1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * Changes Copyright (C) 1994 by Florian La Roche
8 * - Do not copy data too often around in the kernel.
9 * - In nfs_file_read the return value of kmalloc wasn't checked.
10 * - Put in a better version of read look-ahead buffering. Original idea
11 * and implementation by Wai S Kok elekokws@ee.nus.sg.
12 *
13 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 *
15 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 *
17 * nfs regular file handling functions
18 */
19
20 #include <linux/module.h>
21 #include <linux/time.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/fcntl.h>
25 #include <linux/stat.h>
26 #include <linux/nfs_fs.h>
27 #include <linux/nfs_mount.h>
28 #include <linux/mm.h>
29 #include <linux/pagemap.h>
30 #include <linux/gfp.h>
31 #include <linux/swap.h>
32 #include <linux/compaction.h>
33
34 #include <linux/uaccess.h>
35 #include <linux/filelock.h>
36
37 #include "delegation.h"
38 #include "internal.h"
39 #include "iostat.h"
40 #include "fscache.h"
41 #include "pnfs.h"
42
43 #include "nfstrace.h"
44
45 #define NFSDBG_FACILITY NFSDBG_FILE
46
47 static const struct vm_operations_struct nfs_file_vm_ops;
48
nfs_check_flags(int flags)49 int nfs_check_flags(int flags)
50 {
51 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
52 return -EINVAL;
53
54 return 0;
55 }
56 EXPORT_SYMBOL_GPL(nfs_check_flags);
57
58 /*
59 * Open file
60 */
61 static int
nfs_file_open(struct inode * inode,struct file * filp)62 nfs_file_open(struct inode *inode, struct file *filp)
63 {
64 int res;
65
66 dprintk("NFS: open file(%pD2)\n", filp);
67
68 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
69 res = nfs_check_flags(filp->f_flags);
70 if (res)
71 return res;
72
73 res = nfs_open(inode, filp);
74 if (res == 0)
75 filp->f_mode |= FMODE_CAN_ODIRECT;
76 return res;
77 }
78
79 int
nfs_file_release(struct inode * inode,struct file * filp)80 nfs_file_release(struct inode *inode, struct file *filp)
81 {
82 dprintk("NFS: release(%pD2)\n", filp);
83
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 nfs_file_clear_open_context(filp);
86 nfs_fscache_release_file(inode, filp);
87 return 0;
88 }
89 EXPORT_SYMBOL_GPL(nfs_file_release);
90
91 /**
92 * nfs_revalidate_file_size - Revalidate the file size
93 * @inode: pointer to inode struct
94 * @filp: pointer to struct file
95 *
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
101 */
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_INVALID_SIZE))
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 result = nfs_start_io_read(inode);
171 if (result)
172 return result;
173
174 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
175 if (!result) {
176 result = generic_file_read_iter(iocb, to);
177 if (result > 0)
178 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
179 }
180 nfs_end_io_read(inode);
181 return result;
182 }
183 EXPORT_SYMBOL_GPL(nfs_file_read);
184
185 ssize_t
nfs_file_splice_read(struct file * in,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)186 nfs_file_splice_read(struct file *in, loff_t *ppos, struct pipe_inode_info *pipe,
187 size_t len, unsigned int flags)
188 {
189 struct inode *inode = file_inode(in);
190 ssize_t result;
191
192 dprintk("NFS: splice_read(%pD2, %zu@%llu)\n", in, len, *ppos);
193
194 result = nfs_start_io_read(inode);
195 if (result)
196 return result;
197
198 result = nfs_revalidate_mapping(inode, in->f_mapping);
199 if (!result) {
200 result = filemap_splice_read(in, ppos, pipe, len, flags);
201 if (result > 0)
202 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
203 }
204 nfs_end_io_read(inode);
205 return result;
206 }
207 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
208
209 int
nfs_file_mmap_prepare(struct vm_area_desc * desc)210 nfs_file_mmap_prepare(struct vm_area_desc *desc)
211 {
212 struct file *file = desc->file;
213 struct inode *inode = file_inode(file);
214 int status;
215
216 dprintk("NFS: mmap(%pD2)\n", file);
217
218 /* Note: generic_file_mmap_prepare() returns ENOSYS on nommu systems
219 * so we call that before revalidating the mapping
220 */
221 status = generic_file_mmap_prepare(desc);
222 if (!status) {
223 desc->vm_ops = &nfs_file_vm_ops;
224 status = nfs_revalidate_mapping(inode, file->f_mapping);
225 }
226 return status;
227 }
228 EXPORT_SYMBOL_GPL(nfs_file_mmap_prepare);
229
230 /*
231 * Flush any dirty pages for this process, and check for write errors.
232 * The return status from this call provides a reliable indication of
233 * whether any write errors occurred for this process.
234 */
235 static int
nfs_file_fsync_commit(struct file * file,int datasync)236 nfs_file_fsync_commit(struct file *file, int datasync)
237 {
238 struct inode *inode = file_inode(file);
239 int ret, ret2;
240
241 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
242
243 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
244 ret = nfs_commit_inode(inode, FLUSH_SYNC);
245 ret2 = file_check_and_advance_wb_err(file);
246 if (ret2 < 0)
247 return ret2;
248 return ret;
249 }
250
251 int
nfs_file_fsync(struct file * file,loff_t start,loff_t end,int datasync)252 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
253 {
254 struct inode *inode = file_inode(file);
255 struct nfs_inode *nfsi = NFS_I(inode);
256 long save_nredirtied = atomic_long_read(&nfsi->redirtied_pages);
257 long nredirtied;
258 int ret;
259
260 trace_nfs_fsync_enter(inode);
261
262 for (;;) {
263 ret = file_write_and_wait_range(file, start, end);
264 if (ret != 0)
265 break;
266 ret = nfs_file_fsync_commit(file, datasync);
267 if (ret != 0)
268 break;
269 ret = pnfs_sync_inode(inode, !!datasync);
270 if (ret != 0)
271 break;
272 nredirtied = atomic_long_read(&nfsi->redirtied_pages);
273 if (nredirtied == save_nredirtied)
274 break;
275 save_nredirtied = nredirtied;
276 }
277
278 trace_nfs_fsync_exit(inode, ret);
279 return ret;
280 }
281 EXPORT_SYMBOL_GPL(nfs_file_fsync);
282
283 /*
284 * Decide whether a read/modify/write cycle may be more efficient
285 * then a modify/write/read cycle when writing to a page in the
286 * page cache.
287 *
288 * Some pNFS layout drivers can only read/write at a certain block
289 * granularity like all block devices and therefore we must perform
290 * read/modify/write whenever a page hasn't read yet and the data
291 * to be written there is not aligned to a block boundary and/or
292 * smaller than the block size.
293 *
294 * The modify/write/read cycle may occur if a page is read before
295 * being completely filled by the writer. In this situation, the
296 * page must be completely written to stable storage on the server
297 * before it can be refilled by reading in the page from the server.
298 * This can lead to expensive, small, FILE_SYNC mode writes being
299 * done.
300 *
301 * It may be more efficient to read the page first if the file is
302 * open for reading in addition to writing, the page is not marked
303 * as Uptodate, it is not dirty or waiting to be committed,
304 * indicating that it was previously allocated and then modified,
305 * that there were valid bytes of data in that range of the file,
306 * and that the new data won't completely replace the old data in
307 * that range of the file.
308 */
nfs_folio_is_full_write(struct folio * folio,loff_t pos,unsigned int len)309 static bool nfs_folio_is_full_write(struct folio *folio, loff_t pos,
310 unsigned int len)
311 {
312 unsigned int pglen = nfs_folio_length(folio);
313 unsigned int offset = offset_in_folio(folio, pos);
314 unsigned int end = offset + len;
315
316 return !pglen || (end >= pglen && !offset);
317 }
318
nfs_want_read_modify_write(struct file * file,struct folio * folio,loff_t pos,unsigned int len)319 static bool nfs_want_read_modify_write(struct file *file, struct folio *folio,
320 loff_t pos, unsigned int len)
321 {
322 /*
323 * Up-to-date pages, those with ongoing or full-page write
324 * don't need read/modify/write
325 */
326 if (folio_test_uptodate(folio) || folio_test_private(folio) ||
327 nfs_folio_is_full_write(folio, pos, len))
328 return false;
329
330 if (pnfs_ld_read_whole_page(file_inode(file)))
331 return true;
332 /* Open for reading too? */
333 if (file->f_mode & FMODE_READ)
334 return true;
335 return false;
336 }
337
338 /*
339 * This does the "real" work of the write. We must allocate and lock the
340 * page to be sent back to the generic routine, which then copies the
341 * data from user space.
342 *
343 * If the writer ends up delaying the write, the writer needs to
344 * increment the page use counts until he is done with the page.
345 */
nfs_write_begin(const struct kiocb * iocb,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)346 static int nfs_write_begin(const struct kiocb *iocb,
347 struct address_space *mapping,
348 loff_t pos, unsigned len, struct folio **foliop,
349 void **fsdata)
350 {
351 fgf_t fgp = FGP_WRITEBEGIN;
352 struct folio *folio;
353 struct file *file = iocb->ki_filp;
354 int once_thru = 0;
355 int ret;
356
357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
358 file, mapping->host->i_ino, len, (long long) pos);
359
360 fgp |= fgf_set_order(len);
361 start:
362 folio = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp,
363 mapping_gfp_mask(mapping));
364 if (IS_ERR(folio))
365 return PTR_ERR(folio);
366 *foliop = folio;
367
368 ret = nfs_flush_incompatible(file, folio);
369 if (ret) {
370 folio_unlock(folio);
371 folio_put(folio);
372 } else if (!once_thru &&
373 nfs_want_read_modify_write(file, folio, pos, len)) {
374 once_thru = 1;
375 ret = nfs_read_folio(file, folio);
376 folio_put(folio);
377 if (!ret)
378 goto start;
379 }
380 return ret;
381 }
382
nfs_write_end(const struct kiocb * iocb,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)383 static int nfs_write_end(const struct kiocb *iocb,
384 struct address_space *mapping,
385 loff_t pos, unsigned len, unsigned copied,
386 struct folio *folio, void *fsdata)
387 {
388 struct file *file = iocb->ki_filp;
389 struct nfs_open_context *ctx = nfs_file_open_context(file);
390 unsigned offset = offset_in_folio(folio, pos);
391 int status;
392
393 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
394 file, mapping->host->i_ino, len, (long long) pos);
395
396 /*
397 * Zero any uninitialised parts of the page, and then mark the page
398 * as up to date if it turns out that we're extending the file.
399 */
400 if (!folio_test_uptodate(folio)) {
401 size_t fsize = folio_size(folio);
402 unsigned pglen = nfs_folio_length(folio);
403 unsigned end = offset + copied;
404
405 if (pglen == 0) {
406 folio_zero_segments(folio, 0, offset, end, fsize);
407 folio_mark_uptodate(folio);
408 } else if (end >= pglen) {
409 folio_zero_segment(folio, end, fsize);
410 if (offset == 0)
411 folio_mark_uptodate(folio);
412 } else
413 folio_zero_segment(folio, pglen, fsize);
414 }
415
416 status = nfs_update_folio(file, folio, offset, copied);
417
418 folio_unlock(folio);
419 folio_put(folio);
420
421 if (status < 0)
422 return status;
423 NFS_I(mapping->host)->write_io += copied;
424
425 if (nfs_ctx_key_to_expire(ctx, mapping->host))
426 nfs_wb_all(mapping->host);
427
428 return copied;
429 }
430
431 /*
432 * Partially or wholly invalidate a page
433 * - Release the private state associated with a page if undergoing complete
434 * page invalidation
435 * - Called if either PG_private or PG_fscache is set on the page
436 * - Caller holds page lock
437 */
nfs_invalidate_folio(struct folio * folio,size_t offset,size_t length)438 static void nfs_invalidate_folio(struct folio *folio, size_t offset,
439 size_t length)
440 {
441 struct inode *inode = folio->mapping->host;
442 dfprintk(PAGECACHE, "NFS: invalidate_folio(%lu, %zu, %zu)\n",
443 folio->index, offset, length);
444
445 if (offset != 0 || length < folio_size(folio))
446 return;
447 /* Cancel any unstarted writes on this page */
448 nfs_wb_folio_cancel(inode, folio);
449 folio_wait_private_2(folio); /* [DEPRECATED] */
450 trace_nfs_invalidate_folio(inode, folio_pos(folio) + offset, length);
451 }
452
453 /*
454 * Attempt to release the private state associated with a folio
455 * - Called if either private or fscache flags are set on the folio
456 * - Caller holds folio lock
457 * - Return true (may release folio) or false (may not)
458 */
nfs_release_folio(struct folio * folio,gfp_t gfp)459 static bool nfs_release_folio(struct folio *folio, gfp_t gfp)
460 {
461 dfprintk(PAGECACHE, "NFS: release_folio(%p)\n", folio);
462
463 /* If the private flag is set, then the folio is not freeable */
464 if (folio_test_private(folio)) {
465 if ((current_gfp_context(gfp) & GFP_KERNEL) != GFP_KERNEL ||
466 current_is_kswapd() || current_is_kcompactd())
467 return false;
468 if (nfs_wb_folio(folio->mapping->host, folio) < 0)
469 return false;
470 }
471 return nfs_fscache_release_folio(folio, gfp);
472 }
473
nfs_check_dirty_writeback(struct folio * folio,bool * dirty,bool * writeback)474 static void nfs_check_dirty_writeback(struct folio *folio,
475 bool *dirty, bool *writeback)
476 {
477 struct nfs_inode *nfsi;
478 struct address_space *mapping = folio->mapping;
479
480 /*
481 * Check if an unstable folio is currently being committed and
482 * if so, have the VM treat it as if the folio is under writeback
483 * so it will not block due to folios that will shortly be freeable.
484 */
485 nfsi = NFS_I(mapping->host);
486 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
487 *writeback = true;
488 return;
489 }
490
491 /*
492 * If the private flag is set, then the folio is not freeable
493 * and as the inode is not being committed, it's not going to
494 * be cleaned in the near future so treat it as dirty
495 */
496 if (folio_test_private(folio))
497 *dirty = true;
498 }
499
500 /*
501 * Attempt to clear the private state associated with a page when an error
502 * occurs that requires the cached contents of an inode to be written back or
503 * destroyed
504 * - Called if either PG_private or fscache is set on the page
505 * - Caller holds page lock
506 * - Return 0 if successful, -error otherwise
507 */
nfs_launder_folio(struct folio * folio)508 static int nfs_launder_folio(struct folio *folio)
509 {
510 struct inode *inode = folio->mapping->host;
511 int ret;
512
513 dfprintk(PAGECACHE, "NFS: launder_folio(%ld, %llu)\n",
514 inode->i_ino, folio_pos(folio));
515
516 folio_wait_private_2(folio); /* [DEPRECATED] */
517 ret = nfs_wb_folio(inode, folio);
518 trace_nfs_launder_folio_done(inode, folio_pos(folio),
519 folio_size(folio), ret);
520 return ret;
521 }
522
nfs_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)523 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
524 sector_t *span)
525 {
526 unsigned long blocks;
527 long long isize;
528 int ret;
529 struct inode *inode = file_inode(file);
530 struct rpc_clnt *clnt = NFS_CLIENT(inode);
531 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
532
533 spin_lock(&inode->i_lock);
534 blocks = inode->i_blocks;
535 isize = inode->i_size;
536 spin_unlock(&inode->i_lock);
537 if (blocks*512 < isize) {
538 pr_warn("swap activate: swapfile has holes\n");
539 return -EINVAL;
540 }
541
542 ret = rpc_clnt_swap_activate(clnt);
543 if (ret)
544 return ret;
545 ret = add_swap_extent(sis, 0, sis->max, 0);
546 if (ret < 0) {
547 rpc_clnt_swap_deactivate(clnt);
548 return ret;
549 }
550
551 *span = sis->pages;
552
553 if (cl->rpc_ops->enable_swap)
554 cl->rpc_ops->enable_swap(inode);
555
556 sis->flags |= SWP_FS_OPS;
557 return ret;
558 }
559
nfs_swap_deactivate(struct file * file)560 static void nfs_swap_deactivate(struct file *file)
561 {
562 struct inode *inode = file_inode(file);
563 struct rpc_clnt *clnt = NFS_CLIENT(inode);
564 struct nfs_client *cl = NFS_SERVER(inode)->nfs_client;
565
566 rpc_clnt_swap_deactivate(clnt);
567 if (cl->rpc_ops->disable_swap)
568 cl->rpc_ops->disable_swap(file_inode(file));
569 }
570
571 const struct address_space_operations nfs_file_aops = {
572 .read_folio = nfs_read_folio,
573 .readahead = nfs_readahead,
574 .dirty_folio = filemap_dirty_folio,
575 .writepages = nfs_writepages,
576 .write_begin = nfs_write_begin,
577 .write_end = nfs_write_end,
578 .invalidate_folio = nfs_invalidate_folio,
579 .release_folio = nfs_release_folio,
580 .migrate_folio = nfs_migrate_folio,
581 .launder_folio = nfs_launder_folio,
582 .is_dirty_writeback = nfs_check_dirty_writeback,
583 .error_remove_folio = generic_error_remove_folio,
584 .swap_activate = nfs_swap_activate,
585 .swap_deactivate = nfs_swap_deactivate,
586 .swap_rw = nfs_swap_rw,
587 };
588
589 /*
590 * Notification that a PTE pointing to an NFS page is about to be made
591 * writable, implying that someone is about to modify the page through a
592 * shared-writable mapping
593 */
nfs_vm_page_mkwrite(struct vm_fault * vmf)594 static vm_fault_t nfs_vm_page_mkwrite(struct vm_fault *vmf)
595 {
596 struct file *filp = vmf->vma->vm_file;
597 struct inode *inode = file_inode(filp);
598 unsigned pagelen;
599 vm_fault_t ret = VM_FAULT_NOPAGE;
600 struct address_space *mapping;
601 struct folio *folio = page_folio(vmf->page);
602
603 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
604 filp, filp->f_mapping->host->i_ino,
605 (long long)folio_pos(folio));
606
607 sb_start_pagefault(inode->i_sb);
608
609 /* make sure the cache has finished storing the page */
610 if (folio_test_private_2(folio) && /* [DEPRECATED] */
611 folio_wait_private_2_killable(folio) < 0) {
612 ret = VM_FAULT_RETRY;
613 goto out;
614 }
615
616 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
617 nfs_wait_bit_killable,
618 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
619
620 folio_lock(folio);
621 mapping = folio->mapping;
622 if (mapping != inode->i_mapping)
623 goto out_unlock;
624
625 folio_wait_writeback(folio);
626
627 pagelen = nfs_folio_length(folio);
628 if (pagelen == 0)
629 goto out_unlock;
630
631 ret = VM_FAULT_LOCKED;
632 if (nfs_flush_incompatible(filp, folio) == 0 &&
633 nfs_update_folio(filp, folio, 0, pagelen) == 0)
634 goto out;
635
636 ret = VM_FAULT_SIGBUS;
637 out_unlock:
638 folio_unlock(folio);
639 out:
640 sb_end_pagefault(inode->i_sb);
641 return ret;
642 }
643
644 static const struct vm_operations_struct nfs_file_vm_ops = {
645 .fault = filemap_fault,
646 .map_pages = filemap_map_pages,
647 .page_mkwrite = nfs_vm_page_mkwrite,
648 };
649
nfs_file_write(struct kiocb * iocb,struct iov_iter * from)650 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
651 {
652 struct file *file = iocb->ki_filp;
653 struct inode *inode = file_inode(file);
654 unsigned int mntflags = NFS_SERVER(inode)->flags;
655 ssize_t result, written;
656 errseq_t since;
657 int error;
658
659 result = nfs_key_timeout_notify(file, inode);
660 if (result)
661 return result;
662
663 if (iocb->ki_flags & IOCB_DIRECT)
664 return nfs_file_direct_write(iocb, from, false);
665
666 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
667 file, iov_iter_count(from), (long long) iocb->ki_pos);
668
669 if (IS_SWAPFILE(inode))
670 goto out_swapfile;
671 /*
672 * O_APPEND implies that we must revalidate the file length.
673 */
674 if (iocb->ki_flags & IOCB_APPEND || iocb->ki_pos > i_size_read(inode)) {
675 result = nfs_revalidate_file_size(inode, file);
676 if (result)
677 return result;
678 }
679
680 nfs_clear_invalid_mapping(file->f_mapping);
681
682 since = filemap_sample_wb_err(file->f_mapping);
683 error = nfs_start_io_write(inode);
684 if (error)
685 return error;
686 result = generic_write_checks(iocb, from);
687 if (result > 0)
688 result = generic_perform_write(iocb, from);
689 nfs_end_io_write(inode);
690 if (result <= 0)
691 goto out;
692
693 written = result;
694 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
695
696 if (mntflags & NFS_MOUNT_WRITE_EAGER) {
697 result = filemap_fdatawrite_range(file->f_mapping,
698 iocb->ki_pos - written,
699 iocb->ki_pos - 1);
700 if (result < 0)
701 goto out;
702 }
703 if (mntflags & NFS_MOUNT_WRITE_WAIT) {
704 filemap_fdatawait_range(file->f_mapping,
705 iocb->ki_pos - written,
706 iocb->ki_pos - 1);
707 }
708 result = generic_write_sync(iocb, written);
709 if (result < 0)
710 return result;
711
712 out:
713 /* Return error values */
714 error = filemap_check_wb_err(file->f_mapping, since);
715 switch (error) {
716 default:
717 break;
718 case -EDQUOT:
719 case -EFBIG:
720 case -ENOSPC:
721 nfs_wb_all(inode);
722 error = file_check_and_advance_wb_err(file);
723 if (error < 0)
724 result = error;
725 }
726 return result;
727
728 out_swapfile:
729 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
730 return -ETXTBSY;
731 }
732 EXPORT_SYMBOL_GPL(nfs_file_write);
733
734 static int
do_getlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)735 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
736 {
737 struct inode *inode = filp->f_mapping->host;
738 int status = 0;
739 unsigned int saved_type = fl->c.flc_type;
740
741 /* Try local locking first */
742 posix_test_lock(filp, fl);
743 if (fl->c.flc_type != F_UNLCK) {
744 /* found a conflict */
745 goto out;
746 }
747 fl->c.flc_type = saved_type;
748
749 if (nfs_have_read_or_write_delegation(inode))
750 goto out_noconflict;
751
752 if (is_local)
753 goto out_noconflict;
754
755 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
756 out:
757 return status;
758 out_noconflict:
759 fl->c.flc_type = F_UNLCK;
760 goto out;
761 }
762
763 static int
do_unlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)764 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
765 {
766 struct inode *inode = filp->f_mapping->host;
767 struct nfs_lock_context *l_ctx;
768 int status;
769
770 /*
771 * Flush all pending writes before doing anything
772 * with locks..
773 */
774 nfs_wb_all(inode);
775
776 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
777 if (!IS_ERR(l_ctx)) {
778 status = nfs_iocounter_wait(l_ctx);
779 nfs_put_lock_context(l_ctx);
780 /* NOTE: special case
781 * If we're signalled while cleaning up locks on process exit, we
782 * still need to complete the unlock.
783 */
784 if (status < 0 && !(fl->c.flc_flags & FL_CLOSE))
785 return status;
786 }
787
788 /*
789 * Use local locking if mounted with "-onolock" or with appropriate
790 * "-olocal_lock="
791 */
792 if (!is_local)
793 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
794 else
795 status = locks_lock_file_wait(filp, fl);
796 return status;
797 }
798
799 static int
do_setlk(struct file * filp,int cmd,struct file_lock * fl,int is_local)800 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
801 {
802 struct inode *inode = filp->f_mapping->host;
803 int status;
804
805 /*
806 * Flush all pending writes before doing anything
807 * with locks..
808 */
809 status = nfs_sync_mapping(filp->f_mapping);
810 if (status != 0)
811 goto out;
812
813 /*
814 * Use local locking if mounted with "-onolock" or with appropriate
815 * "-olocal_lock="
816 */
817 if (!is_local)
818 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
819 else
820 status = locks_lock_file_wait(filp, fl);
821 if (status < 0)
822 goto out;
823
824 /*
825 * Invalidate cache to prevent missing any changes. If
826 * the file is mapped, clear the page cache as well so
827 * those mappings will be loaded.
828 *
829 * This makes locking act as a cache coherency point.
830 */
831 nfs_sync_mapping(filp->f_mapping);
832 if (!nfs_have_read_or_write_delegation(inode)) {
833 nfs_zap_caches(inode);
834 if (mapping_mapped(filp->f_mapping))
835 nfs_revalidate_mapping(inode, filp->f_mapping);
836 }
837 out:
838 return status;
839 }
840
841 /*
842 * Lock a (portion of) a file
843 */
nfs_lock(struct file * filp,int cmd,struct file_lock * fl)844 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
845 {
846 struct inode *inode = filp->f_mapping->host;
847 int ret = -ENOLCK;
848 int is_local = 0;
849
850 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
851 filp, fl->c.flc_type, fl->c.flc_flags,
852 (long long)fl->fl_start, (long long)fl->fl_end);
853
854 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
855
856 if (fl->c.flc_flags & FL_RECLAIM)
857 return -ENOGRACE;
858
859 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
860 is_local = 1;
861
862 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
863 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
864 if (ret < 0)
865 goto out_err;
866 }
867
868 if (IS_GETLK(cmd))
869 ret = do_getlk(filp, cmd, fl, is_local);
870 else if (lock_is_unlock(fl))
871 ret = do_unlk(filp, cmd, fl, is_local);
872 else
873 ret = do_setlk(filp, cmd, fl, is_local);
874 out_err:
875 return ret;
876 }
877 EXPORT_SYMBOL_GPL(nfs_lock);
878
879 /*
880 * Lock a (portion of) a file
881 */
nfs_flock(struct file * filp,int cmd,struct file_lock * fl)882 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
883 {
884 struct inode *inode = filp->f_mapping->host;
885 int is_local = 0;
886
887 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
888 filp, fl->c.flc_type, fl->c.flc_flags);
889
890 if (!(fl->c.flc_flags & FL_FLOCK))
891 return -ENOLCK;
892
893 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
894 is_local = 1;
895
896 /* We're simulating flock() locks using posix locks on the server */
897 if (lock_is_unlock(fl))
898 return do_unlk(filp, cmd, fl, is_local);
899 return do_setlk(filp, cmd, fl, is_local);
900 }
901 EXPORT_SYMBOL_GPL(nfs_flock);
902
903 const struct file_operations nfs_file_operations = {
904 .llseek = nfs_file_llseek,
905 .read_iter = nfs_file_read,
906 .write_iter = nfs_file_write,
907 .mmap_prepare = nfs_file_mmap_prepare,
908 .open = nfs_file_open,
909 .flush = nfs_file_flush,
910 .release = nfs_file_release,
911 .fsync = nfs_file_fsync,
912 .lock = nfs_lock,
913 .flock = nfs_flock,
914 .splice_read = nfs_file_splice_read,
915 .splice_write = iter_file_splice_write,
916 .check_flags = nfs_check_flags,
917 .setlease = simple_nosetlease,
918 };
919 EXPORT_SYMBOL_GPL(nfs_file_operations);
920