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
46 static const struct vm_operations_struct nfs_file_vm_ops;
47
48 int 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 }
55 EXPORT_SYMBOL_GPL(nfs_check_flags);
56
57 /*
58 * Open file
59 */
60 static int
61 nfs_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
78 int
79 nfs_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 }
88 EXPORT_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 */
101 static 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;
110 force_reval:
111 return __nfs_revalidate_inode(server, inode);
112 }
113
114 loff_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 }
133 EXPORT_SYMBOL_GPL(nfs_file_llseek);
134
135 /*
136 * Flush all dirty pages, and check for write errors.
137 */
138 static int
139 nfs_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
156 ssize_t
157 nfs_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 }
179 EXPORT_SYMBOL_GPL(nfs_file_read);
180
181 ssize_t
182 nfs_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 }
200 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
201
202 int
203 nfs_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 }
220 EXPORT_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 */
227 static int
228 nfs_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
243 int
244 nfs_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 }
273 EXPORT_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 */
301 static 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
311 static 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 */
338 static 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 fgf_t fgp = FGP_WRITEBEGIN;
343 struct folio *folio;
344 int once_thru = 0;
345 int ret;
346
347 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
348 file, mapping->host->i_ino, len, (long long) pos);
349
350 fgp |= fgf_set_order(len);
351 start:
352 folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp,
353 mapping_gfp_mask(mapping));
354 if (IS_ERR(folio))
355 return PTR_ERR(folio);
356 *pagep = &folio->page;
357
358 ret = nfs_flush_incompatible(file, folio);
359 if (ret) {
360 folio_unlock(folio);
361 folio_put(folio);
362 } else if (!once_thru &&
363 nfs_want_read_modify_write(file, folio, pos, len)) {
364 once_thru = 1;
365 ret = nfs_read_folio(file, folio);
366 folio_put(folio);
367 if (!ret)
368 goto start;
369 }
370 return ret;
371 }
372
373 static int nfs_write_end(struct file *file, struct address_space *mapping,
374 loff_t pos, unsigned len, unsigned copied,
375 struct page *page, void *fsdata)
376 {
377 struct nfs_open_context *ctx = nfs_file_open_context(file);
378 struct folio *folio = page_folio(page);
379 unsigned offset = offset_in_folio(folio, pos);
380 int status;
381
382 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
383 file, mapping->host->i_ino, len, (long long) pos);
384
385 /*
386 * Zero any uninitialised parts of the page, and then mark the page
387 * as up to date if it turns out that we're extending the file.
388 */
389 if (!folio_test_uptodate(folio)) {
390 size_t fsize = folio_size(folio);
391 unsigned pglen = nfs_folio_length(folio);
392 unsigned end = offset + copied;
393
394 if (pglen == 0) {
395 folio_zero_segments(folio, 0, offset, end, fsize);
396 folio_mark_uptodate(folio);
397 } else if (end >= pglen) {
398 folio_zero_segment(folio, end, fsize);
399 if (offset == 0)
400 folio_mark_uptodate(folio);
401 } else
402 folio_zero_segment(folio, pglen, fsize);
403 }
404
405 status = nfs_update_folio(file, folio, offset, copied);
406
407 folio_unlock(folio);
408 folio_put(folio);
409
410 if (status < 0)
411 return status;
412 NFS_I(mapping->host)->write_io += copied;
413
414 if (nfs_ctx_key_to_expire(ctx, mapping->host))
415 nfs_wb_all(mapping->host);
416
417 return copied;
418 }
419
420 /*
421 * Partially or wholly invalidate a page
422 * - Release the private state associated with a page if undergoing complete
423 * page invalidation
424 * - Called if either PG_private or PG_fscache is set on the page
425 * - Caller holds page lock
426 */
427 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
428 size_t length)
429 {
430 struct inode *inode = folio->mapping->host;
431 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
432 folio->index, offset, length);
433
434 if (offset != 0 || length < folio_size(folio))
435 return;
436 /* Cancel any unstarted writes on this page */
437 nfs_wb_folio_cancel(inode, folio);
438 folio_wait_private_2(folio); /* [DEPRECATED] */
439 trace_nfs_invalidate_folio(inode, folio_pos(folio) + offset, length);
440 }
441
442 /*
443 * Attempt to release the private state associated with a folio
444 * - Called if either private or fscache flags are set on the folio
445 * - Caller holds folio lock
446 * - Return true (may release folio) or false (may not)
447 */
448 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
449 {
450 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
451
452 /* If the private flag is set, then the folio is not freeable */
453 if (folio_test_private(folio)) {
454 if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
455 current_is_kswapd())
456 return false;
457 if (nfs_wb_folio(folio->mapping->host, folio) < 0)
458 return false;
459 }
460 return nfs_fscache_release_folio(folio, gfp);
461 }
462
463 static void nfs_check_dirty_writeback(struct folio *folio,
464 bool *dirty, bool *writeback)
465 {
466 struct nfs_inode *nfsi;
467 struct address_space *mapping = folio->mapping;
468
469 /*
470 * Check if an unstable folio is currently being committed and
471 * if so, have the VM treat it as if the folio is under writeback
472 * so it will not block due to folios that will shortly be freeable.
473 */
474 nfsi = NFS_I(mapping->host);
475 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
476 *writeback = true;
477 return;
478 }
479
480 /*
481 * If the private flag is set, then the folio is not freeable
482 * and as the inode is not being committed, it's not going to
483 * be cleaned in the near future so treat it as dirty
484 */
485 if (folio_test_private(folio))
486 *dirty = true;
487 }
488
489 /*
490 * Attempt to clear the private state associated with a page when an error
491 * occurs that requires the cached contents of an inode to be written back or
492 * destroyed
493 * - Called if either PG_private or fscache is set on the page
494 * - Caller holds page lock
495 * - Return 0 if successful, -error otherwise
496 */
497 static int nfs_launder_folio(struct folio *folio)
498 {
499 struct inode *inode = folio->mapping->host;
500 int ret;
501
502 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
503 inode->i_ino, folio_pos(folio));
504
505 folio_wait_private_2(folio); /* [DEPRECATED] */
506 ret = nfs_wb_folio(inode, folio);
507 trace_nfs_launder_folio_done(inode, folio_pos(folio),
508 folio_size(folio), ret);
509 return ret;
510 }
511
512 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
513 sector_t *span)
514 {
515 unsigned long blocks;
516 long long isize;
517 int ret;
518 struct inode *inode = file_inode(file);
519 struct rpc_clnt *clnt = NFS_CLIENT(inode);
520 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
521
522 spin_lock(&inode->i_lock);
523 blocks = inode->i_blocks;
524 isize = inode->i_size;
525 spin_unlock(&inode->i_lock);
526 if (blocks*512 < isize) {
527 pr_warn("swap activate: swapfile has holes\n");
528 return -EINVAL;
529 }
530
531 ret = rpc_clnt_swap_activate(clnt);
532 if (ret)
533 return ret;
534 ret = add_swap_extent(sis, 0, sis->max, 0);
535 if (ret < 0) {
536 rpc_clnt_swap_deactivate(clnt);
537 return ret;
538 }
539
540 *span = sis->pages;
541
542 if (cl->rpc_ops->enable_swap)
543 cl->rpc_ops->enable_swap(inode);
544
545 sis->flags |= SWP_FS_OPS;
546 return ret;
547 }
548
549 static void nfs_swap_deactivate(struct file *file)
550 {
551 struct inode *inode = file_inode(file);
552 struct rpc_clnt *clnt = NFS_CLIENT(inode);
553 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
554
555 rpc_clnt_swap_deactivate(clnt);
556 if (cl->rpc_ops->disable_swap)
557 cl->rpc_ops->disable_swap(file_inode(file));
558 }
559
560 const struct address_space_operations nfs_file_aops = {
561 .read_folio = nfs_read_folio,
562 .readahead = nfs_readahead,
563 .dirty_folio = filemap_dirty_folio,
564 .writepages = nfs_writepages,
565 .write_begin = nfs_write_begin,
566 .write_end = nfs_write_end,
567 .invalidate_folio = nfs_invalidate_folio,
568 .release_folio = nfs_release_folio,
569 .migrate_folio = nfs_migrate_folio,
570 .launder_folio = nfs_launder_folio,
571 .is_dirty_writeback = nfs_check_dirty_writeback,
572 .error_remove_folio = generic_error_remove_folio,
573 .swap_activate = nfs_swap_activate,
574 .swap_deactivate = nfs_swap_deactivate,
575 .swap_rw = nfs_swap_rw,
576 };
577
578 /*
579 * Notification that a PTE pointing to an NFS page is about to be made
580 * writable, implying that someone is about to modify the page through a
581 * shared-writable mapping
582 */
583 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
584 {
585 struct file *filp = vmf->vma->vm_file;
586 struct inode *inode = file_inode(filp);
587 unsigned pagelen;
588 vm_fault_t ret = VM_FAULT_NOPAGE;
589 struct address_space *mapping;
590 struct folio *folio = page_folio(vmf->page);
591
592 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
593 filp, filp->f_mapping->host->i_ino,
594 (long long)folio_pos(folio));
595
596 sb_start_pagefault(inode->i_sb);
597
598 /* make sure the cache has finished storing the page */
599 if (folio_test_private_2(folio) && /* [DEPRECATED] */
600 folio_wait_private_2_killable(folio) < 0) {
601 ret = VM_FAULT_RETRY;
602 goto out;
603 }
604
605 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
606 nfs_wait_bit_killable,
607 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
608
609 folio_lock(folio);
610 mapping = folio->mapping;
611 if (mapping != inode->i_mapping)
612 goto out_unlock;
613
614 folio_wait_writeback(folio);
615
616 pagelen = nfs_folio_length(folio);
617 if (pagelen == 0)
618 goto out_unlock;
619
620 ret = VM_FAULT_LOCKED;
621 if (nfs_flush_incompatible(filp, folio) == 0 &&
622 nfs_update_folio(filp, folio, 0, pagelen) == 0)
623 goto out;
624
625 ret = VM_FAULT_SIGBUS;
626 out_unlock:
627 folio_unlock(folio);
628 out:
629 sb_end_pagefault(inode->i_sb);
630 return ret;
631 }
632
633 static const struct vm_operations_struct nfs_file_vm_ops = {
634 .fault = filemap_fault,
635 .map_pages = filemap_map_pages,
636 .page_mkwrite = nfs_vm_page_mkwrite,
637 };
638
639 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
640 {
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
643 unsigned int mntflags = NFS_SERVER(inode)->flags;
644 ssize_t result, written;
645 errseq_t since;
646 int error;
647
648 result = nfs_key_timeout_notify(file, inode);
649 if (result)
650 return result;
651
652 if (iocb->ki_flags & IOCB_DIRECT)
653 return nfs_file_direct_write(iocb, from, false);
654
655 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
656 file, iov_iter_count(from), (long long) iocb->ki_pos);
657
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
663 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
664 result = nfs_revalidate_file_size(inode, file);
665 if (result)
666 return result;
667 }
668
669 nfs_clear_invalid_mapping(file->f_mapping);
670
671 since = filemap_sample_wb_err(file->f_mapping);
672 nfs_start_io_write(inode);
673 result = generic_write_checks(iocb, from);
674 if (result > 0)
675 result = generic_perform_write(iocb, from);
676 nfs_end_io_write(inode);
677 if (result <= 0)
678 goto out;
679
680 written = result;
681 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
682
683 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
684 result = filemap_fdatawrite_range(file->f_mapping,
685 iocb->ki_pos - written,
686 iocb->ki_pos - 1);
687 if (result < 0)
688 goto out;
689 }
690 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
691 filemap_fdatawait_range(file->f_mapping,
692 iocb->ki_pos - written,
693 iocb->ki_pos - 1);
694 }
695 result = generic_write_sync(iocb, written);
696 if (result < 0)
697 return result;
698
699 out:
700 /* Return error values */
701 error = filemap_check_wb_err(file->f_mapping, since);
702 switch (error) {
703 default:
704 break;
705 case -EDQUOT:
706 case -EFBIG:
707 case -ENOSPC:
708 nfs_wb_all(inode);
709 error = file_check_and_advance_wb_err(file);
710 if (error < 0)
711 result = error;
712 }
713 return result;
714
715 out_swapfile:
716 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
717 return -ETXTBSY;
718 }
719 EXPORT_SYMBOL_GPL(nfs_file_write);
720
721 static int
722 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
723 {
724 struct inode *inode = filp->f_mapping->host;
725 int status = 0;
726 unsigned int saved_type = fl->c.flc_type;
727
728 /* Try local locking first */
729 posix_test_lock(filp, fl);
730 if (fl->c.flc_type != F_UNLCK) {
731 /* found a conflict */
732 goto out;
733 }
734 fl->c.flc_type = saved_type;
735
736 if (nfs_have_read_or_write_delegation(inode))
737 goto out_noconflict;
738
739 if (is_local)
740 goto out_noconflict;
741
742 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
743 out:
744 return status;
745 out_noconflict:
746 fl->c.flc_type = F_UNLCK;
747 goto out;
748 }
749
750 static int
751 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
752 {
753 struct inode *inode = filp->f_mapping->host;
754 struct nfs_lock_context *l_ctx;
755 int status;
756
757 /*
758 * Flush all pending writes before doing anything
759 * with locks..
760 */
761 nfs_wb_all(inode);
762
763 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
764 if (!IS_ERR(l_ctx)) {
765 status = nfs_iocounter_wait(l_ctx);
766 nfs_put_lock_context(l_ctx);
767 /* NOTE: special case
768 * If we're signalled while cleaning up locks on process exit, we
769 * still need to complete the unlock.
770 */
771 if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
772 return status;
773 }
774
775 /*
776 * Use local locking if mounted with "-onolock" or with appropriate
777 * "-olocal_lock="
778 */
779 if (!is_local)
780 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
781 else
782 status = locks_lock_file_wait(filp, fl);
783 return status;
784 }
785
786 static int
787 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
788 {
789 struct inode *inode = filp->f_mapping->host;
790 int status;
791
792 /*
793 * Flush all pending writes before doing anything
794 * with locks..
795 */
796 status = nfs_sync_mapping(filp->f_mapping);
797 if (status != 0)
798 goto out;
799
800 /*
801 * Use local locking if mounted with "-onolock" or with appropriate
802 * "-olocal_lock="
803 */
804 if (!is_local)
805 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
806 else
807 status = locks_lock_file_wait(filp, fl);
808 if (status < 0)
809 goto out;
810
811 /*
812 * Invalidate cache to prevent missing any changes. If
813 * the file is mapped, clear the page cache as well so
814 * those mappings will be loaded.
815 *
816 * This makes locking act as a cache coherency point.
817 */
818 nfs_sync_mapping(filp->f_mapping);
819 if (!nfs_have_read_or_write_delegation(inode)) {
820 nfs_zap_caches(inode);
821 if (mapping_mapped(filp->f_mapping))
822 nfs_revalidate_mapping(inode, filp->f_mapping);
823 }
824 out:
825 return status;
826 }
827
828 /*
829 * Lock a (portion of) a file
830 */
831 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
832 {
833 struct inode *inode = filp->f_mapping->host;
834 int ret = -ENOLCK;
835 int is_local = 0;
836
837 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
838 filp, fl->c.flc_type, fl->c.flc_flags,
839 (long long)fl->fl_start, (long long)fl->fl_end);
840
841 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
842
843 if (fl->c.flc_flags & FL_RECLAIM)
844 return -ENOGRACE;
845
846 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
847 is_local = 1;
848
849 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
850 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
851 if (ret < 0)
852 goto out_err;
853 }
854
855 if (IS_GETLK(cmd))
856 ret = do_getlk(filp, cmd, fl, is_local);
857 else if (lock_is_unlock(fl))
858 ret = do_unlk(filp, cmd, fl, is_local);
859 else
860 ret = do_setlk(filp, cmd, fl, is_local);
861 out_err:
862 return ret;
863 }
864 EXPORT_SYMBOL_GPL(nfs_lock);
865
866 /*
867 * Lock a (portion of) a file
868 */
869 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
870 {
871 struct inode *inode = filp->f_mapping->host;
872 int is_local = 0;
873
874 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
875 filp, fl->c.flc_type, fl->c.flc_flags);
876
877 if (!(fl->c.flc_flags & FL_FLOCK))
878 return -ENOLCK;
879
880 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
881 is_local = 1;
882
883 /* We're simulating flock() locks using posix locks on the server */
884 if (lock_is_unlock(fl))
885 return do_unlk(filp, cmd, fl, is_local);
886 return do_setlk(filp, cmd, fl, is_local);
887 }
888 EXPORT_SYMBOL_GPL(nfs_flock);
889
890 const struct file_operations nfs_file_operations = {
891 .llseek = nfs_file_llseek,
892 .read_iter = nfs_file_read,
893 .write_iter = nfs_file_write,
894 .mmap = nfs_file_mmap,
895 .open = nfs_file_open,
896 .flush = nfs_file_flush,
897 .release = nfs_file_release,
898 .fsync = nfs_file_fsync,
899 .lock = nfs_lock,
900 .flock = nfs_flock,
901 .splice_read = nfs_file_splice_read,
902 .splice_write = iter_file_splice_write,
903 .check_flags = nfs_check_flags,
904 .setlease = simple_nosetlease,
905 };
906 EXPORT_SYMBOL_GPL(nfs_file_operations);
907
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.