Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * Changes Copyright (C) 1994 by Florian La Roche
8 * - Do not copy data too often around in the kernel.
9 * - In nfs_file_read the return value of kmalloc wasn't checked.
10 * - Put in a better version of read look-ahead buffering. Original idea
11 * and implementation by Wai S Kok elekokws@ee.nus.sg.
12 *
13 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 *
15 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 *
17 * nfs regular file handling functions
18 */
19
20#include <linux/module.h>
21#include <linux/time.h>
22#include <linux/kernel.h>
23#include <linux/errno.h>
24#include <linux/fcntl.h>
25#include <linux/stat.h>
26#include <linux/nfs_fs.h>
27#include <linux/nfs_mount.h>
28#include <linux/mm.h>
29#include <linux/pagemap.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32
33#include <linux/uaccess.h>
34#include <linux/filelock.h>
35
36#include "delegation.h"
37#include "internal.h"
38#include "iostat.h"
39#include "fscache.h"
40#include "pnfs.h"
41
42#include "nfstrace.h"
43
44#define NFSDBG_FACILITY NFSDBG_FILE
45
46static const struct vm_operations_struct nfs_file_vm_ops;
47
48int nfs_check_flags(int flags)
49{
50 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
51 return -EINVAL;
52
53 return 0;
54}
55EXPORT_SYMBOL_GPL(nfs_check_flags);
56
57/*
58 * Open file
59 */
60static int
61nfs_file_open(struct inode *inode, struct file *filp)
62{
63 int res;
64
65 dprintk("NFS: open file(%pD2)\n", filp);
66
67 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
68 res = nfs_check_flags(filp->f_flags);
69 if (res)
70 return res;
71
72 res = nfs_open(inode, filp);
73 if (res == 0)
74 filp->f_mode |= FMODE_CAN_ODIRECT;
75 return res;
76}
77
78int
79nfs_file_release(struct inode *inode, struct file *filp)
80{
81 dprintk("NFS: release(%pD2)\n", filp);
82
83 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
84 nfs_file_clear_open_context(filp);
85 nfs_fscache_release_file(inode, filp);
86 return 0;
87}
88EXPORT_SYMBOL_GPL(nfs_file_release);
89
90/**
91 * nfs_revalidate_file_size - Revalidate the file size
92 * @inode: pointer to inode struct
93 * @filp: 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_INVALID_SIZE))
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 errseq_t since;
143
144 dprintk("NFS: flush(%pD2)\n", file);
145
146 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
147 if ((file->f_mode & FMODE_WRITE) == 0)
148 return 0;
149
150 /* Flush writes to the server and return any errors */
151 since = filemap_sample_wb_err(file->f_mapping);
152 nfs_wb_all(inode);
153 return filemap_check_wb_err(file->f_mapping, since);
154}
155
156ssize_t
157nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
158{
159 struct inode *inode = file_inode(iocb->ki_filp);
160 ssize_t result;
161
162 if (iocb->ki_flags & IOCB_DIRECT)
163 return nfs_file_direct_read(iocb, to, false);
164
165 dprintk("NFS: read(%pD2, %zu@%lu)\n",
166 iocb->ki_filp,
167 iov_iter_count(to), (unsigned long) iocb->ki_pos);
168
169 nfs_start_io_read(inode);
170 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
171 if (!result) {
172 result = generic_file_read_iter(iocb, to);
173 if (result > 0)
174 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
175 }
176 nfs_end_io_read(inode);
177 return result;
178}
179EXPORT_SYMBOL_GPL(nfs_file_read);
180
181ssize_t
182nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
183 size_t len, unsigned int flags)
184{
185 struct inode *inode = file_inode(in);
186 ssize_t result;
187
188 dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
189
190 nfs_start_io_read(inode);
191 result = nfs_revalidate_mapping(inode, in->f_mapping);
192 if (!result) {
193 result = filemap_splice_read(in, ppos, pipe, len, flags);
194 if (result > 0)
195 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
196 }
197 nfs_end_io_read(inode);
198 return result;
199}
200EXPORT_SYMBOL_GPL(nfs_file_splice_read);
201
202int
203nfs_file_mmap(struct file *file, struct vm_area_struct *vma)
204{
205 struct inode *inode = file_inode(file);
206 int status;
207
208 dprintk("NFS: mmap(%pD2)\n", file);
209
210 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
211 * so we call that before revalidating the mapping
212 */
213 status = generic_file_mmap(file, vma);
214 if (!status) {
215 vma->vm_ops = &nfs_file_vm_ops;
216 status = nfs_revalidate_mapping(inode, file->f_mapping);
217 }
218 return status;
219}
220EXPORT_SYMBOL_GPL(nfs_file_mmap);
221
222/*
223 * Flush any dirty pages for this process, and check for write errors.
224 * The return status from this call provides a reliable indication of
225 * whether any write errors occurred for this process.
226 */
227static int
228nfs_file_fsync_commit(struct file *file, int datasync)
229{
230 struct inode *inode = file_inode(file);
231 int ret, ret2;
232
233 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
234
235 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
236 ret = nfs_commit_inode(inode, FLUSH_SYNC);
237 ret2 = file_check_and_advance_wb_err(file);
238 if (ret2 < 0)
239 return ret2;
240 return ret;
241}
242
243int
244nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
245{
246 struct inode *inode = file_inode(file);
247 struct nfs_inode *nfsi = NFS_I(inode);
248 long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
249 long nredirtied;
250 int ret;
251
252 trace_nfs_fsync_enter(inode);
253
254 for (;;) {
255 ret = file_write_and_wait_range(file, start, end);
256 if (ret != 0)
257 break;
258 ret = nfs_file_fsync_commit(file, datasync);
259 if (ret != 0)
260 break;
261 ret = pnfs_sync_inode(inode, !!datasync);
262 if (ret != 0)
263 break;
264 nredirtied = atomic_long_read(&nfsi->redirtied_pages);
265 if (nredirtied == save_nredirtied)
266 break;
267 save_nredirtied = nredirtied;
268 }
269
270 trace_nfs_fsync_exit(inode, ret);
271 return ret;
272}
273EXPORT_SYMBOL_GPL(nfs_file_fsync);
274
275/*
276 * Decide whether a read/modify/write cycle may be more efficient
277 * then a modify/write/read cycle when writing to a page in the
278 * page cache.
279 *
280 * Some pNFS layout drivers can only read/write at a certain block
281 * granularity like all block devices and therefore we must perform
282 * read/modify/write whenever a page hasn't read yet and the data
283 * to be written there is not aligned to a block boundary and/or
284 * smaller than the block size.
285 *
286 * The modify/write/read cycle may occur if a page is read before
287 * being completely filled by the writer. In this situation, the
288 * page must be completely written to stable storage on the server
289 * before it can be refilled by reading in the page from the server.
290 * This can lead to expensive, small, FILE_SYNC mode writes being
291 * done.
292 *
293 * It may be more efficient to read the page first if the file is
294 * open for reading in addition to writing, the page is not marked
295 * as Uptodate, it is not dirty or waiting to be committed,
296 * indicating that it was previously allocated and then modified,
297 * that there were valid bytes of data in that range of the file,
298 * and that the new data won't completely replace the old data in
299 * that range of the file.
300 */
301static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
302 unsigned int len)
303{
304 unsigned int pglen = nfs_folio_length(folio);
305 unsigned int offset = offset_in_folio(folio, pos);
306 unsigned int end = offset + len;
307
308 return !pglen || (end >= pglen && !offset);
309}
310
311static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
312 loff_t pos, unsigned int len)
313{
314 /*
315 * Up-to-date pages, those with ongoing or full-page write
316 * don't need read/modify/write
317 */
318 if (folio_test_uptodate(folio) || folio_test_private(folio) ||
319 nfs_folio_is_full_write(folio, pos, len))
320 return false;
321
322 if (pnfs_ld_read_whole_page(file_inode(file)))
323 return true;
324 /* Open for reading too? */
325 if (file->f_mode & FMODE_READ)
326 return true;
327 return false;
328}
329
330/*
331 * This does the "real" work of the write. We must allocate and lock the
332 * page to be sent back to the generic routine, which then copies the
333 * data from user space.
334 *
335 * If the writer ends up delaying the write, the writer needs to
336 * increment the page use counts until he is done with the page.
337 */
338static int nfs_write_begin(struct file *file, struct address_space *mapping,
339 loff_t pos, unsigned len, struct page **pagep,
340 void **fsdata)
341{
342 struct folio *folio;
343 int once_thru = 0;
344 int ret;
345
346 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
347 file, mapping->host->i_ino, len, (long long) pos);
348
349start:
350 folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, FGP_WRITEBEGIN,
351 mapping_gfp_mask(mapping));
352 if (IS_ERR(folio))
353 return PTR_ERR(folio);
354 *pagep = &folio->page;
355
356 ret = nfs_flush_incompatible(file, folio);
357 if (ret) {
358 folio_unlock(folio);
359 folio_put(folio);
360 } else if (!once_thru &&
361 nfs_want_read_modify_write(file, folio, pos, len)) {
362 once_thru = 1;
363 ret = nfs_read_folio(file, folio);
364 folio_put(folio);
365 if (!ret)
366 goto start;
367 }
368 return ret;
369}
370
371static int nfs_write_end(struct file *file, struct address_space *mapping,
372 loff_t pos, unsigned len, unsigned copied,
373 struct page *page, void *fsdata)
374{
375 struct nfs_open_context *ctx = nfs_file_open_context(file);
376 struct folio *folio = page_folio(page);
377 unsigned offset = offset_in_folio(folio, pos);
378 int status;
379
380 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
381 file, mapping->host->i_ino, len, (long long) pos);
382
383 /*
384 * Zero any uninitialised parts of the page, and then mark the page
385 * as up to date if it turns out that we're extending the file.
386 */
387 if (!folio_test_uptodate(folio)) {
388 size_t fsize = folio_size(folio);
389 unsigned pglen = nfs_folio_length(folio);
390 unsigned end = offset + copied;
391
392 if (pglen == 0) {
393 folio_zero_segments(folio, 0, offset, end, fsize);
394 folio_mark_uptodate(folio);
395 } else if (end >= pglen) {
396 folio_zero_segment(folio, end, fsize);
397 if (offset == 0)
398 folio_mark_uptodate(folio);
399 } else
400 folio_zero_segment(folio, pglen, fsize);
401 }
402
403 status = nfs_update_folio(file, folio, offset, copied);
404
405 folio_unlock(folio);
406 folio_put(folio);
407
408 if (status < 0)
409 return status;
410 NFS_I(mapping->host)->write_io += copied;
411
412 if (nfs_ctx_key_to_expire(ctx, mapping->host))
413 nfs_wb_all(mapping->host);
414
415 return copied;
416}
417
418/*
419 * Partially or wholly invalidate a page
420 * - Release the private state associated with a page if undergoing complete
421 * page invalidation
422 * - Called if either PG_private or PG_fscache is set on the page
423 * - Caller holds page lock
424 */
425static void nfs_invalidate_folio(struct folio *folio, size_t offset,
426 size_t length)
427{
428 struct inode *inode = folio_file_mapping(folio)->host;
429 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
430 folio->index, offset, length);
431
432 if (offset != 0 || length < folio_size(folio))
433 return;
434 /* Cancel any unstarted writes on this page */
435 nfs_wb_folio_cancel(inode, folio);
436 folio_wait_fscache(folio);
437 trace_nfs_invalidate_folio(inode, folio);
438}
439
440/*
441 * Attempt to release the private state associated with a folio
442 * - Called if either private or fscache flags are set on the folio
443 * - Caller holds folio lock
444 * - Return true (may release folio) or false (may not)
445 */
446static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
447{
448 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
449
450 /* If the private flag is set, then the folio is not freeable */
451 if (folio_test_private(folio)) {
452 if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
453 current_is_kswapd())
454 return false;
455 if (nfs_wb_folio(folio_file_mapping(folio)->host, folio) < 0)
456 return false;
457 }
458 return nfs_fscache_release_folio(folio, gfp);
459}
460
461static void nfs_check_dirty_writeback(struct folio *folio,
462 bool *dirty, bool *writeback)
463{
464 struct nfs_inode *nfsi;
465 struct address_space *mapping = folio->mapping;
466
467 /*
468 * Check if an unstable folio is currently being committed and
469 * if so, have the VM treat it as if the folio is under writeback
470 * so it will not block due to folios that will shortly be freeable.
471 */
472 nfsi = NFS_I(mapping->host);
473 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
474 *writeback = true;
475 return;
476 }
477
478 /*
479 * If the private flag is set, then the folio is not freeable
480 * and as the inode is not being committed, it's not going to
481 * be cleaned in the near future so treat it as dirty
482 */
483 if (folio_test_private(folio))
484 *dirty = true;
485}
486
487/*
488 * Attempt to clear the private state associated with a page when an error
489 * occurs that requires the cached contents of an inode to be written back or
490 * destroyed
491 * - Called if either PG_private or fscache is set on the page
492 * - Caller holds page lock
493 * - Return 0 if successful, -error otherwise
494 */
495static int nfs_launder_folio(struct folio *folio)
496{
497 struct inode *inode = folio->mapping->host;
498 int ret;
499
500 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
501 inode->i_ino, folio_pos(folio));
502
503 folio_wait_fscache(folio);
504 ret = nfs_wb_folio(inode, folio);
505 trace_nfs_launder_folio_done(inode, folio, ret);
506 return ret;
507}
508
509static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
510 sector_t *span)
511{
512 unsigned long blocks;
513 long long isize;
514 int ret;
515 struct inode *inode = file_inode(file);
516 struct rpc_clnt *clnt = NFS_CLIENT(inode);
517 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
518
519 spin_lock(&inode->i_lock);
520 blocks = inode->i_blocks;
521 isize = inode->i_size;
522 spin_unlock(&inode->i_lock);
523 if (blocks*512 < isize) {
524 pr_warn("swap activate: swapfile has holes\n");
525 return -EINVAL;
526 }
527
528 ret = rpc_clnt_swap_activate(clnt);
529 if (ret)
530 return ret;
531 ret = add_swap_extent(sis, 0, sis->max, 0);
532 if (ret < 0) {
533 rpc_clnt_swap_deactivate(clnt);
534 return ret;
535 }
536
537 *span = sis->pages;
538
539 if (cl->rpc_ops->enable_swap)
540 cl->rpc_ops->enable_swap(inode);
541
542 sis->flags |= SWP_FS_OPS;
543 return ret;
544}
545
546static void nfs_swap_deactivate(struct file *file)
547{
548 struct inode *inode = file_inode(file);
549 struct rpc_clnt *clnt = NFS_CLIENT(inode);
550 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
551
552 rpc_clnt_swap_deactivate(clnt);
553 if (cl->rpc_ops->disable_swap)
554 cl->rpc_ops->disable_swap(file_inode(file));
555}
556
557const struct address_space_operations nfs_file_aops = {
558 .read_folio = nfs_read_folio,
559 .readahead = nfs_readahead,
560 .dirty_folio = filemap_dirty_folio,
561 .writepages = nfs_writepages,
562 .write_begin = nfs_write_begin,
563 .write_end = nfs_write_end,
564 .invalidate_folio = nfs_invalidate_folio,
565 .release_folio = nfs_release_folio,
566 .migrate_folio = nfs_migrate_folio,
567 .launder_folio = nfs_launder_folio,
568 .is_dirty_writeback = nfs_check_dirty_writeback,
569 .error_remove_folio = generic_error_remove_folio,
570 .swap_activate = nfs_swap_activate,
571 .swap_deactivate = nfs_swap_deactivate,
572 .swap_rw = nfs_swap_rw,
573};
574
575/*
576 * Notification that a PTE pointing to an NFS page is about to be made
577 * writable, implying that someone is about to modify the page through a
578 * shared-writable mapping
579 */
580static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
581{
582 struct file *filp = vmf->vma->vm_file;
583 struct inode *inode = file_inode(filp);
584 unsigned pagelen;
585 vm_fault_t ret = VM_FAULT_NOPAGE;
586 struct address_space *mapping;
587 struct folio *folio = page_folio(vmf->page);
588
589 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
590 filp, filp->f_mapping->host->i_ino,
591 (long long)folio_file_pos(folio));
592
593 sb_start_pagefault(inode->i_sb);
594
595 /* make sure the cache has finished storing the page */
596 if (folio_test_fscache(folio) &&
597 folio_wait_fscache_killable(folio) < 0) {
598 ret = VM_FAULT_RETRY;
599 goto out;
600 }
601
602 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
603 nfs_wait_bit_killable,
604 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
605
606 folio_lock(folio);
607 mapping = folio_file_mapping(folio);
608 if (mapping != inode->i_mapping)
609 goto out_unlock;
610
611 folio_wait_writeback(folio);
612
613 pagelen = nfs_folio_length(folio);
614 if (pagelen == 0)
615 goto out_unlock;
616
617 ret = VM_FAULT_LOCKED;
618 if (nfs_flush_incompatible(filp, folio) == 0 &&
619 nfs_update_folio(filp, folio, 0, pagelen) == 0)
620 goto out;
621
622 ret = VM_FAULT_SIGBUS;
623out_unlock:
624 folio_unlock(folio);
625out:
626 sb_end_pagefault(inode->i_sb);
627 return ret;
628}
629
630static const struct vm_operations_struct nfs_file_vm_ops = {
631 .fault = filemap_fault,
632 .map_pages = filemap_map_pages,
633 .page_mkwrite = nfs_vm_page_mkwrite,
634};
635
636ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
637{
638 struct file *file = iocb->ki_filp;
639 struct inode *inode = file_inode(file);
640 unsigned int mntflags = NFS_SERVER(inode)->flags;
641 ssize_t result, written;
642 errseq_t since;
643 int error;
644
645 result = nfs_key_timeout_notify(file, inode);
646 if (result)
647 return result;
648
649 if (iocb->ki_flags & IOCB_DIRECT)
650 return nfs_file_direct_write(iocb, from, false);
651
652 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
653 file, iov_iter_count(from), (long long) iocb->ki_pos);
654
655 if (IS_SWAPFILE(inode))
656 goto out_swapfile;
657 /*
658 * O_APPEND implies that we must revalidate the file length.
659 */
660 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
661 result = nfs_revalidate_file_size(inode, file);
662 if (result)
663 return result;
664 }
665
666 nfs_clear_invalid_mapping(file->f_mapping);
667
668 since = filemap_sample_wb_err(file->f_mapping);
669 nfs_start_io_write(inode);
670 result = generic_write_checks(iocb, from);
671 if (result > 0)
672 result = generic_perform_write(iocb, from);
673 nfs_end_io_write(inode);
674 if (result <= 0)
675 goto out;
676
677 written = result;
678 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
679
680 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
681 result = filemap_fdatawrite_range(file->f_mapping,
682 iocb->ki_pos - written,
683 iocb->ki_pos - 1);
684 if (result < 0)
685 goto out;
686 }
687 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
688 filemap_fdatawait_range(file->f_mapping,
689 iocb->ki_pos - written,
690 iocb->ki_pos - 1);
691 }
692 result = generic_write_sync(iocb, written);
693 if (result < 0)
694 return result;
695
696out:
697 /* Return error values */
698 error = filemap_check_wb_err(file->f_mapping, since);
699 switch (error) {
700 default:
701 break;
702 case -EDQUOT:
703 case -EFBIG:
704 case -ENOSPC:
705 nfs_wb_all(inode);
706 error = file_check_and_advance_wb_err(file);
707 if (error < 0)
708 result = error;
709 }
710 return result;
711
712out_swapfile:
713 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
714 return -ETXTBSY;
715}
716EXPORT_SYMBOL_GPL(nfs_file_write);
717
718static int
719do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
720{
721 struct inode *inode = filp->f_mapping->host;
722 int status = 0;
723 unsigned int saved_type = fl->fl_type;
724
725 /* Try local locking first */
726 posix_test_lock(filp, fl);
727 if (fl->fl_type != F_UNLCK) {
728 /* found a conflict */
729 goto out;
730 }
731 fl->fl_type = saved_type;
732
733 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
734 goto out_noconflict;
735
736 if (is_local)
737 goto out_noconflict;
738
739 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
740out:
741 return status;
742out_noconflict:
743 fl->fl_type = F_UNLCK;
744 goto out;
745}
746
747static int
748do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
749{
750 struct inode *inode = filp->f_mapping->host;
751 struct nfs_lock_context *l_ctx;
752 int status;
753
754 /*
755 * Flush all pending writes before doing anything
756 * with locks..
757 */
758 nfs_wb_all(inode);
759
760 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
761 if (!IS_ERR(l_ctx)) {
762 status = nfs_iocounter_wait(l_ctx);
763 nfs_put_lock_context(l_ctx);
764 /* NOTE: special case
765 * If we're signalled while cleaning up locks on process exit, we
766 * still need to complete the unlock.
767 */
768 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
769 return status;
770 }
771
772 /*
773 * Use local locking if mounted with "-onolock" or with appropriate
774 * "-olocal_lock="
775 */
776 if (!is_local)
777 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
778 else
779 status = locks_lock_file_wait(filp, fl);
780 return status;
781}
782
783static int
784do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
785{
786 struct inode *inode = filp->f_mapping->host;
787 int status;
788
789 /*
790 * Flush all pending writes before doing anything
791 * with locks..
792 */
793 status = nfs_sync_mapping(filp->f_mapping);
794 if (status != 0)
795 goto out;
796
797 /*
798 * Use local locking if mounted with "-onolock" or with appropriate
799 * "-olocal_lock="
800 */
801 if (!is_local)
802 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
803 else
804 status = locks_lock_file_wait(filp, fl);
805 if (status < 0)
806 goto out;
807
808 /*
809 * Invalidate cache to prevent missing any changes. If
810 * the file is mapped, clear the page cache as well so
811 * those mappings will be loaded.
812 *
813 * This makes locking act as a cache coherency point.
814 */
815 nfs_sync_mapping(filp->f_mapping);
816 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
817 nfs_zap_caches(inode);
818 if (mapping_mapped(filp->f_mapping))
819 nfs_revalidate_mapping(inode, filp->f_mapping);
820 }
821out:
822 return status;
823}
824
825/*
826 * Lock a (portion of) a file
827 */
828int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
829{
830 struct inode *inode = filp->f_mapping->host;
831 int ret = -ENOLCK;
832 int is_local = 0;
833
834 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
835 filp, fl->fl_type, fl->fl_flags,
836 (long long)fl->fl_start, (long long)fl->fl_end);
837
838 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
839
840 if (fl->fl_flags & FL_RECLAIM)
841 return -ENOGRACE;
842
843 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
844 is_local = 1;
845
846 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
847 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
848 if (ret < 0)
849 goto out_err;
850 }
851
852 if (IS_GETLK(cmd))
853 ret = do_getlk(filp, cmd, fl, is_local);
854 else if (fl->fl_type == F_UNLCK)
855 ret = do_unlk(filp, cmd, fl, is_local);
856 else
857 ret = do_setlk(filp, cmd, fl, is_local);
858out_err:
859 return ret;
860}
861EXPORT_SYMBOL_GPL(nfs_lock);
862
863/*
864 * Lock a (portion of) a file
865 */
866int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
867{
868 struct inode *inode = filp->f_mapping->host;
869 int is_local = 0;
870
871 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
872 filp, fl->fl_type, fl->fl_flags);
873
874 if (!(fl->fl_flags & FL_FLOCK))
875 return -ENOLCK;
876
877 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
878 is_local = 1;
879
880 /* We're simulating flock() locks using posix locks on the server */
881 if (fl->fl_type == F_UNLCK)
882 return do_unlk(filp, cmd, fl, is_local);
883 return do_setlk(filp, cmd, fl, is_local);
884}
885EXPORT_SYMBOL_GPL(nfs_flock);
886
887const struct file_operations nfs_file_operations = {
888 .llseek = nfs_file_llseek,
889 .read_iter = nfs_file_read,
890 .write_iter = nfs_file_write,
891 .mmap = nfs_file_mmap,
892 .open = nfs_file_open,
893 .flush = nfs_file_flush,
894 .release = nfs_file_release,
895 .fsync = nfs_file_fsync,
896 .lock = nfs_lock,
897 .flock = nfs_flock,
898 .splice_read = nfs_file_splice_read,
899 .splice_write = iter_file_splice_write,
900 .check_flags = nfs_check_flags,
901 .setlease = simple_nosetlease,
902};
903EXPORT_SYMBOL_GPL(nfs_file_operations);
1/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
19#include <linux/module.h>
20#include <linux/time.h>
21#include <linux/kernel.h>
22#include <linux/errno.h>
23#include <linux/fcntl.h>
24#include <linux/stat.h>
25#include <linux/nfs_fs.h>
26#include <linux/nfs_mount.h>
27#include <linux/mm.h>
28#include <linux/pagemap.h>
29#include <linux/aio.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32
33#include <asm/uaccess.h>
34
35#include "delegation.h"
36#include "internal.h"
37#include "iostat.h"
38#include "fscache.h"
39
40#include "nfstrace.h"
41
42#define NFSDBG_FACILITY NFSDBG_FILE
43
44static const struct vm_operations_struct nfs_file_vm_ops;
45
46/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
51int nfs_check_flags(int flags)
52{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
58EXPORT_SYMBOL_GPL(nfs_check_flags);
59
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
66 int res;
67
68 dprintk("NFS: open file(%pD2)\n", filp);
69
70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
75 res = nfs_open(inode, filp);
76 return res;
77}
78
79int
80nfs_file_release(struct inode *inode, struct file *filp)
81{
82 dprintk("NFS: release(%pD2)\n", filp);
83
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 return nfs_release(inode, filp);
86}
87EXPORT_SYMBOL_GPL(nfs_file_release);
88
89/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
121{
122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
124
125 /*
126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
127 * the cached file length
128 */
129 if (whence != SEEK_SET && whence != SEEK_CUR) {
130 struct inode *inode = filp->f_mapping->host;
131
132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
135 }
136
137 return generic_file_llseek(filp, offset, whence);
138}
139EXPORT_SYMBOL_GPL(nfs_file_llseek);
140
141/*
142 * Flush all dirty pages, and check for write errors.
143 */
144int
145nfs_file_flush(struct file *file, fl_owner_t id)
146{
147 struct inode *inode = file_inode(file);
148
149 dprintk("NFS: flush(%pD2)\n", file);
150
151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
154
155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
160 return filemap_fdatawrite(file->f_mapping);
161
162 /* Flush writes to the server and return any errors */
163 return vfs_fsync(file, 0);
164}
165EXPORT_SYMBOL_GPL(nfs_file_flush);
166
167ssize_t
168nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
170{
171 struct inode *inode = file_inode(iocb->ki_filp);
172 ssize_t result;
173
174 if (iocb->ki_filp->f_flags & O_DIRECT)
175 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
176
177 dprintk("NFS: read(%pD2, %lu@%lu)\n",
178 iocb->ki_filp,
179 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
180
181 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
182 if (!result) {
183 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
184 if (result > 0)
185 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 }
187 return result;
188}
189EXPORT_SYMBOL_GPL(nfs_file_read);
190
191ssize_t
192nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 struct pipe_inode_info *pipe, size_t count,
194 unsigned int flags)
195{
196 struct inode *inode = file_inode(filp);
197 ssize_t res;
198
199 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 filp, (unsigned long) count, (unsigned long long) *ppos);
201
202 res = nfs_revalidate_mapping(inode, filp->f_mapping);
203 if (!res) {
204 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
205 if (res > 0)
206 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 }
208 return res;
209}
210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
211
212int
213nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214{
215 struct inode *inode = file_inode(file);
216 int status;
217
218 dprintk("NFS: mmap(%pD2)\n", file);
219
220 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 * so we call that before revalidating the mapping
222 */
223 status = generic_file_mmap(file, vma);
224 if (!status) {
225 vma->vm_ops = &nfs_file_vm_ops;
226 status = nfs_revalidate_mapping(inode, file->f_mapping);
227 }
228 return status;
229}
230EXPORT_SYMBOL_GPL(nfs_file_mmap);
231
232/*
233 * Flush any dirty pages for this process, and check for write errors.
234 * The return status from this call provides a reliable indication of
235 * whether any write errors occurred for this process.
236 *
237 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238 * disk, but it retrieves and clears ctx->error after synching, despite
239 * the two being set at the same time in nfs_context_set_write_error().
240 * This is because the former is used to notify the _next_ call to
241 * nfs_file_write() that a write error occurred, and hence cause it to
242 * fall back to doing a synchronous write.
243 */
244int
245nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
246{
247 struct nfs_open_context *ctx = nfs_file_open_context(file);
248 struct inode *inode = file_inode(file);
249 int have_error, do_resend, status;
250 int ret = 0;
251
252 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
253
254 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
255 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
256 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 status = nfs_commit_inode(inode, FLUSH_SYNC);
258 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
259 if (have_error) {
260 ret = xchg(&ctx->error, 0);
261 if (ret)
262 goto out;
263 }
264 if (status < 0) {
265 ret = status;
266 goto out;
267 }
268 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 if (do_resend)
270 ret = -EAGAIN;
271out:
272 return ret;
273}
274EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
275
276static int
277nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278{
279 int ret;
280 struct inode *inode = file_inode(file);
281
282 trace_nfs_fsync_enter(inode);
283
284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
298 } while (ret == -EAGAIN);
299
300 trace_nfs_fsync_exit(inode, ret);
301 return ret;
302}
303
304/*
305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
331 if ((file->f_mode & FMODE_READ) && /* open for read? */
332 !PageUptodate(page) && /* Uptodate? */
333 !PagePrivate(page) && /* i/o request already? */
334 pglen && /* valid bytes of file? */
335 (end < pglen || offset)) /* replace all valid bytes? */
336 return 1;
337 return 0;
338}
339
340/*
341 * This does the "real" work of the write. We must allocate and lock the
342 * page to be sent back to the generic routine, which then copies the
343 * data from user space.
344 *
345 * If the writer ends up delaying the write, the writer needs to
346 * increment the page use counts until he is done with the page.
347 */
348static int nfs_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
351{
352 int ret;
353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
354 struct page *page;
355 int once_thru = 0;
356
357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
358 file, mapping->host->i_ino, len, (long long) pos);
359
360start:
361 /*
362 * Prevent starvation issues if someone is doing a consistency
363 * sync-to-disk
364 */
365 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
366 nfs_wait_bit_killable, TASK_KILLABLE);
367 if (ret)
368 return ret;
369
370 page = grab_cache_page_write_begin(mapping, index, flags);
371 if (!page)
372 return -ENOMEM;
373 *pagep = page;
374
375 ret = nfs_flush_incompatible(file, page);
376 if (ret) {
377 unlock_page(page);
378 page_cache_release(page);
379 } else if (!once_thru &&
380 nfs_want_read_modify_write(file, page, pos, len)) {
381 once_thru = 1;
382 ret = nfs_readpage(file, page);
383 page_cache_release(page);
384 if (!ret)
385 goto start;
386 }
387 return ret;
388}
389
390static int nfs_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
393{
394 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
395 struct nfs_open_context *ctx = nfs_file_open_context(file);
396 int status;
397
398 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
399 file, mapping->host->i_ino, len, (long long) pos);
400
401 /*
402 * Zero any uninitialised parts of the page, and then mark the page
403 * as up to date if it turns out that we're extending the file.
404 */
405 if (!PageUptodate(page)) {
406 unsigned pglen = nfs_page_length(page);
407 unsigned end = offset + len;
408
409 if (pglen == 0) {
410 zero_user_segments(page, 0, offset,
411 end, PAGE_CACHE_SIZE);
412 SetPageUptodate(page);
413 } else if (end >= pglen) {
414 zero_user_segment(page, end, PAGE_CACHE_SIZE);
415 if (offset == 0)
416 SetPageUptodate(page);
417 } else
418 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
419 }
420
421 status = nfs_updatepage(file, page, offset, copied);
422
423 unlock_page(page);
424 page_cache_release(page);
425
426 if (status < 0)
427 return status;
428 NFS_I(mapping->host)->write_io += copied;
429
430 if (nfs_ctx_key_to_expire(ctx)) {
431 status = nfs_wb_all(mapping->host);
432 if (status < 0)
433 return status;
434 }
435
436 return copied;
437}
438
439/*
440 * Partially or wholly invalidate a page
441 * - Release the private state associated with a page if undergoing complete
442 * page invalidation
443 * - Called if either PG_private or PG_fscache is set on the page
444 * - Caller holds page lock
445 */
446static void nfs_invalidate_page(struct page *page, unsigned int offset,
447 unsigned int length)
448{
449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
450 page, offset, length);
451
452 if (offset != 0 || length < PAGE_CACHE_SIZE)
453 return;
454 /* Cancel any unstarted writes on this page */
455 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
456
457 nfs_fscache_invalidate_page(page, page->mapping->host);
458}
459
460/*
461 * Attempt to release the private state associated with a page
462 * - Called if either PG_private or PG_fscache is set on the page
463 * - Caller holds page lock
464 * - Return true (may release page) or false (may not)
465 */
466static int nfs_release_page(struct page *page, gfp_t gfp)
467{
468 struct address_space *mapping = page->mapping;
469
470 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
471
472 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
473 * doing this memory reclaim for a fs-related allocation.
474 */
475 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
476 !(current->flags & PF_FSTRANS)) {
477 int how = FLUSH_SYNC;
478
479 /* Don't let kswapd deadlock waiting for OOM RPC calls */
480 if (current_is_kswapd())
481 how = 0;
482 nfs_commit_inode(mapping->host, how);
483 }
484 /* If PagePrivate() is set, then the page is not freeable */
485 if (PagePrivate(page))
486 return 0;
487 return nfs_fscache_release_page(page, gfp);
488}
489
490static void nfs_check_dirty_writeback(struct page *page,
491 bool *dirty, bool *writeback)
492{
493 struct nfs_inode *nfsi;
494 struct address_space *mapping = page_file_mapping(page);
495
496 if (!mapping || PageSwapCache(page))
497 return;
498
499 /*
500 * Check if an unstable page is currently being committed and
501 * if so, have the VM treat it as if the page is under writeback
502 * so it will not block due to pages that will shortly be freeable.
503 */
504 nfsi = NFS_I(mapping->host);
505 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
506 *writeback = true;
507 return;
508 }
509
510 /*
511 * If PagePrivate() is set, then the page is not freeable and as the
512 * inode is not being committed, it's not going to be cleaned in the
513 * near future so treat it as dirty
514 */
515 if (PagePrivate(page))
516 *dirty = true;
517}
518
519/*
520 * Attempt to clear the private state associated with a page when an error
521 * occurs that requires the cached contents of an inode to be written back or
522 * destroyed
523 * - Called if either PG_private or fscache is set on the page
524 * - Caller holds page lock
525 * - Return 0 if successful, -error otherwise
526 */
527static int nfs_launder_page(struct page *page)
528{
529 struct inode *inode = page_file_mapping(page)->host;
530 struct nfs_inode *nfsi = NFS_I(inode);
531
532 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
533 inode->i_ino, (long long)page_offset(page));
534
535 nfs_fscache_wait_on_page_write(nfsi, page);
536 return nfs_wb_page(inode, page);
537}
538
539#ifdef CONFIG_NFS_SWAP
540static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
541 sector_t *span)
542{
543 *span = sis->pages;
544 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
545}
546
547static void nfs_swap_deactivate(struct file *file)
548{
549 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
550}
551#endif
552
553const struct address_space_operations nfs_file_aops = {
554 .readpage = nfs_readpage,
555 .readpages = nfs_readpages,
556 .set_page_dirty = __set_page_dirty_nobuffers,
557 .writepage = nfs_writepage,
558 .writepages = nfs_writepages,
559 .write_begin = nfs_write_begin,
560 .write_end = nfs_write_end,
561 .invalidatepage = nfs_invalidate_page,
562 .releasepage = nfs_release_page,
563 .direct_IO = nfs_direct_IO,
564 .migratepage = nfs_migrate_page,
565 .launder_page = nfs_launder_page,
566 .is_dirty_writeback = nfs_check_dirty_writeback,
567 .error_remove_page = generic_error_remove_page,
568#ifdef CONFIG_NFS_SWAP
569 .swap_activate = nfs_swap_activate,
570 .swap_deactivate = nfs_swap_deactivate,
571#endif
572};
573
574/*
575 * Notification that a PTE pointing to an NFS page is about to be made
576 * writable, implying that someone is about to modify the page through a
577 * shared-writable mapping
578 */
579static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
580{
581 struct page *page = vmf->page;
582 struct file *filp = vma->vm_file;
583 struct inode *inode = file_inode(filp);
584 unsigned pagelen;
585 int ret = VM_FAULT_NOPAGE;
586 struct address_space *mapping;
587
588 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
589 filp, filp->f_mapping->host->i_ino,
590 (long long)page_offset(page));
591
592 /* make sure the cache has finished storing the page */
593 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
594
595 lock_page(page);
596 mapping = page_file_mapping(page);
597 if (mapping != inode->i_mapping)
598 goto out_unlock;
599
600 wait_on_page_writeback(page);
601
602 pagelen = nfs_page_length(page);
603 if (pagelen == 0)
604 goto out_unlock;
605
606 ret = VM_FAULT_LOCKED;
607 if (nfs_flush_incompatible(filp, page) == 0 &&
608 nfs_updatepage(filp, page, 0, pagelen) == 0)
609 goto out;
610
611 ret = VM_FAULT_SIGBUS;
612out_unlock:
613 unlock_page(page);
614out:
615 return ret;
616}
617
618static const struct vm_operations_struct nfs_file_vm_ops = {
619 .fault = filemap_fault,
620 .map_pages = filemap_map_pages,
621 .page_mkwrite = nfs_vm_page_mkwrite,
622 .remap_pages = generic_file_remap_pages,
623};
624
625static int nfs_need_sync_write(struct file *filp, struct inode *inode)
626{
627 struct nfs_open_context *ctx;
628
629 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
630 return 1;
631 ctx = nfs_file_open_context(filp);
632 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
633 nfs_ctx_key_to_expire(ctx))
634 return 1;
635 return 0;
636}
637
638ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
639 unsigned long nr_segs, loff_t pos)
640{
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
643 unsigned long written = 0;
644 ssize_t result;
645 size_t count = iov_length(iov, nr_segs);
646
647 result = nfs_key_timeout_notify(file, inode);
648 if (result)
649 return result;
650
651 if (file->f_flags & O_DIRECT)
652 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
653
654 dprintk("NFS: write(%pD2, %lu@%Ld)\n",
655 file, (unsigned long) count, (long long) pos);
656
657 result = -EBUSY;
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
663 if (file->f_flags & O_APPEND) {
664 result = nfs_revalidate_file_size(inode, file);
665 if (result)
666 goto out;
667 }
668
669 result = count;
670 if (!count)
671 goto out;
672
673 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
674 if (result > 0)
675 written = result;
676
677 /* Return error values for O_DSYNC and IS_SYNC() */
678 if (result >= 0 && nfs_need_sync_write(file, inode)) {
679 int err = vfs_fsync(file, 0);
680 if (err < 0)
681 result = err;
682 }
683 if (result > 0)
684 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
685out:
686 return result;
687
688out_swapfile:
689 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 goto out;
691}
692EXPORT_SYMBOL_GPL(nfs_file_write);
693
694ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
695 struct file *filp, loff_t *ppos,
696 size_t count, unsigned int flags)
697{
698 struct inode *inode = file_inode(filp);
699 unsigned long written = 0;
700 ssize_t ret;
701
702 dprintk("NFS splice_write(%pD2, %lu@%llu)\n",
703 filp, (unsigned long) count, (unsigned long long) *ppos);
704
705 /*
706 * The combination of splice and an O_APPEND destination is disallowed.
707 */
708
709 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
710 if (ret > 0)
711 written = ret;
712
713 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
714 int err = vfs_fsync(filp, 0);
715 if (err < 0)
716 ret = err;
717 }
718 if (ret > 0)
719 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
720 return ret;
721}
722EXPORT_SYMBOL_GPL(nfs_file_splice_write);
723
724static int
725do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
726{
727 struct inode *inode = filp->f_mapping->host;
728 int status = 0;
729 unsigned int saved_type = fl->fl_type;
730
731 /* Try local locking first */
732 posix_test_lock(filp, fl);
733 if (fl->fl_type != F_UNLCK) {
734 /* found a conflict */
735 goto out;
736 }
737 fl->fl_type = saved_type;
738
739 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
740 goto out_noconflict;
741
742 if (is_local)
743 goto out_noconflict;
744
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746out:
747 return status;
748out_noconflict:
749 fl->fl_type = F_UNLCK;
750 goto out;
751}
752
753static int do_vfs_lock(struct file *file, struct file_lock *fl)
754{
755 int res = 0;
756 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
757 case FL_POSIX:
758 res = posix_lock_file_wait(file, fl);
759 break;
760 case FL_FLOCK:
761 res = flock_lock_file_wait(file, fl);
762 break;
763 default:
764 BUG();
765 }
766 return res;
767}
768
769static int
770do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
771{
772 struct inode *inode = filp->f_mapping->host;
773 struct nfs_lock_context *l_ctx;
774 int status;
775
776 /*
777 * Flush all pending writes before doing anything
778 * with locks..
779 */
780 nfs_sync_mapping(filp->f_mapping);
781
782 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
783 if (!IS_ERR(l_ctx)) {
784 status = nfs_iocounter_wait(&l_ctx->io_count);
785 nfs_put_lock_context(l_ctx);
786 if (status < 0)
787 return status;
788 }
789
790 /* NOTE: special case
791 * If we're signalled while cleaning up locks on process exit, we
792 * still need to complete the unlock.
793 */
794 /*
795 * Use local locking if mounted with "-onolock" or with appropriate
796 * "-olocal_lock="
797 */
798 if (!is_local)
799 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
800 else
801 status = do_vfs_lock(filp, fl);
802 return status;
803}
804
805static int
806is_time_granular(struct timespec *ts) {
807 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
808}
809
810static int
811do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
812{
813 struct inode *inode = filp->f_mapping->host;
814 int status;
815
816 /*
817 * Flush all pending writes before doing anything
818 * with locks..
819 */
820 status = nfs_sync_mapping(filp->f_mapping);
821 if (status != 0)
822 goto out;
823
824 /*
825 * Use local locking if mounted with "-onolock" or with appropriate
826 * "-olocal_lock="
827 */
828 if (!is_local)
829 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
830 else
831 status = do_vfs_lock(filp, fl);
832 if (status < 0)
833 goto out;
834
835 /*
836 * Revalidate the cache if the server has time stamps granular
837 * enough to detect subsecond changes. Otherwise, clear the
838 * cache to prevent missing any changes.
839 *
840 * This makes locking act as a cache coherency point.
841 */
842 nfs_sync_mapping(filp->f_mapping);
843 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
844 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
845 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
846 else
847 nfs_zap_caches(inode);
848 }
849out:
850 return status;
851}
852
853/*
854 * Lock a (portion of) a file
855 */
856int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
857{
858 struct inode *inode = filp->f_mapping->host;
859 int ret = -ENOLCK;
860 int is_local = 0;
861
862 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
863 filp, fl->fl_type, fl->fl_flags,
864 (long long)fl->fl_start, (long long)fl->fl_end);
865
866 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
867
868 /* No mandatory locks over NFS */
869 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
870 goto out_err;
871
872 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
873 is_local = 1;
874
875 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
876 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
877 if (ret < 0)
878 goto out_err;
879 }
880
881 if (IS_GETLK(cmd))
882 ret = do_getlk(filp, cmd, fl, is_local);
883 else if (fl->fl_type == F_UNLCK)
884 ret = do_unlk(filp, cmd, fl, is_local);
885 else
886 ret = do_setlk(filp, cmd, fl, is_local);
887out_err:
888 return ret;
889}
890EXPORT_SYMBOL_GPL(nfs_lock);
891
892/*
893 * Lock a (portion of) a file
894 */
895int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
896{
897 struct inode *inode = filp->f_mapping->host;
898 int is_local = 0;
899
900 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
901 filp, fl->fl_type, fl->fl_flags);
902
903 if (!(fl->fl_flags & FL_FLOCK))
904 return -ENOLCK;
905
906 /*
907 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
908 * any standard. In principle we might be able to support LOCK_MAND
909 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
910 * NFS code is not set up for it.
911 */
912 if (fl->fl_type & LOCK_MAND)
913 return -EINVAL;
914
915 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
916 is_local = 1;
917
918 /* We're simulating flock() locks using posix locks on the server */
919 fl->fl_owner = (fl_owner_t)filp;
920 fl->fl_start = 0;
921 fl->fl_end = OFFSET_MAX;
922
923 if (fl->fl_type == F_UNLCK)
924 return do_unlk(filp, cmd, fl, is_local);
925 return do_setlk(filp, cmd, fl, is_local);
926}
927EXPORT_SYMBOL_GPL(nfs_flock);
928
929/*
930 * There is no protocol support for leases, so we have no way to implement
931 * them correctly in the face of opens by other clients.
932 */
933int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
934{
935 dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg);
936 return -EINVAL;
937}
938EXPORT_SYMBOL_GPL(nfs_setlease);
939
940const struct file_operations nfs_file_operations = {
941 .llseek = nfs_file_llseek,
942 .read = do_sync_read,
943 .write = do_sync_write,
944 .aio_read = nfs_file_read,
945 .aio_write = nfs_file_write,
946 .mmap = nfs_file_mmap,
947 .open = nfs_file_open,
948 .flush = nfs_file_flush,
949 .release = nfs_file_release,
950 .fsync = nfs_file_fsync,
951 .lock = nfs_lock,
952 .flock = nfs_flock,
953 .splice_read = nfs_file_splice_read,
954 .splice_write = nfs_file_splice_write,
955 .check_flags = nfs_check_flags,
956 .setlease = nfs_setlease,
957};
958EXPORT_SYMBOL_GPL(nfs_file_operations);