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
35 #include "delegation.h"
36 #include "internal.h"
37 #include "iostat.h"
38 #include "fscache.h"
39 #include "pnfs.h"
40
41 #include "nfstrace.h"
42
43 #define NFSDBG_FACILITY NFSDBG_FILE
44
45 static const struct vm_operations_struct nfs_file_vm_ops;
46
47 /* Hack for future NFS swap support */
48 #ifndef IS_SWAPFILE
49 # define IS_SWAPFILE(inode) (0)
50 #endif
51
nfs_check_flags(int flags)52 int nfs_check_flags(int flags)
53 {
54 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
55 return -EINVAL;
56
57 return 0;
58 }
59 EXPORT_SYMBOL_GPL(nfs_check_flags);
60
61 /*
62 * Open file
63 */
64 static int
nfs_file_open(struct inode * inode,struct file * filp)65 nfs_file_open(struct inode *inode, struct file *filp)
66 {
67 int res;
68
69 dprintk("NFS: open file(%pD2)\n", filp);
70
71 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
72 res = nfs_check_flags(filp->f_flags);
73 if (res)
74 return res;
75
76 res = nfs_open(inode, filp);
77 return res;
78 }
79
80 int
nfs_file_release(struct inode * inode,struct file * filp)81 nfs_file_release(struct inode *inode, struct file *filp)
82 {
83 dprintk("NFS: release(%pD2)\n", filp);
84
85 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
86 nfs_file_clear_open_context(filp);
87 return 0;
88 }
89 EXPORT_SYMBOL_GPL(nfs_file_release);
90
91 /**
92 * nfs_revalidate_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 */
nfs_revalidate_file_size(struct inode * inode,struct file * filp)102 static 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_REVAL_PAGECACHE))
109 goto force_reval;
110 return 0;
111 force_reval:
112 return __nfs_revalidate_inode(server, inode);
113 }
114
nfs_file_llseek(struct file * filp,loff_t offset,int whence)115 loff_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 }
134 EXPORT_SYMBOL_GPL(nfs_file_llseek);
135
136 /*
137 * Flush all dirty pages, and check for write errors.
138 */
139 static int
nfs_file_flush(struct file * file,fl_owner_t id)140 nfs_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
157 ssize_t
nfs_file_read(struct kiocb * iocb,struct iov_iter * to)158 nfs_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 nfs_start_io_read(inode);
171 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
172 if (!result) {
173 result = generic_file_read_iter(iocb, to);
174 if (result > 0)
175 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
176 }
177 nfs_end_io_read(inode);
178 return result;
179 }
180 EXPORT_SYMBOL_GPL(nfs_file_read);
181
182 int
nfs_file_mmap(struct file * file,struct vm_area_struct * vma)183 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
184 {
185 struct inode *inode = file_inode(file);
186 int status;
187
188 dprintk("NFS: mmap(%pD2)\n", file);
189
190 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
191 * so we call that before revalidating the mapping
192 */
193 status = generic_file_mmap(file, vma);
194 if (!status) {
195 vma->vm_ops = &nfs_file_vm_ops;
196 status = nfs_revalidate_mapping(inode, file->f_mapping);
197 }
198 return status;
199 }
200 EXPORT_SYMBOL_GPL(nfs_file_mmap);
201
202 /*
203 * Flush any dirty pages for this process, and check for write errors.
204 * The return status from this call provides a reliable indication of
205 * whether any write errors occurred for this process.
206 */
207 static int
nfs_file_fsync_commit(struct file * file,int datasync)208 nfs_file_fsync_commit(struct file *file, int datasync)
209 {
210 struct inode *inode = file_inode(file);
211 int ret, ret2;
212
213 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
214
215 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
216 ret = nfs_commit_inode(inode, FLUSH_SYNC);
217 ret2 = file_check_and_advance_wb_err(file);
218 if (ret2 < 0)
219 return ret2;
220 return ret;
221 }
222
223 int
nfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)224 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
225 {
226 struct nfs_open_context *ctx = nfs_file_open_context(file);
227 struct inode *inode = file_inode(file);
228 int ret;
229
230 trace_nfs_fsync_enter(inode);
231
232 for (;;) {
233 ret = file_write_and_wait_range(file, start, end);
234 if (ret != 0)
235 break;
236 ret = nfs_file_fsync_commit(file, datasync);
237 if (ret != 0)
238 break;
239 ret = pnfs_sync_inode(inode, !!datasync);
240 if (ret != 0)
241 break;
242 if (!test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags))
243 break;
244 /*
245 * If nfs_file_fsync_commit detected a server reboot, then
246 * resend all dirty pages that might have been covered by
247 * the NFS_CONTEXT_RESEND_WRITES flag
248 */
249 start = 0;
250 end = LLONG_MAX;
251 }
252
253 trace_nfs_fsync_exit(inode, ret);
254 return ret;
255 }
256 EXPORT_SYMBOL_GPL(nfs_file_fsync);
257
258 /*
259 * Decide whether a read/modify/write cycle may be more efficient
260 * then a modify/write/read cycle when writing to a page in the
261 * page cache.
262 *
263 * Some pNFS layout drivers can only read/write at a certain block
264 * granularity like all block devices and therefore we must perform
265 * read/modify/write whenever a page hasn't read yet and the data
266 * to be written there is not aligned to a block boundary and/or
267 * smaller than the block size.
268 *
269 * The modify/write/read cycle may occur if a page is read before
270 * being completely filled by the writer. In this situation, the
271 * page must be completely written to stable storage on the server
272 * before it can be refilled by reading in the page from the server.
273 * This can lead to expensive, small, FILE_SYNC mode writes being
274 * done.
275 *
276 * It may be more efficient to read the page first if the file is
277 * open for reading in addition to writing, the page is not marked
278 * as Uptodate, it is not dirty or waiting to be committed,
279 * indicating that it was previously allocated and then modified,
280 * that there were valid bytes of data in that range of the file,
281 * and that the new data won't completely replace the old data in
282 * that range of the file.
283 */
nfs_full_page_write(struct page * page,loff_t pos,unsigned int len)284 static bool nfs_full_page_write(struct page *page, loff_t pos, unsigned int len)
285 {
286 unsigned int pglen = nfs_page_length(page);
287 unsigned int offset = pos & (PAGE_SIZE - 1);
288 unsigned int end = offset + len;
289
290 return !pglen || (end >= pglen && !offset);
291 }
292
nfs_want_read_modify_write(struct file * file,struct page * page,loff_t pos,unsigned int len)293 static bool nfs_want_read_modify_write(struct file *file, struct page *page,
294 loff_t pos, unsigned int len)
295 {
296 /*
297 * Up-to-date pages, those with ongoing or full-page write
298 * don't need read/modify/write
299 */
300 if (PageUptodate(page) || PagePrivate(page) ||
301 nfs_full_page_write(page, pos, len))
302 return false;
303
304 if (pnfs_ld_read_whole_page(file->f_mapping->host))
305 return true;
306 /* Open for reading too? */
307 if (file->f_mode & FMODE_READ)
308 return true;
309 return false;
310 }
311
312 /*
313 * This does the "real" work of the write. We must allocate and lock the
314 * page to be sent back to the generic routine, which then copies the
315 * data from user space.
316 *
317 * If the writer ends up delaying the write, the writer needs to
318 * increment the page use counts until he is done with the page.
319 */
nfs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)320 static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned flags,
322 struct page **pagep, void **fsdata)
323 {
324 int ret;
325 pgoff_t index = pos >> PAGE_SHIFT;
326 struct page *page;
327 int once_thru = 0;
328
329 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
330 file, mapping->host->i_ino, len, (long long) pos);
331
332 start:
333 page = grab_cache_page_write_begin(mapping, index, flags);
334 if (!page)
335 return -ENOMEM;
336 *pagep = page;
337
338 ret = nfs_flush_incompatible(file, page);
339 if (ret) {
340 unlock_page(page);
341 put_page(page);
342 } else if (!once_thru &&
343 nfs_want_read_modify_write(file, page, pos, len)) {
344 once_thru = 1;
345 ret = nfs_readpage(file, page);
346 put_page(page);
347 if (!ret)
348 goto start;
349 }
350 return ret;
351 }
352
nfs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)353 static int nfs_write_end(struct file *file, struct address_space *mapping,
354 loff_t pos, unsigned len, unsigned copied,
355 struct page *page, void *fsdata)
356 {
357 unsigned offset = pos & (PAGE_SIZE - 1);
358 struct nfs_open_context *ctx = nfs_file_open_context(file);
359 int status;
360
361 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
362 file, mapping->host->i_ino, len, (long long) pos);
363
364 /*
365 * Zero any uninitialised parts of the page, and then mark the page
366 * as up to date if it turns out that we're extending the file.
367 */
368 if (!PageUptodate(page)) {
369 unsigned pglen = nfs_page_length(page);
370 unsigned end = offset + copied;
371
372 if (pglen == 0) {
373 zero_user_segments(page, 0, offset,
374 end, PAGE_SIZE);
375 SetPageUptodate(page);
376 } else if (end >= pglen) {
377 zero_user_segment(page, end, PAGE_SIZE);
378 if (offset == 0)
379 SetPageUptodate(page);
380 } else
381 zero_user_segment(page, pglen, PAGE_SIZE);
382 }
383
384 status = nfs_updatepage(file, page, offset, copied);
385
386 unlock_page(page);
387 put_page(page);
388
389 if (status < 0)
390 return status;
391 NFS_I(mapping->host)->write_io += copied;
392
393 if (nfs_ctx_key_to_expire(ctx, mapping->host))
394 nfs_wb_all(mapping->host);
395
396 return copied;
397 }
398
399 /*
400 * Partially or wholly invalidate a page
401 * - Release the private state associated with a page if undergoing complete
402 * page invalidation
403 * - Called if either PG_private or PG_fscache is set on the page
404 * - Caller holds page lock
405 */
nfs_invalidate_page(struct page * page,unsigned int offset,unsigned int length)406 static void nfs_invalidate_page(struct page *page, unsigned int offset,
407 unsigned int length)
408 {
409 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
410 page, offset, length);
411
412 if (offset != 0 || length < PAGE_SIZE)
413 return;
414 /* Cancel any unstarted writes on this page */
415 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
416
417 nfs_fscache_invalidate_page(page, page->mapping->host);
418 }
419
420 /*
421 * Attempt to release the private state associated with a page
422 * - Called if either PG_private or PG_fscache is set on the page
423 * - Caller holds page lock
424 * - Return true (may release page) or false (may not)
425 */
nfs_release_page(struct page * page,gfp_t gfp)426 static int nfs_release_page(struct page *page, gfp_t gfp)
427 {
428 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
429
430 /* If PagePrivate() is set, then the page is not freeable */
431 if (PagePrivate(page))
432 return 0;
433 return nfs_fscache_release_page(page, gfp);
434 }
435
nfs_check_dirty_writeback(struct page * page,bool * dirty,bool * writeback)436 static void nfs_check_dirty_writeback(struct page *page,
437 bool *dirty, bool *writeback)
438 {
439 struct nfs_inode *nfsi;
440 struct address_space *mapping = page_file_mapping(page);
441
442 if (!mapping || PageSwapCache(page))
443 return;
444
445 /*
446 * Check if an unstable page is currently being committed and
447 * if so, have the VM treat it as if the page is under writeback
448 * so it will not block due to pages that will shortly be freeable.
449 */
450 nfsi = NFS_I(mapping->host);
451 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
452 *writeback = true;
453 return;
454 }
455
456 /*
457 * If PagePrivate() is set, then the page is not freeable and as the
458 * inode is not being committed, it's not going to be cleaned in the
459 * near future so treat it as dirty
460 */
461 if (PagePrivate(page))
462 *dirty = true;
463 }
464
465 /*
466 * Attempt to clear the private state associated with a page when an error
467 * occurs that requires the cached contents of an inode to be written back or
468 * destroyed
469 * - Called if either PG_private or fscache is set on the page
470 * - Caller holds page lock
471 * - Return 0 if successful, -error otherwise
472 */
nfs_launder_page(struct page * page)473 static int nfs_launder_page(struct page *page)
474 {
475 struct inode *inode = page_file_mapping(page)->host;
476 struct nfs_inode *nfsi = NFS_I(inode);
477
478 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
479 inode->i_ino, (long long)page_offset(page));
480
481 nfs_fscache_wait_on_page_write(nfsi, page);
482 return nfs_wb_page(inode, page);
483 }
484
nfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)485 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
486 sector_t *span)
487 {
488 unsigned long blocks;
489 long long isize;
490 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
491 struct inode *inode = file->f_mapping->host;
492
493 spin_lock(&inode->i_lock);
494 blocks = inode->i_blocks;
495 isize = inode->i_size;
496 spin_unlock(&inode->i_lock);
497 if (blocks*512 < isize) {
498 pr_warn("swap activate: swapfile has holes\n");
499 return -EINVAL;
500 }
501
502 *span = sis->pages;
503
504 return rpc_clnt_swap_activate(clnt);
505 }
506
nfs_swap_deactivate(struct file * file)507 static void nfs_swap_deactivate(struct file *file)
508 {
509 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
510
511 rpc_clnt_swap_deactivate(clnt);
512 }
513
514 const struct address_space_operations nfs_file_aops = {
515 .readpage = nfs_readpage,
516 .readpages = nfs_readpages,
517 .set_page_dirty = __set_page_dirty_nobuffers,
518 .writepage = nfs_writepage,
519 .writepages = nfs_writepages,
520 .write_begin = nfs_write_begin,
521 .write_end = nfs_write_end,
522 .invalidatepage = nfs_invalidate_page,
523 .releasepage = nfs_release_page,
524 .direct_IO = nfs_direct_IO,
525 #ifdef CONFIG_MIGRATION
526 .migratepage = nfs_migrate_page,
527 #endif
528 .launder_page = nfs_launder_page,
529 .is_dirty_writeback = nfs_check_dirty_writeback,
530 .error_remove_page = generic_error_remove_page,
531 .swap_activate = nfs_swap_activate,
532 .swap_deactivate = nfs_swap_deactivate,
533 };
534
535 /*
536 * Notification that a PTE pointing to an NFS page is about to be made
537 * writable, implying that someone is about to modify the page through a
538 * shared-writable mapping
539 */
nfs_vm_page_mkwrite(struct vm_fault * vmf)540 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
541 {
542 struct page *page = vmf->page;
543 struct file *filp = vmf->vma->vm_file;
544 struct inode *inode = file_inode(filp);
545 unsigned pagelen;
546 vm_fault_t ret = VM_FAULT_NOPAGE;
547 struct address_space *mapping;
548
549 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
550 filp, filp->f_mapping->host->i_ino,
551 (long long)page_offset(page));
552
553 sb_start_pagefault(inode->i_sb);
554
555 /* make sure the cache has finished storing the page */
556 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
557
558 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
559 nfs_wait_bit_killable, TASK_KILLABLE);
560
561 lock_page(page);
562 mapping = page_file_mapping(page);
563 if (mapping != inode->i_mapping)
564 goto out_unlock;
565
566 wait_on_page_writeback(page);
567
568 pagelen = nfs_page_length(page);
569 if (pagelen == 0)
570 goto out_unlock;
571
572 ret = VM_FAULT_LOCKED;
573 if (nfs_flush_incompatible(filp, page) == 0 &&
574 nfs_updatepage(filp, page, 0, pagelen) == 0)
575 goto out;
576
577 ret = VM_FAULT_SIGBUS;
578 out_unlock:
579 unlock_page(page);
580 out:
581 sb_end_pagefault(inode->i_sb);
582 return ret;
583 }
584
585 static const struct vm_operations_struct nfs_file_vm_ops = {
586 .fault = filemap_fault,
587 .map_pages = filemap_map_pages,
588 .page_mkwrite = nfs_vm_page_mkwrite,
589 };
590
nfs_need_check_write(struct file * filp,struct inode * inode,int error)591 static int nfs_need_check_write(struct file *filp, struct inode *inode,
592 int error)
593 {
594 struct nfs_open_context *ctx;
595
596 ctx = nfs_file_open_context(filp);
597 if (nfs_error_is_fatal_on_server(error) ||
598 nfs_ctx_key_to_expire(ctx, inode))
599 return 1;
600 return 0;
601 }
602
nfs_file_write(struct kiocb * iocb,struct iov_iter * from)603 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
604 {
605 struct file *file = iocb->ki_filp;
606 struct inode *inode = file_inode(file);
607 unsigned long written = 0;
608 ssize_t result;
609 errseq_t since;
610 int error;
611
612 result = nfs_key_timeout_notify(file, inode);
613 if (result)
614 return result;
615
616 if (iocb->ki_flags & IOCB_DIRECT)
617 return nfs_file_direct_write(iocb, from, false);
618
619 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
620 file, iov_iter_count(from), (long long) iocb->ki_pos);
621
622 if (IS_SWAPFILE(inode))
623 goto out_swapfile;
624 /*
625 * O_APPEND implies that we must revalidate the file length.
626 */
627 if (iocb->ki_flags & IOCB_APPEND) {
628 result = nfs_revalidate_file_size(inode, file);
629 if (result)
630 goto out;
631 }
632 if (iocb->ki_pos > i_size_read(inode))
633 nfs_revalidate_mapping(inode, file->f_mapping);
634
635 since = filemap_sample_wb_err(file->f_mapping);
636 nfs_start_io_write(inode);
637 result = generic_write_checks(iocb, from);
638 if (result > 0) {
639 current->backing_dev_info = inode_to_bdi(inode);
640 result = generic_perform_write(file, from, iocb->ki_pos);
641 current->backing_dev_info = NULL;
642 }
643 nfs_end_io_write(inode);
644 if (result <= 0)
645 goto out;
646
647 written = result;
648 iocb->ki_pos += written;
649 result = generic_write_sync(iocb, written);
650 if (result < 0)
651 goto out;
652
653 /* Return error values */
654 error = filemap_check_wb_err(file->f_mapping, since);
655 if (nfs_need_check_write(file, inode, error)) {
656 int err = nfs_wb_all(inode);
657 if (err < 0)
658 result = err;
659 }
660 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
661 out:
662 return result;
663
664 out_swapfile:
665 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
666 return -ETXTBSY;
667 }
668 EXPORT_SYMBOL_GPL(nfs_file_write);
669
670 static int
do_getlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)671 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
672 {
673 struct inode *inode = filp->f_mapping->host;
674 int status = 0;
675 unsigned int saved_type = fl->fl_type;
676
677 /* Try local locking first */
678 posix_test_lock(filp, fl);
679 if (fl->fl_type != F_UNLCK) {
680 /* found a conflict */
681 goto out;
682 }
683 fl->fl_type = saved_type;
684
685 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
686 goto out_noconflict;
687
688 if (is_local)
689 goto out_noconflict;
690
691 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
692 out:
693 return status;
694 out_noconflict:
695 fl->fl_type = F_UNLCK;
696 goto out;
697 }
698
699 static int
do_unlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)700 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
701 {
702 struct inode *inode = filp->f_mapping->host;
703 struct nfs_lock_context *l_ctx;
704 int status;
705
706 /*
707 * Flush all pending writes before doing anything
708 * with locks..
709 */
710 nfs_wb_all(inode);
711
712 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
713 if (!IS_ERR(l_ctx)) {
714 status = nfs_iocounter_wait(l_ctx);
715 nfs_put_lock_context(l_ctx);
716 /* NOTE: special case
717 * If we're signalled while cleaning up locks on process exit, we
718 * still need to complete the unlock.
719 */
720 if (status < 0 && !(fl->fl_flags & FL_CLOSE))
721 return status;
722 }
723
724 /*
725 * Use local locking if mounted with "-onolock" or with appropriate
726 * "-olocal_lock="
727 */
728 if (!is_local)
729 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
730 else
731 status = locks_lock_file_wait(filp, fl);
732 return status;
733 }
734
735 static int
do_setlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)736 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
737 {
738 struct inode *inode = filp->f_mapping->host;
739 int status;
740
741 /*
742 * Flush all pending writes before doing anything
743 * with locks..
744 */
745 status = nfs_sync_mapping(filp->f_mapping);
746 if (status != 0)
747 goto out;
748
749 /*
750 * Use local locking if mounted with "-onolock" or with appropriate
751 * "-olocal_lock="
752 */
753 if (!is_local)
754 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
755 else
756 status = locks_lock_file_wait(filp, fl);
757 if (status < 0)
758 goto out;
759
760 /*
761 * Invalidate cache to prevent missing any changes. If
762 * the file is mapped, clear the page cache as well so
763 * those mappings will be loaded.
764 *
765 * This makes locking act as a cache coherency point.
766 */
767 nfs_sync_mapping(filp->f_mapping);
768 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
769 nfs_zap_caches(inode);
770 if (mapping_mapped(filp->f_mapping))
771 nfs_revalidate_mapping(inode, filp->f_mapping);
772 }
773 out:
774 return status;
775 }
776
777 /*
778 * Lock a (portion of) a file
779 */
nfs_lock(struct file * filp,int cmd,struct file_lock * fl)780 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
781 {
782 struct inode *inode = filp->f_mapping->host;
783 int ret = -ENOLCK;
784 int is_local = 0;
785
786 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
787 filp, fl->fl_type, fl->fl_flags,
788 (long long)fl->fl_start, (long long)fl->fl_end);
789
790 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
791
792 /* No mandatory locks over NFS */
793 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
794 goto out_err;
795
796 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
797 is_local = 1;
798
799 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
800 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
801 if (ret < 0)
802 goto out_err;
803 }
804
805 if (IS_GETLK(cmd))
806 ret = do_getlk(filp, cmd, fl, is_local);
807 else if (fl->fl_type == F_UNLCK)
808 ret = do_unlk(filp, cmd, fl, is_local);
809 else
810 ret = do_setlk(filp, cmd, fl, is_local);
811 out_err:
812 return ret;
813 }
814 EXPORT_SYMBOL_GPL(nfs_lock);
815
816 /*
817 * Lock a (portion of) a file
818 */
nfs_flock(struct file * filp,int cmd,struct file_lock * fl)819 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
820 {
821 struct inode *inode = filp->f_mapping->host;
822 int is_local = 0;
823
824 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
825 filp, fl->fl_type, fl->fl_flags);
826
827 if (!(fl->fl_flags & FL_FLOCK))
828 return -ENOLCK;
829
830 /*
831 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
832 * any standard. In principle we might be able to support LOCK_MAND
833 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
834 * NFS code is not set up for it.
835 */
836 if (fl->fl_type & LOCK_MAND)
837 return -EINVAL;
838
839 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
840 is_local = 1;
841
842 /* We're simulating flock() locks using posix locks on the server */
843 if (fl->fl_type == F_UNLCK)
844 return do_unlk(filp, cmd, fl, is_local);
845 return do_setlk(filp, cmd, fl, is_local);
846 }
847 EXPORT_SYMBOL_GPL(nfs_flock);
848
849 const struct file_operations nfs_file_operations = {
850 .llseek = nfs_file_llseek,
851 .read_iter = nfs_file_read,
852 .write_iter = nfs_file_write,
853 .mmap = nfs_file_mmap,
854 .open = nfs_file_open,
855 .flush = nfs_file_flush,
856 .release = nfs_file_release,
857 .fsync = nfs_file_fsync,
858 .lock = nfs_lock,
859 .flock = nfs_flock,
860 .splice_read = generic_file_splice_read,
861 .splice_write = iter_file_splice_write,
862 .check_flags = nfs_check_flags,
863 .setlease = simple_nosetlease,
864 };
865 EXPORT_SYMBOL_GPL(nfs_file_operations);
866