]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/nfs/file.c
NFS: Fix fdatasync/fsync() when confronted with a server reboot
[mirror_ubuntu-jammy-kernel.git] / fs / nfs / file.c
CommitLineData
1da177e4
LT
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
ddda8e0a 19#include <linux/module.h>
1da177e4
LT
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>
1da177e4 28#include <linux/pagemap.h>
e8edc6e0 29#include <linux/aio.h>
5a0e3ad6 30#include <linux/gfp.h>
b608b283 31#include <linux/swap.h>
1da177e4
LT
32
33#include <asm/uaccess.h>
1da177e4
LT
34
35#include "delegation.h"
94387fb1 36#include "internal.h"
91d5b470 37#include "iostat.h"
545db45f 38#include "fscache.h"
1da177e4
LT
39
40#define NFSDBG_FACILITY NFSDBG_FILE
41
f0f37e2f 42static const struct vm_operations_struct nfs_file_vm_ops;
94387fb1 43
1da177e4
LT
44/* Hack for future NFS swap support */
45#ifndef IS_SWAPFILE
46# define IS_SWAPFILE(inode) (0)
47#endif
48
ce4ef7c0 49int nfs_check_flags(int flags)
1da177e4
LT
50{
51 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
52 return -EINVAL;
53
54 return 0;
55}
89d77c8f 56EXPORT_SYMBOL_GPL(nfs_check_flags);
1da177e4
LT
57
58/*
59 * Open file
60 */
61static int
62nfs_file_open(struct inode *inode, struct file *filp)
63{
1da177e4
LT
64 int res;
65
6da24bc9 66 dprintk("NFS: open file(%s/%s)\n",
cc0dd2d1
CL
67 filp->f_path.dentry->d_parent->d_name.name,
68 filp->f_path.dentry->d_name.name);
69
c2459dc4 70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
1da177e4
LT
71 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
46cb650c 75 res = nfs_open(inode, filp);
1da177e4
LT
76 return res;
77}
78
ce4ef7c0 79int
1da177e4
LT
80nfs_file_release(struct inode *inode, struct file *filp)
81{
6da24bc9 82 dprintk("NFS: release(%s/%s)\n",
6f276e49
RM
83 filp->f_path.dentry->d_parent->d_name.name,
84 filp->f_path.dentry->d_name.name);
6da24bc9 85
91d5b470 86 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
46cb650c 87 return nfs_release(inode, filp);
1da177e4 88}
89d77c8f 89EXPORT_SYMBOL_GPL(nfs_file_release);
1da177e4 90
980802e3
TM
91/**
92 * nfs_revalidate_size - Revalidate the file size
93 * @inode - pointer to inode struct
94 * @file - pointer to struct file
95 *
96 * Revalidates the file length. This is basically a wrapper around
97 * nfs_revalidate_inode() that takes into account the fact that we may
98 * have cached writes (in which case we don't care about the server's
99 * idea of what the file length is), or O_DIRECT (in which case we
100 * shouldn't trust the cache).
101 */
102static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
103{
104 struct nfs_server *server = NFS_SERVER(inode);
105 struct nfs_inode *nfsi = NFS_I(inode);
106
d7cf8dd0
TM
107 if (nfs_have_delegated_attributes(inode))
108 goto out_noreval;
109
980802e3
TM
110 if (filp->f_flags & O_DIRECT)
111 goto force_reval;
d7cf8dd0
TM
112 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
113 goto force_reval;
114 if (nfs_attribute_timeout(inode))
115 goto force_reval;
116out_noreval:
117 return 0;
980802e3
TM
118force_reval:
119 return __nfs_revalidate_inode(server, inode);
120}
121
ce4ef7c0 122loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
980802e3 123{
6da24bc9 124 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
b84e06c5
CL
125 filp->f_path.dentry->d_parent->d_name.name,
126 filp->f_path.dentry->d_name.name,
127 offset, origin);
128
06222e49
JB
129 /*
130 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
131 * the cached file length
132 */
6c529617 133 if (origin != SEEK_SET && origin != SEEK_CUR) {
980802e3 134 struct inode *inode = filp->f_mapping->host;
d5e66348 135
980802e3
TM
136 int retval = nfs_revalidate_file_size(inode, filp);
137 if (retval < 0)
138 return (loff_t)retval;
79835a71 139 }
d5e66348 140
79835a71 141 return generic_file_llseek(filp, offset, origin);
980802e3 142}
89d77c8f 143EXPORT_SYMBOL_GPL(nfs_file_llseek);
980802e3 144
1da177e4
LT
145/*
146 * Flush all dirty pages, and check for write errors.
1da177e4 147 */
ce4ef7c0 148int
75e1fcc0 149nfs_file_flush(struct file *file, fl_owner_t id)
1da177e4 150{
6da24bc9
CL
151 struct dentry *dentry = file->f_path.dentry;
152 struct inode *inode = dentry->d_inode;
1da177e4 153
6da24bc9
CL
154 dprintk("NFS: flush(%s/%s)\n",
155 dentry->d_parent->d_name.name,
156 dentry->d_name.name);
1da177e4 157
c2459dc4 158 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
1da177e4
LT
159 if ((file->f_mode & FMODE_WRITE) == 0)
160 return 0;
7b159fc1 161
14546c33
TM
162 /*
163 * If we're holding a write delegation, then just start the i/o
164 * but don't wait for completion (or send a commit).
165 */
011e2a7f 166 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
14546c33
TM
167 return filemap_fdatawrite(file->f_mapping);
168
7fe5c398 169 /* Flush writes to the server and return any errors */
af7fa165 170 return vfs_fsync(file, 0);
1da177e4 171}
89d77c8f 172EXPORT_SYMBOL_GPL(nfs_file_flush);
1da177e4 173
ce4ef7c0 174ssize_t
027445c3
BP
175nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
176 unsigned long nr_segs, loff_t pos)
1da177e4 177{
01cce933 178 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
1da177e4
LT
179 struct inode * inode = dentry->d_inode;
180 ssize_t result;
181
1da177e4 182 if (iocb->ki_filp->f_flags & O_DIRECT)
a564b8f0 183 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
1da177e4 184
6da24bc9 185 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
1da177e4 186 dentry->d_parent->d_name.name, dentry->d_name.name,
6f276e49 187 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
1da177e4 188
44b11874 189 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
4184dcf2 190 if (!result) {
027445c3 191 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
4184dcf2
CL
192 if (result > 0)
193 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
194 }
1da177e4
LT
195 return result;
196}
89d77c8f 197EXPORT_SYMBOL_GPL(nfs_file_read);
1da177e4 198
ce4ef7c0 199ssize_t
f0930fff
JA
200nfs_file_splice_read(struct file *filp, loff_t *ppos,
201 struct pipe_inode_info *pipe, size_t count,
202 unsigned int flags)
1da177e4 203{
01cce933 204 struct dentry *dentry = filp->f_path.dentry;
1da177e4
LT
205 struct inode *inode = dentry->d_inode;
206 ssize_t res;
207
6da24bc9 208 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
1da177e4
LT
209 dentry->d_parent->d_name.name, dentry->d_name.name,
210 (unsigned long) count, (unsigned long long) *ppos);
211
44b11874 212 res = nfs_revalidate_mapping(inode, filp->f_mapping);
aa2f1ef1 213 if (!res) {
f0930fff 214 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
aa2f1ef1
CL
215 if (res > 0)
216 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
217 }
1da177e4
LT
218 return res;
219}
89d77c8f 220EXPORT_SYMBOL_GPL(nfs_file_splice_read);
1da177e4 221
ce4ef7c0 222int
1da177e4
LT
223nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
224{
01cce933 225 struct dentry *dentry = file->f_path.dentry;
1da177e4
LT
226 struct inode *inode = dentry->d_inode;
227 int status;
228
6da24bc9 229 dprintk("NFS: mmap(%s/%s)\n",
1da177e4
LT
230 dentry->d_parent->d_name.name, dentry->d_name.name);
231
e1ebfd33
TM
232 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
233 * so we call that before revalidating the mapping
234 */
235 status = generic_file_mmap(file, vma);
94387fb1
TM
236 if (!status) {
237 vma->vm_ops = &nfs_file_vm_ops;
e1ebfd33 238 status = nfs_revalidate_mapping(inode, file->f_mapping);
94387fb1 239 }
1da177e4
LT
240 return status;
241}
89d77c8f 242EXPORT_SYMBOL_GPL(nfs_file_mmap);
1da177e4
LT
243
244/*
245 * Flush any dirty pages for this process, and check for write errors.
246 * The return status from this call provides a reliable indication of
247 * whether any write errors occurred for this process.
af7fa165
TM
248 *
249 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
250 * disk, but it retrieves and clears ctx->error after synching, despite
251 * the two being set at the same time in nfs_context_set_write_error().
252 * This is because the former is used to notify the _next_ call to
25985edc 253 * nfs_file_write() that a write error occurred, and hence cause it to
af7fa165 254 * fall back to doing a synchronous write.
1da177e4 255 */
ce4ef7c0 256int
a5c58892 257nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
1da177e4 258{
7ea80859 259 struct dentry *dentry = file->f_path.dentry;
cd3758e3 260 struct nfs_open_context *ctx = nfs_file_open_context(file);
1da177e4 261 struct inode *inode = dentry->d_inode;
05990d1b 262 int have_error, do_resend, status;
af7fa165
TM
263 int ret = 0;
264
6da24bc9 265 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
54917786
CL
266 dentry->d_parent->d_name.name, dentry->d_name.name,
267 datasync);
1da177e4 268
91d5b470 269 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
05990d1b 270 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
af7fa165
TM
271 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
272 status = nfs_commit_inode(inode, FLUSH_SYNC);
273 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
05990d1b 274 if (have_error) {
af7fa165 275 ret = xchg(&ctx->error, 0);
05990d1b
TM
276 if (ret)
277 goto out;
278 }
279 if (status < 0) {
af7fa165 280 ret = status;
05990d1b
TM
281 goto out;
282 }
283 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
284 if (do_resend)
285 ret = -EAGAIN;
286out:
a5c58892
BS
287 return ret;
288}
89d77c8f 289EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
a5c58892
BS
290
291static int
292nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
293{
294 int ret;
295 struct inode *inode = file->f_path.dentry->d_inode;
296
05990d1b
TM
297 do {
298 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
299 if (ret != 0)
300 break;
301 mutex_lock(&inode->i_mutex);
302 ret = nfs_file_fsync_commit(file, start, end, datasync);
303 mutex_unlock(&inode->i_mutex);
304 } while (ret == -EAGAIN);
305
af7fa165 306 return ret;
1da177e4
LT
307}
308
38c73044
PS
309/*
310 * Decide whether a read/modify/write cycle may be more efficient
311 * then a modify/write/read cycle when writing to a page in the
312 * page cache.
313 *
314 * The modify/write/read cycle may occur if a page is read before
315 * being completely filled by the writer. In this situation, the
316 * page must be completely written to stable storage on the server
317 * before it can be refilled by reading in the page from the server.
318 * This can lead to expensive, small, FILE_SYNC mode writes being
319 * done.
320 *
321 * It may be more efficient to read the page first if the file is
322 * open for reading in addition to writing, the page is not marked
323 * as Uptodate, it is not dirty or waiting to be committed,
324 * indicating that it was previously allocated and then modified,
325 * that there were valid bytes of data in that range of the file,
326 * and that the new data won't completely replace the old data in
327 * that range of the file.
328 */
329static int nfs_want_read_modify_write(struct file *file, struct page *page,
330 loff_t pos, unsigned len)
331{
332 unsigned int pglen = nfs_page_length(page);
333 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
334 unsigned int end = offset + len;
335
336 if ((file->f_mode & FMODE_READ) && /* open for read? */
337 !PageUptodate(page) && /* Uptodate? */
338 !PagePrivate(page) && /* i/o request already? */
339 pglen && /* valid bytes of file? */
340 (end < pglen || offset)) /* replace all valid bytes? */
341 return 1;
342 return 0;
343}
344
1da177e4 345/*
4899f9c8
NP
346 * This does the "real" work of the write. We must allocate and lock the
347 * page to be sent back to the generic routine, which then copies the
348 * data from user space.
1da177e4
LT
349 *
350 * If the writer ends up delaying the write, the writer needs to
351 * increment the page use counts until he is done with the page.
352 */
4899f9c8
NP
353static int nfs_write_begin(struct file *file, struct address_space *mapping,
354 loff_t pos, unsigned len, unsigned flags,
355 struct page **pagep, void **fsdata)
1da177e4 356{
4899f9c8 357 int ret;
38c73044 358 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
4899f9c8 359 struct page *page;
38c73044 360 int once_thru = 0;
4899f9c8 361
b7eaefaa
CL
362 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
363 file->f_path.dentry->d_parent->d_name.name,
364 file->f_path.dentry->d_name.name,
365 mapping->host->i_ino, len, (long long) pos);
366
38c73044 367start:
72cb77f4
TM
368 /*
369 * Prevent starvation issues if someone is doing a consistency
370 * sync-to-disk
371 */
372 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
373 nfs_wait_bit_killable, TASK_KILLABLE);
374 if (ret)
375 return ret;
376
54566b2c 377 page = grab_cache_page_write_begin(mapping, index, flags);
4899f9c8
NP
378 if (!page)
379 return -ENOMEM;
380 *pagep = page;
381
382 ret = nfs_flush_incompatible(file, page);
383 if (ret) {
384 unlock_page(page);
385 page_cache_release(page);
38c73044
PS
386 } else if (!once_thru &&
387 nfs_want_read_modify_write(file, page, pos, len)) {
388 once_thru = 1;
389 ret = nfs_readpage(file, page);
390 page_cache_release(page);
391 if (!ret)
392 goto start;
4899f9c8
NP
393 }
394 return ret;
1da177e4
LT
395}
396
4899f9c8
NP
397static int nfs_write_end(struct file *file, struct address_space *mapping,
398 loff_t pos, unsigned len, unsigned copied,
399 struct page *page, void *fsdata)
1da177e4 400{
4899f9c8
NP
401 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
402 int status;
1da177e4 403
b7eaefaa
CL
404 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
405 file->f_path.dentry->d_parent->d_name.name,
406 file->f_path.dentry->d_name.name,
407 mapping->host->i_ino, len, (long long) pos);
408
efc91ed0
TM
409 /*
410 * Zero any uninitialised parts of the page, and then mark the page
411 * as up to date if it turns out that we're extending the file.
412 */
413 if (!PageUptodate(page)) {
414 unsigned pglen = nfs_page_length(page);
415 unsigned end = offset + len;
416
417 if (pglen == 0) {
418 zero_user_segments(page, 0, offset,
419 end, PAGE_CACHE_SIZE);
420 SetPageUptodate(page);
421 } else if (end >= pglen) {
422 zero_user_segment(page, end, PAGE_CACHE_SIZE);
423 if (offset == 0)
424 SetPageUptodate(page);
425 } else
426 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
427 }
428
4899f9c8 429 status = nfs_updatepage(file, page, offset, copied);
4899f9c8
NP
430
431 unlock_page(page);
432 page_cache_release(page);
433
3d509e54
CL
434 if (status < 0)
435 return status;
2701d086 436 NFS_I(mapping->host)->write_io += copied;
3d509e54 437 return copied;
1da177e4
LT
438}
439
6b9b3514
DH
440/*
441 * Partially or wholly invalidate a page
442 * - Release the private state associated with a page if undergoing complete
443 * page invalidation
545db45f 444 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
445 * - Caller holds page lock
446 */
2ff28e22 447static void nfs_invalidate_page(struct page *page, unsigned long offset)
cd52ed35 448{
b7eaefaa
CL
449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
450
1c75950b
TM
451 if (offset != 0)
452 return;
d2ccddf0 453 /* Cancel any unstarted writes on this page */
d56b4ddf 454 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
545db45f
DH
455
456 nfs_fscache_invalidate_page(page, page->mapping->host);
cd52ed35
TM
457}
458
6b9b3514
DH
459/*
460 * Attempt to release the private state associated with a page
545db45f 461 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
462 * - Caller holds page lock
463 * - Return true (may release page) or false (may not)
464 */
cd52ed35
TM
465static int nfs_release_page(struct page *page, gfp_t gfp)
466{
b608b283
TM
467 struct address_space *mapping = page->mapping;
468
b7eaefaa
CL
469 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
470
5cf02d09
JL
471 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
472 * doing this memory reclaim for a fs-related allocation.
473 */
474 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
475 !(current->flags & PF_FSTRANS)) {
b608b283
TM
476 int how = FLUSH_SYNC;
477
478 /* Don't let kswapd deadlock waiting for OOM RPC calls */
479 if (current_is_kswapd())
480 how = 0;
481 nfs_commit_inode(mapping->host, how);
482 }
e3db7691 483 /* If PagePrivate() is set, then the page is not freeable */
545db45f
DH
484 if (PagePrivate(page))
485 return 0;
486 return nfs_fscache_release_page(page, gfp);
e3db7691
TM
487}
488
6b9b3514
DH
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
545db45f 493 * - Called if either PG_private or fscache is set on the page
6b9b3514
DH
494 * - Caller holds page lock
495 * - Return 0 if successful, -error otherwise
496 */
e3db7691
TM
497static int nfs_launder_page(struct page *page)
498{
d56b4ddf 499 struct inode *inode = page_file_mapping(page)->host;
545db45f 500 struct nfs_inode *nfsi = NFS_I(inode);
b7eaefaa
CL
501
502 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
503 inode->i_ino, (long long)page_offset(page));
504
545db45f 505 nfs_fscache_wait_on_page_write(nfsi, page);
b7eaefaa 506 return nfs_wb_page(inode, page);
cd52ed35
TM
507}
508
a564b8f0
MG
509#ifdef CONFIG_NFS_SWAP
510static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
511 sector_t *span)
512{
513 *span = sis->pages;
514 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
515}
516
517static void nfs_swap_deactivate(struct file *file)
518{
519 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
520}
521#endif
522
f5e54d6e 523const struct address_space_operations nfs_file_aops = {
1da177e4
LT
524 .readpage = nfs_readpage,
525 .readpages = nfs_readpages,
9cccef95 526 .set_page_dirty = __set_page_dirty_nobuffers,
1da177e4
LT
527 .writepage = nfs_writepage,
528 .writepages = nfs_writepages,
4899f9c8
NP
529 .write_begin = nfs_write_begin,
530 .write_end = nfs_write_end,
cd52ed35
TM
531 .invalidatepage = nfs_invalidate_page,
532 .releasepage = nfs_release_page,
1da177e4 533 .direct_IO = nfs_direct_IO,
074cc1de 534 .migratepage = nfs_migrate_page,
e3db7691 535 .launder_page = nfs_launder_page,
f590f333 536 .error_remove_page = generic_error_remove_page,
a564b8f0
MG
537#ifdef CONFIG_NFS_SWAP
538 .swap_activate = nfs_swap_activate,
539 .swap_deactivate = nfs_swap_deactivate,
540#endif
1da177e4
LT
541};
542
6b9b3514
DH
543/*
544 * Notification that a PTE pointing to an NFS page is about to be made
545 * writable, implying that someone is about to modify the page through a
546 * shared-writable mapping
547 */
c2ec175c 548static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
94387fb1 549{
c2ec175c 550 struct page *page = vmf->page;
94387fb1 551 struct file *filp = vma->vm_file;
b7eaefaa 552 struct dentry *dentry = filp->f_path.dentry;
94387fb1 553 unsigned pagelen;
bc4866b6 554 int ret = VM_FAULT_NOPAGE;
4899f9c8 555 struct address_space *mapping;
94387fb1 556
b7eaefaa
CL
557 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
558 dentry->d_parent->d_name.name, dentry->d_name.name,
559 filp->f_mapping->host->i_ino,
560 (long long)page_offset(page));
561
545db45f
DH
562 /* make sure the cache has finished storing the page */
563 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
564
94387fb1 565 lock_page(page);
d56b4ddf 566 mapping = page_file_mapping(page);
b7eaefaa 567 if (mapping != dentry->d_inode->i_mapping)
8b1f9ee5
TM
568 goto out_unlock;
569
2aeb98f4
TM
570 wait_on_page_writeback(page);
571
94387fb1 572 pagelen = nfs_page_length(page);
8b1f9ee5
TM
573 if (pagelen == 0)
574 goto out_unlock;
4899f9c8 575
bc4866b6
TM
576 ret = VM_FAULT_LOCKED;
577 if (nfs_flush_incompatible(filp, page) == 0 &&
578 nfs_updatepage(filp, page, 0, pagelen) == 0)
579 goto out;
8b1f9ee5 580
bc4866b6 581 ret = VM_FAULT_SIGBUS;
8b1f9ee5
TM
582out_unlock:
583 unlock_page(page);
bc4866b6
TM
584out:
585 return ret;
94387fb1
TM
586}
587
f0f37e2f 588static const struct vm_operations_struct nfs_file_vm_ops = {
94387fb1
TM
589 .fault = filemap_fault,
590 .page_mkwrite = nfs_vm_page_mkwrite,
591};
592
7b159fc1
TM
593static int nfs_need_sync_write(struct file *filp, struct inode *inode)
594{
595 struct nfs_open_context *ctx;
596
6b2f3d1f 597 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
7b159fc1 598 return 1;
cd3758e3 599 ctx = nfs_file_open_context(filp);
7b159fc1
TM
600 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
601 return 1;
602 return 0;
603}
604
ce4ef7c0
BS
605ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
606 unsigned long nr_segs, loff_t pos)
1da177e4 607{
01cce933 608 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
1da177e4 609 struct inode * inode = dentry->d_inode;
7e381172 610 unsigned long written = 0;
1da177e4 611 ssize_t result;
027445c3 612 size_t count = iov_length(iov, nr_segs);
1da177e4 613
1da177e4 614 if (iocb->ki_filp->f_flags & O_DIRECT)
a564b8f0 615 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
1da177e4 616
6da24bc9 617 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
1da177e4 618 dentry->d_parent->d_name.name, dentry->d_name.name,
6da24bc9 619 (unsigned long) count, (long long) pos);
1da177e4
LT
620
621 result = -EBUSY;
622 if (IS_SWAPFILE(inode))
623 goto out_swapfile;
7d52e862
TM
624 /*
625 * O_APPEND implies that we must revalidate the file length.
626 */
627 if (iocb->ki_filp->f_flags & O_APPEND) {
628 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
629 if (result)
630 goto out;
fe51beec 631 }
1da177e4
LT
632
633 result = count;
634 if (!count)
635 goto out;
636
027445c3 637 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
7e381172
CL
638 if (result > 0)
639 written = result;
640
6b2f3d1f 641 /* Return error values for O_DSYNC and IS_SYNC() */
7b159fc1 642 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
af7fa165 643 int err = vfs_fsync(iocb->ki_filp, 0);
200baa21
TM
644 if (err < 0)
645 result = err;
646 }
7e381172
CL
647 if (result > 0)
648 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
1da177e4
LT
649out:
650 return result;
651
652out_swapfile:
653 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
654 goto out;
655}
89d77c8f 656EXPORT_SYMBOL_GPL(nfs_file_write);
1da177e4 657
ce4ef7c0
BS
658ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
659 struct file *filp, loff_t *ppos,
660 size_t count, unsigned int flags)
bf40d343
SJ
661{
662 struct dentry *dentry = filp->f_path.dentry;
663 struct inode *inode = dentry->d_inode;
7e381172 664 unsigned long written = 0;
bf40d343
SJ
665 ssize_t ret;
666
667 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
668 dentry->d_parent->d_name.name, dentry->d_name.name,
669 (unsigned long) count, (unsigned long long) *ppos);
670
671 /*
672 * The combination of splice and an O_APPEND destination is disallowed.
673 */
674
bf40d343 675 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
7e381172
CL
676 if (ret > 0)
677 written = ret;
678
bf40d343 679 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
af7fa165 680 int err = vfs_fsync(filp, 0);
bf40d343
SJ
681 if (err < 0)
682 ret = err;
683 }
7e381172
CL
684 if (ret > 0)
685 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
bf40d343
SJ
686 return ret;
687}
89d77c8f 688EXPORT_SYMBOL_GPL(nfs_file_splice_write);
bf40d343 689
5eebde23
SJ
690static int
691do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
692{
693 struct inode *inode = filp->f_mapping->host;
694 int status = 0;
21ac19d4 695 unsigned int saved_type = fl->fl_type;
1da177e4 696
039c4d7a 697 /* Try local locking first */
6d34ac19
BF
698 posix_test_lock(filp, fl);
699 if (fl->fl_type != F_UNLCK) {
700 /* found a conflict */
039c4d7a 701 goto out;
1da177e4 702 }
21ac19d4 703 fl->fl_type = saved_type;
039c4d7a 704
011e2a7f 705 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
039c4d7a
TM
706 goto out_noconflict;
707
5eebde23 708 if (is_local)
039c4d7a
TM
709 goto out_noconflict;
710
711 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
712out:
1da177e4 713 return status;
039c4d7a
TM
714out_noconflict:
715 fl->fl_type = F_UNLCK;
716 goto out;
1da177e4
LT
717}
718
719static int do_vfs_lock(struct file *file, struct file_lock *fl)
720{
721 int res = 0;
722 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
723 case FL_POSIX:
724 res = posix_lock_file_wait(file, fl);
725 break;
726 case FL_FLOCK:
727 res = flock_lock_file_wait(file, fl);
728 break;
729 default:
730 BUG();
731 }
1da177e4
LT
732 return res;
733}
734
5eebde23
SJ
735static int
736do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
737{
738 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
739 int status;
740
1da177e4
LT
741 /*
742 * Flush all pending writes before doing anything
743 * with locks..
744 */
29884df0 745 nfs_sync_mapping(filp->f_mapping);
1da177e4
LT
746
747 /* NOTE: special case
748 * If we're signalled while cleaning up locks on process exit, we
749 * still need to complete the unlock.
750 */
5eebde23
SJ
751 /*
752 * Use local locking if mounted with "-onolock" or with appropriate
753 * "-olocal_lock="
754 */
755 if (!is_local)
1da177e4
LT
756 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
757 else
758 status = do_vfs_lock(filp, fl);
1da177e4
LT
759 return status;
760}
761
6b96724e
RL
762static int
763is_time_granular(struct timespec *ts) {
764 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
765}
766
5eebde23
SJ
767static int
768do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
769{
770 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
771 int status;
772
1da177e4
LT
773 /*
774 * Flush all pending writes before doing anything
775 * with locks..
776 */
29884df0
TM
777 status = nfs_sync_mapping(filp->f_mapping);
778 if (status != 0)
1da177e4
LT
779 goto out;
780
5eebde23
SJ
781 /*
782 * Use local locking if mounted with "-onolock" or with appropriate
783 * "-olocal_lock="
784 */
785 if (!is_local)
1da177e4 786 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
c4d7c402 787 else
1da177e4 788 status = do_vfs_lock(filp, fl);
1da177e4
LT
789 if (status < 0)
790 goto out;
6b96724e 791
1da177e4 792 /*
6b96724e
RL
793 * Revalidate the cache if the server has time stamps granular
794 * enough to detect subsecond changes. Otherwise, clear the
795 * cache to prevent missing any changes.
796 *
1da177e4
LT
797 * This makes locking act as a cache coherency point.
798 */
29884df0 799 nfs_sync_mapping(filp->f_mapping);
011e2a7f 800 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
6b96724e
RL
801 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
802 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
803 else
804 nfs_zap_caches(inode);
805 }
1da177e4 806out:
1da177e4
LT
807 return status;
808}
809
810/*
811 * Lock a (portion of) a file
812 */
ce4ef7c0 813int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 814{
6da24bc9 815 struct inode *inode = filp->f_mapping->host;
2116271a 816 int ret = -ENOLCK;
5eebde23 817 int is_local = 0;
1da177e4 818
6da24bc9
CL
819 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
820 filp->f_path.dentry->d_parent->d_name.name,
821 filp->f_path.dentry->d_name.name,
1da177e4
LT
822 fl->fl_type, fl->fl_flags,
823 (long long)fl->fl_start, (long long)fl->fl_end);
6da24bc9 824
91d5b470 825 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
1da177e4
LT
826
827 /* No mandatory locks over NFS */
dfad9441 828 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
2116271a
TM
829 goto out_err;
830
5eebde23
SJ
831 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
832 is_local = 1;
833
2116271a
TM
834 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
835 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
836 if (ret < 0)
837 goto out_err;
838 }
1da177e4
LT
839
840 if (IS_GETLK(cmd))
5eebde23 841 ret = do_getlk(filp, cmd, fl, is_local);
2116271a 842 else if (fl->fl_type == F_UNLCK)
5eebde23 843 ret = do_unlk(filp, cmd, fl, is_local);
2116271a 844 else
5eebde23 845 ret = do_setlk(filp, cmd, fl, is_local);
2116271a
TM
846out_err:
847 return ret;
1da177e4 848}
89d77c8f 849EXPORT_SYMBOL_GPL(nfs_lock);
1da177e4
LT
850
851/*
852 * Lock a (portion of) a file
853 */
ce4ef7c0 854int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 855{
5eebde23
SJ
856 struct inode *inode = filp->f_mapping->host;
857 int is_local = 0;
858
6da24bc9
CL
859 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
860 filp->f_path.dentry->d_parent->d_name.name,
861 filp->f_path.dentry->d_name.name,
1da177e4
LT
862 fl->fl_type, fl->fl_flags);
863
1da177e4
LT
864 if (!(fl->fl_flags & FL_FLOCK))
865 return -ENOLCK;
866
ad0fcd4e
JL
867 /*
868 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
869 * any standard. In principle we might be able to support LOCK_MAND
870 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
871 * NFS code is not set up for it.
872 */
873 if (fl->fl_type & LOCK_MAND)
874 return -EINVAL;
875
5eebde23
SJ
876 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
877 is_local = 1;
878
1da177e4
LT
879 /* We're simulating flock() locks using posix locks on the server */
880 fl->fl_owner = (fl_owner_t)filp;
881 fl->fl_start = 0;
882 fl->fl_end = OFFSET_MAX;
883
884 if (fl->fl_type == F_UNLCK)
5eebde23
SJ
885 return do_unlk(filp, cmd, fl, is_local);
886 return do_setlk(filp, cmd, fl, is_local);
1da177e4 887}
89d77c8f 888EXPORT_SYMBOL_GPL(nfs_flock);
370f6599 889
6da24bc9
CL
890/*
891 * There is no protocol support for leases, so we have no way to implement
892 * them correctly in the face of opens by other clients.
893 */
ce4ef7c0 894int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
370f6599 895{
6da24bc9
CL
896 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
897 file->f_path.dentry->d_parent->d_name.name,
898 file->f_path.dentry->d_name.name, arg);
370f6599
BF
899 return -EINVAL;
900}
89d77c8f 901EXPORT_SYMBOL_GPL(nfs_setlease);
1788ea6e 902
0486958f
JL
903const struct file_operations nfs_file_operations = {
904 .llseek = nfs_file_llseek,
905 .read = do_sync_read,
906 .write = do_sync_write,
907 .aio_read = nfs_file_read,
908 .aio_write = nfs_file_write,
909 .mmap = nfs_file_mmap,
910 .open = nfs_file_open,
911 .flush = nfs_file_flush,
912 .release = nfs_file_release,
913 .fsync = nfs_file_fsync,
914 .lock = nfs_lock,
915 .flock = nfs_flock,
916 .splice_read = nfs_file_splice_read,
917 .splice_write = nfs_file_splice_write,
918 .check_flags = nfs_check_flags,
919 .setlease = nfs_setlease,
920};
ddda8e0a 921EXPORT_SYMBOL_GPL(nfs_file_operations);