb->ki_pos);
169
170 result = nfs_start_io_read(inode);
171 if (result)
172 return result;
173
174 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
175 if (!result) {
176 result = generic_file_read_iter(iocb, to);
177 if (result > 0)
178 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
179 }
180 nfs_end_io_read(inode);
181 return result;
182}
183EXPORT_SYMBOL_GPL(nfs_file_read);
184
185ssize_t
186nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
187 size_t len, unsigned int flags)
188{
189 struct inode *inode = file_inode(in);
190 ssize_t result;
191
192 dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
193
194 result = nfs_start_io_read(inode);
195 if (result)
196 return result;
197
198 result = nfs_revalidate_mapping(inode, in->f_mapping);
199 if (!result) {
200 result = filemap_splice_read(in, ppos, pipe, len, flags);
201 if (result > 0)
202 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
203 }
204 nfs_end_io_read(inode);
205 return result;
206}
207EXPORT_SYMBOL_GPL(nfs_file_splice_read);
208
209int
210nfs_file_mmap(struct file *file, struct vm_area_struct *vma)
211{
212 struct inode *inode = file_inode(file);
213 int status;
214
215 dprintk("NFS: mmap(%pD2)\n", file);
216
217 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
218 * so we call that before revalidating the mapping
219 */
220 status = generic_file_mmap(file, vma);
221 if (!status) {
222 vma->vm_ops = &nfs_file_vm_ops;
223 status = nfs_revalidate_mapping(inode, file->f_mapping);
224 }
225 return status;
226}
227EXPORT_SYMBOL_GPL(nfs_file_mmap);
228
229/*
230 * Flush any dirty pages for this process, and check for write errors.
231 * The return status from this call provides a reliable indication of
232 * whether any write errors occurred for this process.
233 */
234static int
235nfs_file_fsync_commit(struct file *file, int datasync)
236{
237 struct inode *inode = file_inode(file);
238 int ret, ret2;
239
240 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
241
242 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
243 ret = nfs_commit_inode(inode, FLUSH_SYNC);
244 ret2 = file_check_and_advance_wb_err(file);
245 if (ret2 < 0)
246 return ret2;
247 return ret;
248}
249
250int
251nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
252{
253 struct inode *inode = file_inode(file);
254 struct nfs_inode *nfsi = NFS_I(inode);
255 long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
256 long nredirtied;
257 int ret;
258
259 trace_nfs_fsync_enter(inode);
260
261 for (;;) {
262 ret = file_write_and_wait_range(file, start, end);
263 if (ret != 0)
264 break;
265 ret = nfs_file_fsync_commit(file, datasync);
266 if (ret != 0)
267 break;
268 ret = pnfs_sync_inode(inode, !!datasync);
269 if (ret != 0)
270 break;
271 nredirtied = atomic_long_read(&nfsi->redirtied_pages);
272 if (nredirtied == save_nredirtied)
273 break;
274 save_nredirtied = nredirtied;
275 }
276
277 trace_nfs_fsync_exit(inode, ret);
278 return ret;
279}
280EXPORT_SYMBOL_GPL(nfs_file_fsync);
281
282/*
283 * Decide whether a read/modify/write cycle may be more efficient
284 * then a modify/write/read cycle when writing to a page in the
285 * page cache.
286 *
287 * Some pNFS layout drivers can only read/write at a certain block
288 * granularity like all block devices and therefore we must perform
289 * read/modify/write whenever a page hasn't read yet and the data
290 * to be written there is not aligned to a block boundary and/or
291 * smaller than the block size.
292 *
293 * The modify/write/read cycle may occur if a page is read before
294 * being completely filled by the writer. In this situation, the
295 * page must be completely written to stable storage on the server
296 * before it can be refilled by reading in the page from the server.
297 * This can lead to expensive, small, FILE_SYNC mode writes being
298 * done.
299 *
300 * It may be more efficient to read the page first if the file is
301 * open for reading in addition to writing, the page is not marked
302 * as Uptodate, it is not dirty or waiting to be committed,
303 * indicating that it was previously allocated and then modified,
304 * that there were valid bytes of data in that range of the file,
305 * and that the new data won't completely replace the old data in
306 * that range of the file.
307 */
308static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
309 unsigned int len)
310{
311 unsigned int pglen = nfs_folio_length(folio);
312 unsigned int offset = offset_in_folio(folio, pos);
313 unsigned int end = offset + len;
314
315 return !pglen || (end >= pglen && !offset);
316}
317
318static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
319 loff_t pos, unsigned int len)
320{
321 /*
322 * Up-to-date pages, those with ongoing or full-page write
323 * don't need read/modify/write
324 */
325 if (folio_test_uptodate(folio) || folio_test_private(folio) ||
326 nfs_folio_is_full_write(folio, pos, len))
327 return false;
328
329 if (pnfs_ld_read_whole_page(file_inode(file)))
330 return true;
331 /* Open for reading too? */
332 if (file->f_mode & FMODE_READ)
333 return true;
334 return false;
335}
336
337/*
338 * This does the "real" work of the write. We must allocate and lock the
339 * page to be sent back to the generic routine, which then copies the
340 * data from user space.
341 *
342 * If the writer ends up delaying the write, the writer needs to
343 * increment the page use counts until he is done with the page.
344 */
345static int nfs_write_begin(struct file *file, struct address_space *mapping,
346 loff_t pos, unsigned len, struct folio **foliop,
347 void **fsdata)
348{
349 fgf_t fgp = FGP_WRITEBEGIN;
350 struct folio *folio;
351 int once_thru = 0;
352 int ret;
353
354 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
355 file, mapping->host->i_ino, len, (long long) pos);
356
357 fgp |= fgf_set_order(len);
358start:
359 folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp,
360 mapping_gfp_mask(mapping));
361 if (IS_ERR(folio))
362 return PTR_ERR(folio);
363 *foliop = folio;
364
365 ret = nfs_flush_incompatible(file, folio);
366 if (ret) {
367 folio_unlock(folio);
368 folio_put(folio);
369 } else if (!once_thru &&
370 nfs_want_read_modify_write(file, folio, pos, len)) {
371 once_thru = 1;
372 ret = nfs_read_folio(file, folio);
373 folio_put(folio);
374 if (!ret)
375 goto start;
376 }
377 return ret;
378}
379
380static int nfs_write_end(struct file *file, struct address_space *mapping,
381 loff_t pos, unsigned len, unsigned copied,
382 struct folio *folio, void *fsdata)
383{
384 struct nfs_open_context *ctx = nfs_file_open_context(file);
385 unsigned offset = offset_in_folio(folio, pos);
386 int status;
387
388 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
389 file, mapping->host->i_ino, len, (long long) pos);
390
391 /*
392 * Zero any uninitialised parts of the page, and then mark the page
393 * as up to date if it turns out that we're extending the file.
394 */
395 if (!folio_test_uptodate(folio)) {
396 size_t fsize = folio_size(folio);
397 unsigned pglen = nfs_folio_length(folio);
398 unsigned end = offset + copied;
399
400 if (pglen == 0) {
401 folio_zero_segments(folio, 0, offset, end, fsize);
402 folio_mark_uptodate(folio);
403 } else if (end >= pglen) {
404 folio_zero_segment(folio, end, fsize);
405 if (offset == 0)
406 folio_mark_uptodate(folio);
407 } else
408 folio_zero_segment(folio, pglen, fsize);
409 }
410
411 status = nfs_update_folio(file, folio, offset, copied);
412
413 folio_unlock(folio);
414 folio_put(folio);
415
416 if (status < 0)
417 return status;
418 NFS_I(mapping->host)->write_io += copied;
419
420 if (nfs_ctx_key_to_expire(ctx, mapping->host))
421 nfs_wb_all(mapping->host);
422
423 return copied;
424}
425
426/*
427 * Partially or wholly invalidate a page
428 * - Release the private state associated with a page if undergoing complete
429 * page invalidation
430 * - Called if either PG_private or PG_fscache is set on the page
431 * - Caller holds page lock
432 */
433static void nfs_invalidate_folio(struct folio *folio, size_t offset,
434 size_t length)
435{
436 struct inode *inode = folio->mapping->host;
437 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
438 folio->index, offset, length);
439
440 if (offset != 0 || length < folio_size(folio))
441 return;
442 /* Cancel any unstarted writes on this page */
443 nfs_wb_folio_cancel(inode, folio);
444 folio_wait_private_2(folio); /* [DEPRECATED] */
445 trace_nfs_invalidate_folio(inode, folio_pos(folio) + offset, length);
446}
447
448/*
449 * Attempt to release the private state associated with a folio
450 * - Called if either private or fscache flags are set on the folio
451 * - Caller holds folio lock
452 * - Return true (may release folio) or false (may not)
453 */
454static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
455{
456 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
457
458 /* If the private flag is set, then the folio is not freeable */
459 if (folio_test_private(folio)) {
460 if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
461 current_is_kswapd() || current_is_kcompactd())
462 return false;
463 if (nfs_wb_folio(folio->mapping->host, folio) < 0)
464 return false;
465 }
466 return nfs_fscache_release_folio(folio, gfp);
467}
468
469static void nfs_check_dirty_writeback(struct folio *folio,
470 bool *dirty, bool *writeback)
471{
472 struct nfs_inode *nfsi;
473 struct address_space *mapping = folio->mapping;
474
475 /*
476 * Check if an unstable folio is currently being committed and
477 * if so, have the VM treat it as if the folio is under writeback
478 * so it will not block due to folios that will shortly be freeable.
479 */
480 nfsi = NFS_I(mapping->host);
481 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
482 *writeback = true;
483 return;
484 }
485
486 /*
487 * If the private flag is set, then the folio is not freeable
488 * and as the inode is not being committed, it's not going to
489 * be cleaned in the near future so treat it as dirty
490 */
491 if (folio_test_private(folio))
492 *dirty = true;
493}
494
495/*
496 * Attempt to clear the private state associated with a page when an error
497 * occurs that requires the cached contents of an inode to be written back or
498 * destroyed
499 * - Called if either PG_private or fscache is set on the page
500 * - Caller holds page lock
501 * - Return 0 if successful, -error otherwise
502 */
503static int nfs_launder_folio(struct folio *folio)
504{
505 struct inode *inode = folio->mapping->host;
506 int ret;
507
508 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
509 inode->i_ino, folio_pos(folio));
510
511 folio_wait_private_2(folio); /* [DEPRECATED] */
512 ret = nfs_wb_folio(inode, folio);
513 trace_nfs_launder_folio_done(inode, folio_pos(folio),
514 folio_size(folio), ret);
515 return ret;
516}
517
518static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
519 sector_t *span)
520{
521 unsigned long blocks;
522 long long isize;
523 int ret;
524 struct inode *inode = file_inode(file);
525 struct rpc_clnt *clnt = NFS_CLIENT(inode);
526 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
527
528 spin_lock(&inode->i_lock);
529 blocks = inode->i_blocks;
530 isize = inode->i_size;
531 spin_unlock(&inode->i_lock);
532 if (blocks*512 < isize) {
533 pr_warn("swap activate: swapfile has holes\n");
534 return -EINVAL;
535 }
536
537 ret = rpc_clnt_swap_activate(clnt);
538 if (ret)
539 return ret;
540 ret = add_swap_extent(sis, 0, sis->max, 0);
541 if (ret < 0) {
542 rpc_clnt_swap_deactivate(clnt);
543 return ret;
544 }
545
546 *span = sis->pages;
547
548 if (cl->rpc_ops->enable_swap)
549 cl->rpc_ops->enable_swap(inode);
550
551 sis->flags |= SWP_FS_OPS;
552 return ret;
553}
554
555static void nfs_swap_deactivate(struct file *file)
556{
557 struct inode *inode = file_inode(file);
558 struct rpc_clnt *clnt = NFS_CLIENT(inode);
559 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
560
561 rpc_clnt_swap_deactivate(clnt);
562 if (cl->rpc_ops->disable_swap)
563 cl->rpc_ops->disable_swap(file_inode(file));
564}
565
566const struct address_space_operations nfs_file_aops = {
567 .read_folio = nfs_read_folio,
568 .readahead = nfs_readahead,
569 .dirty_folio = filemap_dirty_folio,
570 .writepages = nfs_writepages,
571 .write_begin = nfs_write_begin,
572 .write_end = nfs_write_end,
573 .invalidate_folio = nfs_invalidate_folio,
574 .release_folio = nfs_release_folio,
575 .migrate_folio = nfs_migrate_folio,
576 .launder_folio = nfs_launder_folio,
577 .is_dirty_writeback = nfs_check_dirty_writeback,
578 .error_remove_folio = generic_error_remove_folio,
579 .swap_activate = nfs_swap_activate,
580 .swap_deactivate = nfs_swap_deactivate,
581 .swap_rw = nfs_swap_rw,
582};
583
584/*
585 * Notification that a PTE pointing to an NFS page is about to be made
586 * writable, implying that someone is about to modify the page through a
587 * shared-writable mapping
588 */
589static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
590{
591 struct file *filp = vmf->vma->vm_file;
592 struct inode *inode = file_inode(filp);
593 unsigned pagelen;
594 vm_fault_t ret = VM_FAULT_NOPAGE;
595 struct address_space *mapping;
596 struct folio *folio = page_folio(vmf->page);
597
598 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
599 filp, filp->f_mapping->host->i_ino,
600 (long long)folio_pos(folio));
601
602 sb_start_pagefault(inode->i_sb);
603
604 /* make sure the cache has finished storing the page */
605 if (folio_test_private_2(folio) && /* [DEPRECATED] */
606 folio_wait_private_2_killable(folio) < 0) {
607 ret = VM_FAULT_RETRY;
608 goto out;
609 }
610
611 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
612 nfs_wait_bit_killable,
613 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
614
615 folio_lock(folio);
616 mapping = folio->mapping;
617 if (mapping != inode->i_mapping)
618 goto out_unlock;
619
620 folio_wait_writeback(folio);
621
622 pagelen = nfs_folio_length(folio);
623 if (pagelen == 0)
624 goto out_unlock;
625
626 ret = VM_FAULT_LOCKED;
627 if (nfs_flush_incompatible(filp, folio) == 0 &&
628 nfs_update_folio(filp, folio, 0, pagelen) == 0)
629 goto out;
630
631 ret = VM_FAULT_SIGBUS;
632out_unlock:
633 folio_unlock(folio);
634out:
635 sb_end_pagefault(inode->i_sb);
636 return ret;
637}
638
639static const struct vm_operations_struct nfs_file_vm_ops = {
640 .fault = filemap_fault,
641 .map_pages = filemap_map_pages,
642 .page_mkwrite = nfs_vm_page_mkwrite,
643};
644
645ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
646{
647 struct file *file = iocb->ki_filp;
648 struct inode *inode = file_inode(file);
649 unsigned int mntflags = NFS_SERVER(inode)->flags;
650 ssize_t result, written;
651 errseq_t since;
652 int error;
653
654 result = nfs_key_timeout_notify(file, inode);
655 if (result)
656 return result;
657
658 if (iocb->ki_flags & IOCB_DIRECT)
659 return nfs_file_direct_write(iocb, from, false);
660
661 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
662 file, iov_iter_count(from), (long long) iocb->ki_pos);
663
664 if (IS_SWAPFILE(inode))
665 goto out_swapfile;
666 /*
667 * O_APPEND implies that we must revalidate the file length.
668 */
669 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
670 result = nfs_revalidate_file_size(inode, file);
671 if (result)
672 return result;
673 }
674
675 nfs_clear_invalid_mapping(file->f_mapping);
676
677 since = filemap_sample_wb_err(file->f_mapping);
678 error = nfs_start_io_write(inode);
679 if (error)
680 return error;
681 result = generic_write_checks(iocb, from);
682 if (result > 0)
683 result = generic_perform_write(iocb, from);
684 nfs_end_io_write(inode);
685 if (result <= 0)
686 goto out;
687
688 written = result;
689 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
690
691 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
692 result = filemap_fdatawrite_range(file->f_mapping,
693 iocb->ki_pos - written,
694 iocb->ki_pos - 1);
695 if (result < 0)
696 goto out;
697 }
698 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
699 filemap_fdatawait_range(file->f_mapping,
700 iocb->ki_pos - written,
701 iocb->ki_pos - 1);
702 }
703 result = generic_write_sync(iocb, written);
704 if (result < 0)
705 return result;
706
707out:
708 /* Return error values */
709 error = filemap_check_wb_err(file->f_mapping, since);
710 switch (error) {
711 default:
712 break;
713 case -EDQUOT:
714 case -EFBIG:
715 case -ENOSPC:
716 nfs_wb_all(inode);
717 error = file_check_and_advance_wb_err(file);
718 if (error < 0)
719 result = error;
720 }
721 return result;
722
723out_swapfile:
724 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
725 return -ETXTBSY;
726}
727EXPORT_SYMBOL_GPL(nfs_file_write);
728
729static int
730do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
731{
732 struct inode *inode = filp->f_mapping->host;
733 int status = 0;
734 unsigned int saved_type = fl->c.flc_type;
735
736 /* Try local locking first */
737 posix_test_lock(filp, fl);
738 if (fl->c.flc_type != F_UNLCK) {
739 /* found a conflict */
740 goto out;
741 }
742 fl->c.flc_type = saved_type;
743
744 if (nfs_have_read_or_write_delegation(inode))
745 goto out_noconflict;
746
747 if (is_local)
748 goto out_noconflict;
749
750 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
751out:
752 return status;
753out_noconflict:
754 fl->c.flc_type = F_UNLCK;
755 goto out;
756}
757
758static int
759do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
760{
761 struct inode *inode = filp->f_mapping->host;
762 struct nfs_lock_context *l_ctx;
763 int status;
764
765 /*
766 * Flush all pending writes before doing anything
767 * with locks..
768 */
769 nfs_wb_all(inode);
770
771 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
772 if (!IS_ERR(l_ctx)) {
773 status = nfs_iocounter_wait(l_ctx);
774 nfs_put_lock_context(l_ctx);
775 /* NOTE: special case
776 * If we're signalled while cleaning up locks on process exit, we
777 * still need to complete the unlock.
778 */
779 if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
780 return status;
781 }
782
783 /*
784 * Use local locking if mounted with "-onolock" or with appropriate
785 * "-olocal_lock="
786 */
787 if (!is_local)
788 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
789 else
790 status = locks_lock_file_wait(filp, fl);
791 return status;
792}
793
794static int
795do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
796{
797 struct inode *inode = filp->f_mapping->host;
798 int status;
799
800 /*
801 * Flush all pending writes before doing anything
802 * with locks..
803 */
804 status = nfs_sync_mapping(filp->f_mapping);
805 if (status != 0)
806 goto out;
807
808 /*
809 * Use local locking if mounted with "-onolock" or with appropriate
810 * "-olocal_lock="
811 */
812 if (!is_local)
813 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
814 else
815 status = locks_lock_file_wait(filp, fl);
816 if (status < 0)
817 goto out;
818
819 /*
820 * Invalidate cache to prevent missing any changes. If
821 * the file is mapped, clear the page cache as well so
822 * those mappings will be loaded.
823 *
824 * This makes locking act as a cache coherency point.
825 */
826 nfs_sync_mapping(filp->f_mapping);
827 if (!nfs_have_read_or_write_delegation(inode)) {
828 nfs_zap_caches(inode);
829 if (mapping_mapped(filp->f_mapping))
830 nfs_revalidate_mapping(inode, filp->f_mapping);
831 }
832out:
833 return status;
834}
835
836/*
837 * Lock a (portion of) a file
838 */
839int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
840{
841 struct inode *inode = filp->f_mapping->host;
842 int ret = -ENOLCK;
843 int is_local = 0;
844
845 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
846 filp, fl->c.flc_type, fl->c.flc_flags,
847 (long long)fl->fl_start, (long long)fl->fl_end);
848
849 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
850
851 if (fl->c.flc_flags & FL_RECLAIM)
852 return -ENOGRACE;
853
854 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
855 is_local = 1;
856
857 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
858 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
859 if (ret < 0)
860 goto out_err;
861 }
862
863 if (IS_GETLK(cmd))
864 ret = do_getlk(filp, cmd, fl, is_local);
865 else if (lock_is_unlock(fl))
866 ret = do_unlk(filp, cmd, fl, is_local);
867 else
868 ret = do_setlk(filp, cmd, fl, is_local);
869out_err:
870 return ret;
871}
872EXPORT_SYMBOL_GPL(nfs_lock);
873
874/*
875 * Lock a (portion of) a file
876 */
877int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
878{
879 struct inode *inode = filp->f_mapping->host;
880 int is_local = 0;
881
882 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
883 filp, fl->c.flc_type, fl->c.flc_flags);
884
885 if (!(fl->c.flc_flags & FL_FLOCK))
886 return -ENOLCK;
887
888 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
889 is_local = 1;
890
891 /* We're simulating flock() locks using posix locks on the server */
892 if (lock_is_unlock(fl))
893 return do_unlk(filp, cmd, fl, is_local);
894 return do_setlk(filp, cmd, fl, is_local);
895}
896EXPORT_SYMBOL_GPL(nfs_flock);
897
898const struct file_operations nfs_file_operations = {
899 .llseek = nfs_file_llseek,
900 .read_iter = nfs_file_read,
901 .write_iter = nfs_file_write,
902 .mmap = nfs_file_mmap,
903 .open = nfs_file_open,
904 .flush = nfs_file_flush,
905 .release = nfs_file_release,
906 .fsync = nfs_file_fsync,
907 .lock = nfs_lock,
908 .flock = nfs_flock,
909 .splice_read = nfs_file_splice_read,
910 .splice_write = iter_file_splice_write,
911 .check_flags = nfs_check_flags,
912 .setlease = simple_nosetlease,
913};
914EXPORT_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/time.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/fcntl.h>
23#include <linux/stat.h>
24#include <linux/nfs_fs.h>
25#include <linux/nfs_mount.h>
26#include <linux/mm.h>
27#include <linux/pagemap.h>
28#include <linux/aio.h>
29#include <linux/gfp.h>
30#include <linux/swap.h>
31
32#include <asm/uaccess.h>
33#include <asm/system.h>
34
35#include "delegation.h"
36#include "internal.h"
37#include "iostat.h"
38#include "fscache.h"
39#include "pnfs.h"
40
41#define NFSDBG_FACILITY NFSDBG_FILE
42
43static int nfs_file_open(struct inode *, struct file *);
44static int nfs_file_release(struct inode *, struct file *);
45static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
46static int nfs_file_mmap(struct file *, struct vm_area_struct *);
47static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
48 struct pipe_inode_info *pipe,
49 size_t count, unsigned int flags);
50static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
51 unsigned long nr_segs, loff_t pos);
52static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
53 struct file *filp, loff_t *ppos,
54 size_t count, unsigned int flags);
55static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
56 unsigned long nr_segs, loff_t pos);
57static int nfs_file_flush(struct file *, fl_owner_t id);
58static int nfs_file_fsync(struct file *, loff_t, loff_t, int datasync);
59static int nfs_check_flags(int flags);
60static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
61static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
62static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
63
64static const struct vm_operations_struct nfs_file_vm_ops;
65
66const struct file_operations nfs_file_operations = {
67 .llseek = nfs_file_llseek,
68 .read = do_sync_read,
69 .write = do_sync_write,
70 .aio_read = nfs_file_read,
71 .aio_write = nfs_file_write,
72 .mmap = nfs_file_mmap,
73 .open = nfs_file_open,
74 .flush = nfs_file_flush,
75 .release = nfs_file_release,
76 .fsync = nfs_file_fsync,
77 .lock = nfs_lock,
78 .flock = nfs_flock,
79 .splice_read = nfs_file_splice_read,
80 .splice_write = nfs_file_splice_write,
81 .check_flags = nfs_check_flags,
82 .setlease = nfs_setlease,
83};
84
85const struct inode_operations nfs_file_inode_operations = {
86 .permission = nfs_permission,
87 .getattr = nfs_getattr,
88 .setattr = nfs_setattr,
89};
90
91#ifdef CONFIG_NFS_V3
92const struct inode_operations nfs3_file_inode_operations = {
93 .permission = nfs_permission,
94 .getattr = nfs_getattr,
95 .setattr = nfs_setattr,
96 .listxattr = nfs3_listxattr,
97 .getxattr = nfs3_getxattr,
98 .setxattr = nfs3_setxattr,
99 .removexattr = nfs3_removexattr,
100};
101#endif /* CONFIG_NFS_v3 */
102
103/* Hack for future NFS swap support */
104#ifndef IS_SWAPFILE
105# define IS_SWAPFILE(inode) (0)
106#endif
107
108static int nfs_check_flags(int flags)
109{
110 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
111 return -EINVAL;
112
113 return 0;
114}
115
116/*
117 * Open file
118 */
119static int
120nfs_file_open(struct inode *inode, struct file *filp)
121{
122 int res;
123
124 dprintk("NFS: open file(%s/%s)\n",
125 filp->f_path.dentry->d_parent->d_name.name,
126 filp->f_path.dentry->d_name.name);
127
128 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
129 res = nfs_check_flags(filp->f_flags);
130 if (res)
131 return res;
132
133 res = nfs_open(inode, filp);
134 return res;
135}
136
137static int
138nfs_file_release(struct inode *inode, struct file *filp)
139{
140 struct dentry *dentry = filp->f_path.dentry;
141
142 dprintk("NFS: release(%s/%s)\n",
143 dentry->d_parent->d_name.name,
144 dentry->d_name.name);
145
146 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
147 return nfs_release(inode, filp);
148}
149
150/**
151 * nfs_revalidate_size - Revalidate the file size
152 * @inode - pointer to inode struct
153 * @file - pointer to struct file
154 *
155 * Revalidates the file length. This is basically a wrapper around
156 * nfs_revalidate_inode() that takes into account the fact that we may
157 * have cached writes (in which case we don't care about the server's
158 * idea of what the file length is), or O_DIRECT (in which case we
159 * shouldn't trust the cache).
160 */
161static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
162{
163 struct nfs_server *server = NFS_SERVER(inode);
164 struct nfs_inode *nfsi = NFS_I(inode);
165
166 if (nfs_have_delegated_attributes(inode))
167 goto out_noreval;
168
169 if (filp->f_flags & O_DIRECT)
170 goto force_reval;
171 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
172 goto force_reval;
173 if (nfs_attribute_timeout(inode))
174 goto force_reval;
175out_noreval:
176 return 0;
177force_reval:
178 return __nfs_revalidate_inode(server, inode);
179}
180
181static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
182{
183 loff_t loff;
184
185 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
186 filp->f_path.dentry->d_parent->d_name.name,
187 filp->f_path.dentry->d_name.name,
188 offset, origin);
189
190 /*
191 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
192 * the cached file length
193 */
194 if (origin != SEEK_SET || origin != SEEK_CUR) {
195 struct inode *inode = filp->f_mapping->host;
196
197 int retval = nfs_revalidate_file_size(inode, filp);
198 if (retval < 0)
199 return (loff_t)retval;
200
201 spin_lock(&inode->i_lock);
202 loff = generic_file_llseek_unlocked(filp, offset, origin);
203 spin_unlock(&inode->i_lock);
204 } else
205 loff = generic_file_llseek_unlocked(filp, offset, origin);
206 return loff;
207}
208
209/*
210 * Flush all dirty pages, and check for write errors.
211 */
212static int
213nfs_file_flush(struct file *file, fl_owner_t id)
214{
215 struct dentry *dentry = file->f_path.dentry;
216 struct inode *inode = dentry->d_inode;
217
218 dprintk("NFS: flush(%s/%s)\n",
219 dentry->d_parent->d_name.name,
220 dentry->d_name.name);
221
222 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
223 if ((file->f_mode & FMODE_WRITE) == 0)
224 return 0;
225
226 /* Flush writes to the server and return any errors */
227 return vfs_fsync(file, 0);
228}
229
230static ssize_t
231nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
232 unsigned long nr_segs, loff_t pos)
233{
234 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
235 struct inode * inode = dentry->d_inode;
236 ssize_t result;
237 size_t count = iov_length(iov, nr_segs);
238
239 if (iocb->ki_filp->f_flags & O_DIRECT)
240 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
241
242 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
243 dentry->d_parent->d_name.name, dentry->d_name.name,
244 (unsigned long) count, (unsigned long) pos);
245
246 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
247 if (!result) {
248 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
249 if (result > 0)
250 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
251 }
252 return result;
253}
254
255static ssize_t
256nfs_file_splice_read(struct file *filp, loff_t *ppos,
257 struct pipe_inode_info *pipe, size_t count,
258 unsigned int flags)
259{
260 struct dentry *dentry = filp->f_path.dentry;
261 struct inode *inode = dentry->d_inode;
262 ssize_t res;
263
264 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
265 dentry->d_parent->d_name.name, dentry->d_name.name,
266 (unsigned long) count, (unsigned long long) *ppos);
267
268 res = nfs_revalidate_mapping(inode, filp->f_mapping);
269 if (!res) {
270 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
271 if (res > 0)
272 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
273 }
274 return res;
275}
276
277static int
278nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
279{
280 struct dentry *dentry = file->f_path.dentry;
281 struct inode *inode = dentry->d_inode;
282 int status;
283
284 dprintk("NFS: mmap(%s/%s)\n",
285 dentry->d_parent->d_name.name, dentry->d_name.name);
286
287 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
288 * so we call that before revalidating the mapping
289 */
290 status = generic_file_mmap(file, vma);
291 if (!status) {
292 vma->vm_ops = &nfs_file_vm_ops;
293 status = nfs_revalidate_mapping(inode, file->f_mapping);
294 }
295 return status;
296}
297
298/*
299 * Flush any dirty pages for this process, and check for write errors.
300 * The return status from this call provides a reliable indication of
301 * whether any write errors occurred for this process.
302 *
303 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
304 * disk, but it retrieves and clears ctx->error after synching, despite
305 * the two being set at the same time in nfs_context_set_write_error().
306 * This is because the former is used to notify the _next_ call to
307 * nfs_file_write() that a write error occurred, and hence cause it to
308 * fall back to doing a synchronous write.
309 */
310static int
311nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
312{
313 struct dentry *dentry = file->f_path.dentry;
314 struct nfs_open_context *ctx = nfs_file_open_context(file);
315 struct inode *inode = dentry->d_inode;
316 int have_error, status;
317 int ret = 0;
318
319 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
320 dentry->d_parent->d_name.name, dentry->d_name.name,
321 datasync);
322
323 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
324 if (ret)
325 return ret;
326 mutex_lock(&inode->i_mutex);
327
328 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
329 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
330 status = nfs_commit_inode(inode, FLUSH_SYNC);
331 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
332 if (have_error)
333 ret = xchg(&ctx->error, 0);
334 if (!ret && status < 0)
335 ret = status;
336 if (!ret && !datasync)
337 /* application has asked for meta-data sync */
338 ret = pnfs_layoutcommit_inode(inode, true);
339 mutex_unlock(&inode->i_mutex);
340 return ret;
341}
342
343/*
344 * Decide whether a read/modify/write cycle may be more efficient
345 * then a modify/write/read cycle when writing to a page in the
346 * page cache.
347 *
348 * The modify/write/read cycle may occur if a page is read before
349 * being completely filled by the writer. In this situation, the
350 * page must be completely written to stable storage on the server
351 * before it can be refilled by reading in the page from the server.
352 * This can lead to expensive, small, FILE_SYNC mode writes being
353 * done.
354 *
355 * It may be more efficient to read the page first if the file is
356 * open for reading in addition to writing, the page is not marked
357 * as Uptodate, it is not dirty or waiting to be committed,
358 * indicating that it was previously allocated and then modified,
359 * that there were valid bytes of data in that range of the file,
360 * and that the new data won't completely replace the old data in
361 * that range of the file.
362 */
363static int nfs_want_read_modify_write(struct file *file, struct page *page,
364 loff_t pos, unsigned len)
365{
366 unsigned int pglen = nfs_page_length(page);
367 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
368 unsigned int end = offset + len;
369
370 if ((file->f_mode & FMODE_READ) && /* open for read? */
371 !PageUptodate(page) && /* Uptodate? */
372 !PagePrivate(page) && /* i/o request already? */
373 pglen && /* valid bytes of file? */
374 (end < pglen || offset)) /* replace all valid bytes? */
375 return 1;
376 return 0;
377}
378
379/*
380 * This does the "real" work of the write. We must allocate and lock the
381 * page to be sent back to the generic routine, which then copies the
382 * data from user space.
383 *
384 * If the writer ends up delaying the write, the writer needs to
385 * increment the page use counts until he is done with the page.
386 */
387static int nfs_write_begin(struct file *file, struct address_space *mapping,
388 loff_t pos, unsigned len, unsigned flags,
389 struct page **pagep, void **fsdata)
390{
391 int ret;
392 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
393 struct page *page;
394 int once_thru = 0;
395
396 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
397 file->f_path.dentry->d_parent->d_name.name,
398 file->f_path.dentry->d_name.name,
399 mapping->host->i_ino, len, (long long) pos);
400
401start:
402 /*
403 * Prevent starvation issues if someone is doing a consistency
404 * sync-to-disk
405 */
406 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
407 nfs_wait_bit_killable, TASK_KILLABLE);
408 if (ret)
409 return ret;
410
411 page = grab_cache_page_write_begin(mapping, index, flags);
412 if (!page)
413 return -ENOMEM;
414 *pagep = page;
415
416 ret = nfs_flush_incompatible(file, page);
417 if (ret) {
418 unlock_page(page);
419 page_cache_release(page);
420 } else if (!once_thru &&
421 nfs_want_read_modify_write(file, page, pos, len)) {
422 once_thru = 1;
423 ret = nfs_readpage(file, page);
424 page_cache_release(page);
425 if (!ret)
426 goto start;
427 }
428 return ret;
429}
430
431static int nfs_write_end(struct file *file, struct address_space *mapping,
432 loff_t pos, unsigned len, unsigned copied,
433 struct page *page, void *fsdata)
434{
435 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
436 int status;
437
438 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
439 file->f_path.dentry->d_parent->d_name.name,
440 file->f_path.dentry->d_name.name,
441 mapping->host->i_ino, len, (long long) pos);
442
443 /*
444 * Zero any uninitialised parts of the page, and then mark the page
445 * as up to date if it turns out that we're extending the file.
446 */
447 if (!PageUptodate(page)) {
448 unsigned pglen = nfs_page_length(page);
449 unsigned end = offset + len;
450
451 if (pglen == 0) {
452 zero_user_segments(page, 0, offset,
453 end, PAGE_CACHE_SIZE);
454 SetPageUptodate(page);
455 } else if (end >= pglen) {
456 zero_user_segment(page, end, PAGE_CACHE_SIZE);
457 if (offset == 0)
458 SetPageUptodate(page);
459 } else
460 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
461 }
462
463 status = nfs_updatepage(file, page, offset, copied);
464
465 unlock_page(page);
466 page_cache_release(page);
467
468 if (status < 0)
469 return status;
470 return copied;
471}
472
473/*
474 * Partially or wholly invalidate a page
475 * - Release the private state associated with a page if undergoing complete
476 * page invalidation
477 * - Called if either PG_private or PG_fscache is set on the page
478 * - Caller holds page lock
479 */
480static void nfs_invalidate_page(struct page *page, unsigned long offset)
481{
482 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
483
484 if (offset != 0)
485 return;
486 /* Cancel any unstarted writes on this page */
487 nfs_wb_page_cancel(page->mapping->host, page);
488
489 nfs_fscache_invalidate_page(page, page->mapping->host);
490}
491
492/*
493 * Attempt to release the private state associated with a page
494 * - Called if either PG_private or PG_fscache is set on the page
495 * - Caller holds page lock
496 * - Return true (may release page) or false (may not)
497 */
498static int nfs_release_page(struct page *page, gfp_t gfp)
499{
500 struct address_space *mapping = page->mapping;
501
502 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
503
504 /* Only do I/O if gfp is a superset of GFP_KERNEL */
505 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) {
506 int how = FLUSH_SYNC;
507
508 /* Don't let kswapd deadlock waiting for OOM RPC calls */
509 if (current_is_kswapd())
510 how = 0;
511 nfs_commit_inode(mapping->host, how);
512 }
513 /* If PagePrivate() is set, then the page is not freeable */
514 if (PagePrivate(page))
515 return 0;
516 return nfs_fscache_release_page(page, gfp);
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->mapping->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
539const struct address_space_operations nfs_file_aops = {
540 .readpage = nfs_readpage,
541 .readpages = nfs_readpages,
542 .set_page_dirty = __set_page_dirty_nobuffers,
543 .writepage = nfs_writepage,
544 .writepages = nfs_writepages,
545 .write_begin = nfs_write_begin,
546 .write_end = nfs_write_end,
547 .invalidatepage = nfs_invalidate_page,
548 .releasepage = nfs_release_page,
549 .direct_IO = nfs_direct_IO,
550 .migratepage = nfs_migrate_page,
551 .launder_page = nfs_launder_page,
552 .error_remove_page = generic_error_remove_page,
553};
554
555/*
556 * Notification that a PTE pointing to an NFS page is about to be made
557 * writable, implying that someone is about to modify the page through a
558 * shared-writable mapping
559 */
560static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
561{
562 struct page *page = vmf->page;
563 struct file *filp = vma->vm_file;
564 struct dentry *dentry = filp->f_path.dentry;
565 unsigned pagelen;
566 int ret = VM_FAULT_NOPAGE;
567 struct address_space *mapping;
568
569 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
570 dentry->d_parent->d_name.name, dentry->d_name.name,
571 filp->f_mapping->host->i_ino,
572 (long long)page_offset(page));
573
574 /* make sure the cache has finished storing the page */
575 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
576
577 lock_page(page);
578 mapping = page->mapping;
579 if (mapping != dentry->d_inode->i_mapping)
580 goto out_unlock;
581
582 pagelen = nfs_page_length(page);
583 if (pagelen == 0)
584 goto out_unlock;
585
586 ret = VM_FAULT_LOCKED;
587 if (nfs_flush_incompatible(filp, page) == 0 &&
588 nfs_updatepage(filp, page, 0, pagelen) == 0)
589 goto out;
590
591 ret = VM_FAULT_SIGBUS;
592out_unlock:
593 unlock_page(page);
594out:
595 return ret;
596}
597
598static const struct vm_operations_struct nfs_file_vm_ops = {
599 .fault = filemap_fault,
600 .page_mkwrite = nfs_vm_page_mkwrite,
601};
602
603static int nfs_need_sync_write(struct file *filp, struct inode *inode)
604{
605 struct nfs_open_context *ctx;
606
607 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
608 return 1;
609 ctx = nfs_file_open_context(filp);
610 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
611 return 1;
612 return 0;
613}
614
615static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
616 unsigned long nr_segs, loff_t pos)
617{
618 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
619 struct inode * inode = dentry->d_inode;
620 unsigned long written = 0;
621 ssize_t result;
622 size_t count = iov_length(iov, nr_segs);
623
624 if (iocb->ki_filp->f_flags & O_DIRECT)
625 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
626
627 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
628 dentry->d_parent->d_name.name, dentry->d_name.name,
629 (unsigned long) count, (long long) pos);
630
631 result = -EBUSY;
632 if (IS_SWAPFILE(inode))
633 goto out_swapfile;
634 /*
635 * O_APPEND implies that we must revalidate the file length.
636 */
637 if (iocb->ki_filp->f_flags & O_APPEND) {
638 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
639 if (result)
640 goto out;
641 }
642
643 result = count;
644 if (!count)
645 goto out;
646
647 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
648 if (result > 0)
649 written = result;
650
651 /* Return error values for O_DSYNC and IS_SYNC() */
652 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
653 int err = vfs_fsync(iocb->ki_filp, 0);
654 if (err < 0)
655 result = err;
656 }
657 if (result > 0)
658 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
659out:
660 return result;
661
662out_swapfile:
663 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
664 goto out;
665}
666
667static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
668 struct file *filp, loff_t *ppos,
669 size_t count, unsigned int flags)
670{
671 struct dentry *dentry = filp->f_path.dentry;
672 struct inode *inode = dentry->d_inode;
673 unsigned long written = 0;
674 ssize_t ret;
675
676 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
677 dentry->d_parent->d_name.name, dentry->d_name.name,
678 (unsigned long) count, (unsigned long long) *ppos);
679
680 /*
681 * The combination of splice and an O_APPEND destination is disallowed.
682 */
683
684 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
685 if (ret > 0)
686 written = ret;
687
688 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
689 int err = vfs_fsync(filp, 0);
690 if (err < 0)
691 ret = err;
692 }
693 if (ret > 0)
694 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
695 return ret;
696}
697
698static int
699do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
700{
701 struct inode *inode = filp->f_mapping->host;
702 int status = 0;
703 unsigned int saved_type = fl->fl_type;
704
705 /* Try local locking first */
706 posix_test_lock(filp, fl);
707 if (fl->fl_type != F_UNLCK) {
708 /* found a conflict */
709 goto out;
710 }
711 fl->fl_type = saved_type;
712
713 if (nfs_have_delegation(inode, FMODE_READ))
714 goto out_noconflict;
715
716 if (is_local)
717 goto out_noconflict;
718
719 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
720out:
721 return status;
722out_noconflict:
723 fl->fl_type = F_UNLCK;
724 goto out;
725}
726
727static int do_vfs_lock(struct file *file, struct file_lock *fl)
728{
729 int res = 0;
730 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
731 case FL_POSIX:
732 res = posix_lock_file_wait(file, fl);
733 break;
734 case FL_FLOCK:
735 res = flock_lock_file_wait(file, fl);
736 break;
737 default:
738 BUG();
739 }
740 return res;
741}
742
743static int
744do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
745{
746 struct inode *inode = filp->f_mapping->host;
747 int status;
748
749 /*
750 * Flush all pending writes before doing anything
751 * with locks..
752 */
753 nfs_sync_mapping(filp->f_mapping);
754
755 /* NOTE: special case
756 * If we're signalled while cleaning up locks on process exit, we
757 * still need to complete the unlock.
758 */
759 /*
760 * Use local locking if mounted with "-onolock" or with appropriate
761 * "-olocal_lock="
762 */
763 if (!is_local)
764 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
765 else
766 status = do_vfs_lock(filp, fl);
767 return status;
768}
769
770static int
771is_time_granular(struct timespec *ts) {
772 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
773}
774
775static int
776do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
777{
778 struct inode *inode = filp->f_mapping->host;
779 int status;
780
781 /*
782 * Flush all pending writes before doing anything
783 * with locks..
784 */
785 status = nfs_sync_mapping(filp->f_mapping);
786 if (status != 0)
787 goto out;
788
789 /*
790 * Use local locking if mounted with "-onolock" or with appropriate
791 * "-olocal_lock="
792 */
793 if (!is_local)
794 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
795 else
796 status = do_vfs_lock(filp, fl);
797 if (status < 0)
798 goto out;
799
800 /*
801 * Revalidate the cache if the server has time stamps granular
802 * enough to detect subsecond changes. Otherwise, clear the
803 * cache to prevent missing any changes.
804 *
805 * This makes locking act as a cache coherency point.
806 */
807 nfs_sync_mapping(filp->f_mapping);
808 if (!nfs_have_delegation(inode, FMODE_READ)) {
809 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
810 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
811 else
812 nfs_zap_caches(inode);
813 }
814out:
815 return status;
816}
817
818/*
819 * Lock a (portion of) a file
820 */
821static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
822{
823 struct inode *inode = filp->f_mapping->host;
824 int ret = -ENOLCK;
825 int is_local = 0;
826
827 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
828 filp->f_path.dentry->d_parent->d_name.name,
829 filp->f_path.dentry->d_name.name,
830 fl->fl_type, fl->fl_flags,
831 (long long)fl->fl_start, (long long)fl->fl_end);
832
833 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
834
835 /* No mandatory locks over NFS */
836 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
837 goto out_err;
838
839 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
840 is_local = 1;
841
842 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
843 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
844 if (ret < 0)
845 goto out_err;
846 }
847
848 if (IS_GETLK(cmd))
849 ret = do_getlk(filp, cmd, fl, is_local);
850 else if (fl->fl_type == F_UNLCK)
851 ret = do_unlk(filp, cmd, fl, is_local);
852 else
853 ret = do_setlk(filp, cmd, fl, is_local);
854out_err:
855 return ret;
856}
857
858/*
859 * Lock a (portion of) a file
860 */
861static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
862{
863 struct inode *inode = filp->f_mapping->host;
864 int is_local = 0;
865
866 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
867 filp->f_path.dentry->d_parent->d_name.name,
868 filp->f_path.dentry->d_name.name,
869 fl->fl_type, fl->fl_flags);
870
871 if (!(fl->fl_flags & FL_FLOCK))
872 return -ENOLCK;
873
874 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
875 is_local = 1;
876
877 /* We're simulating flock() locks using posix locks on the server */
878 fl->fl_owner = (fl_owner_t)filp;
879 fl->fl_start = 0;
880 fl->fl_end = OFFSET_MAX;
881
882 if (fl->fl_type == F_UNLCK)
883 return do_unlk(filp, cmd, fl, is_local);
884 return do_setlk(filp, cmd, fl, is_local);
885}
886
887/*
888 * There is no protocol support for leases, so we have no way to implement
889 * them correctly in the face of opens by other clients.
890 */
891static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
892{
893 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
894 file->f_path.dentry->d_parent->d_name.name,
895 file->f_path.dentry->d_name.name, arg);
896 return -EINVAL;
897}