]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/nfs/file.c
ext4: replace cut'n'pasted llseek code with generic_file_llseek_size
[mirror_ubuntu-zesty-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
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/fcntl.h>
23#include <linux/stat.h>
24#include <linux/nfs_fs.h>
25#include <linux/nfs_mount.h>
26#include <linux/mm.h>
1da177e4 27#include <linux/pagemap.h>
e8edc6e0 28#include <linux/aio.h>
5a0e3ad6 29#include <linux/gfp.h>
b608b283 30#include <linux/swap.h>
1da177e4
LT
31
32#include <asm/uaccess.h>
33#include <asm/system.h>
34
35#include "delegation.h"
94387fb1 36#include "internal.h"
91d5b470 37#include "iostat.h"
545db45f 38#include "fscache.h"
e5e94017 39#include "pnfs.h"
1da177e4
LT
40
41#define NFSDBG_FACILITY NFSDBG_FILE
42
43static int nfs_file_open(struct inode *, struct file *);
44static int nfs_file_release(struct inode *, struct file *);
980802e3 45static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
1da177e4 46static int nfs_file_mmap(struct file *, struct vm_area_struct *);
f0930fff
JA
47static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
48 struct pipe_inode_info *pipe,
49 size_t count, unsigned int flags);
027445c3
BP
50static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
51 unsigned long nr_segs, loff_t pos);
bf40d343
SJ
52static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
53 struct file *filp, loff_t *ppos,
54 size_t count, unsigned int flags);
027445c3
BP
55static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
56 unsigned long nr_segs, loff_t pos);
75e1fcc0 57static int nfs_file_flush(struct file *, fl_owner_t id);
02c24a82 58static int nfs_file_fsync(struct file *, loff_t, loff_t, int datasync);
1da177e4
LT
59static int nfs_check_flags(int flags);
60static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
61static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
370f6599 62static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
1da177e4 63
f0f37e2f 64static const struct vm_operations_struct nfs_file_vm_ops;
94387fb1 65
4b6f5d20 66const struct file_operations nfs_file_operations = {
980802e3 67 .llseek = nfs_file_llseek,
1da177e4
LT
68 .read = do_sync_read,
69 .write = do_sync_write,
027445c3
BP
70 .aio_read = nfs_file_read,
71 .aio_write = nfs_file_write,
1da177e4
LT
72 .mmap = nfs_file_mmap,
73 .open = nfs_file_open,
74 .flush = nfs_file_flush,
75 .release = nfs_file_release,
54917786 76 .fsync = nfs_file_fsync,
1da177e4
LT
77 .lock = nfs_lock,
78 .flock = nfs_flock,
f0930fff 79 .splice_read = nfs_file_splice_read,
bf40d343 80 .splice_write = nfs_file_splice_write,
1da177e4 81 .check_flags = nfs_check_flags,
370f6599 82 .setlease = nfs_setlease,
1da177e4
LT
83};
84
92e1d5be 85const struct inode_operations nfs_file_inode_operations = {
1da177e4
LT
86 .permission = nfs_permission,
87 .getattr = nfs_getattr,
88 .setattr = nfs_setattr,
89};
90
b7fa0554 91#ifdef CONFIG_NFS_V3
92e1d5be 92const struct inode_operations nfs3_file_inode_operations = {
b7fa0554
AG
93 .permission = nfs_permission,
94 .getattr = nfs_getattr,
95 .setattr = nfs_setattr,
96 .listxattr = nfs3_listxattr,
97 .getxattr = nfs3_getxattr,
98 .setxattr = nfs3_setxattr,
99 .removexattr = nfs3_removexattr,
100};
101#endif /* CONFIG_NFS_v3 */
102
1da177e4
LT
103/* Hack for future NFS swap support */
104#ifndef IS_SWAPFILE
105# define IS_SWAPFILE(inode) (0)
106#endif
107
108static int nfs_check_flags(int flags)
109{
110 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
111 return -EINVAL;
112
113 return 0;
114}
115
116/*
117 * Open file
118 */
119static int
120nfs_file_open(struct inode *inode, struct file *filp)
121{
1da177e4
LT
122 int res;
123
6da24bc9 124 dprintk("NFS: open file(%s/%s)\n",
cc0dd2d1
CL
125 filp->f_path.dentry->d_parent->d_name.name,
126 filp->f_path.dentry->d_name.name);
127
c2459dc4 128 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
1da177e4
LT
129 res = nfs_check_flags(filp->f_flags);
130 if (res)
131 return res;
132
46cb650c 133 res = nfs_open(inode, filp);
1da177e4
LT
134 return res;
135}
136
137static int
138nfs_file_release(struct inode *inode, struct file *filp)
139{
6da24bc9
CL
140 struct dentry *dentry = filp->f_path.dentry;
141
142 dprintk("NFS: release(%s/%s)\n",
143 dentry->d_parent->d_name.name,
144 dentry->d_name.name);
145
91d5b470 146 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
46cb650c 147 return nfs_release(inode, filp);
1da177e4
LT
148}
149
980802e3
TM
150/**
151 * nfs_revalidate_size - Revalidate the file size
152 * @inode - pointer to inode struct
153 * @file - pointer to struct file
154 *
155 * Revalidates the file length. This is basically a wrapper around
156 * nfs_revalidate_inode() that takes into account the fact that we may
157 * have cached writes (in which case we don't care about the server's
158 * idea of what the file length is), or O_DIRECT (in which case we
159 * shouldn't trust the cache).
160 */
161static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
162{
163 struct nfs_server *server = NFS_SERVER(inode);
164 struct nfs_inode *nfsi = NFS_I(inode);
165
d7cf8dd0
TM
166 if (nfs_have_delegated_attributes(inode))
167 goto out_noreval;
168
980802e3
TM
169 if (filp->f_flags & O_DIRECT)
170 goto force_reval;
d7cf8dd0
TM
171 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
172 goto force_reval;
173 if (nfs_attribute_timeout(inode))
174 goto force_reval;
175out_noreval:
176 return 0;
980802e3
TM
177force_reval:
178 return __nfs_revalidate_inode(server, inode);
179}
180
181static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
182{
9465efc9 183 loff_t loff;
e89e896d 184
6da24bc9 185 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
b84e06c5
CL
186 filp->f_path.dentry->d_parent->d_name.name,
187 filp->f_path.dentry->d_name.name,
188 offset, origin);
189
06222e49
JB
190 /*
191 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
192 * the cached file length
193 */
194 if (origin != SEEK_SET || origin != SEEK_CUR) {
980802e3 195 struct inode *inode = filp->f_mapping->host;
d5e66348 196
980802e3
TM
197 int retval = nfs_revalidate_file_size(inode, filp);
198 if (retval < 0)
199 return (loff_t)retval;
d5e66348 200
ef3d0fd2 201 /* AK: should drop this lock. Unlikely to be needed. */
d5e66348 202 spin_lock(&inode->i_lock);
ef3d0fd2 203 loff = generic_file_llseek(filp, offset, origin);
d5e66348
TM
204 spin_unlock(&inode->i_lock);
205 } else
ef3d0fd2 206 loff = generic_file_llseek(filp, offset, origin);
9465efc9 207 return loff;
980802e3
TM
208}
209
1da177e4
LT
210/*
211 * Flush all dirty pages, and check for write errors.
1da177e4
LT
212 */
213static int
75e1fcc0 214nfs_file_flush(struct file *file, fl_owner_t id)
1da177e4 215{
6da24bc9
CL
216 struct dentry *dentry = file->f_path.dentry;
217 struct inode *inode = dentry->d_inode;
1da177e4 218
6da24bc9
CL
219 dprintk("NFS: flush(%s/%s)\n",
220 dentry->d_parent->d_name.name,
221 dentry->d_name.name);
1da177e4 222
c2459dc4 223 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
1da177e4
LT
224 if ((file->f_mode & FMODE_WRITE) == 0)
225 return 0;
7b159fc1 226
7fe5c398 227 /* Flush writes to the server and return any errors */
af7fa165 228 return vfs_fsync(file, 0);
1da177e4
LT
229}
230
231static ssize_t
027445c3
BP
232nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
233 unsigned long nr_segs, loff_t pos)
1da177e4 234{
01cce933 235 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
1da177e4
LT
236 struct inode * inode = dentry->d_inode;
237 ssize_t result;
027445c3 238 size_t count = iov_length(iov, nr_segs);
1da177e4 239
1da177e4 240 if (iocb->ki_filp->f_flags & O_DIRECT)
027445c3 241 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
1da177e4 242
6da24bc9 243 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
1da177e4
LT
244 dentry->d_parent->d_name.name, dentry->d_name.name,
245 (unsigned long) count, (unsigned long) pos);
246
44b11874 247 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
4184dcf2 248 if (!result) {
027445c3 249 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
4184dcf2
CL
250 if (result > 0)
251 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
252 }
1da177e4
LT
253 return result;
254}
255
256static ssize_t
f0930fff
JA
257nfs_file_splice_read(struct file *filp, loff_t *ppos,
258 struct pipe_inode_info *pipe, size_t count,
259 unsigned int flags)
1da177e4 260{
01cce933 261 struct dentry *dentry = filp->f_path.dentry;
1da177e4
LT
262 struct inode *inode = dentry->d_inode;
263 ssize_t res;
264
6da24bc9 265 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
1da177e4
LT
266 dentry->d_parent->d_name.name, dentry->d_name.name,
267 (unsigned long) count, (unsigned long long) *ppos);
268
44b11874 269 res = nfs_revalidate_mapping(inode, filp->f_mapping);
aa2f1ef1 270 if (!res) {
f0930fff 271 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
aa2f1ef1
CL
272 if (res > 0)
273 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
274 }
1da177e4
LT
275 return res;
276}
277
278static int
279nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
280{
01cce933 281 struct dentry *dentry = file->f_path.dentry;
1da177e4
LT
282 struct inode *inode = dentry->d_inode;
283 int status;
284
6da24bc9 285 dprintk("NFS: mmap(%s/%s)\n",
1da177e4
LT
286 dentry->d_parent->d_name.name, dentry->d_name.name);
287
e1ebfd33
TM
288 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
289 * so we call that before revalidating the mapping
290 */
291 status = generic_file_mmap(file, vma);
94387fb1
TM
292 if (!status) {
293 vma->vm_ops = &nfs_file_vm_ops;
e1ebfd33 294 status = nfs_revalidate_mapping(inode, file->f_mapping);
94387fb1 295 }
1da177e4
LT
296 return status;
297}
298
299/*
300 * Flush any dirty pages for this process, and check for write errors.
301 * The return status from this call provides a reliable indication of
302 * whether any write errors occurred for this process.
af7fa165
TM
303 *
304 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
305 * disk, but it retrieves and clears ctx->error after synching, despite
306 * the two being set at the same time in nfs_context_set_write_error().
307 * This is because the former is used to notify the _next_ call to
25985edc 308 * nfs_file_write() that a write error occurred, and hence cause it to
af7fa165 309 * fall back to doing a synchronous write.
1da177e4
LT
310 */
311static int
02c24a82 312nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1da177e4 313{
7ea80859 314 struct dentry *dentry = file->f_path.dentry;
cd3758e3 315 struct nfs_open_context *ctx = nfs_file_open_context(file);
1da177e4 316 struct inode *inode = dentry->d_inode;
af7fa165
TM
317 int have_error, status;
318 int ret = 0;
319
6da24bc9 320 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
54917786
CL
321 dentry->d_parent->d_name.name, dentry->d_name.name,
322 datasync);
1da177e4 323
02c24a82
JB
324 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
325 if (ret)
326 return ret;
327 mutex_lock(&inode->i_mutex);
328
91d5b470 329 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
af7fa165
TM
330 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
331 status = nfs_commit_inode(inode, FLUSH_SYNC);
332 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
333 if (have_error)
334 ret = xchg(&ctx->error, 0);
0702099b 335 if (!ret && status < 0)
af7fa165 336 ret = status;
863a3c6c
AA
337 if (!ret && !datasync)
338 /* application has asked for meta-data sync */
ef311537 339 ret = pnfs_layoutcommit_inode(inode, true);
02c24a82 340 mutex_unlock(&inode->i_mutex);
af7fa165 341 return ret;
1da177e4
LT
342}
343
38c73044
PS
344/*
345 * Decide whether a read/modify/write cycle may be more efficient
346 * then a modify/write/read cycle when writing to a page in the
347 * page cache.
348 *
349 * The modify/write/read cycle may occur if a page is read before
350 * being completely filled by the writer. In this situation, the
351 * page must be completely written to stable storage on the server
352 * before it can be refilled by reading in the page from the server.
353 * This can lead to expensive, small, FILE_SYNC mode writes being
354 * done.
355 *
356 * It may be more efficient to read the page first if the file is
357 * open for reading in addition to writing, the page is not marked
358 * as Uptodate, it is not dirty or waiting to be committed,
359 * indicating that it was previously allocated and then modified,
360 * that there were valid bytes of data in that range of the file,
361 * and that the new data won't completely replace the old data in
362 * that range of the file.
363 */
364static int nfs_want_read_modify_write(struct file *file, struct page *page,
365 loff_t pos, unsigned len)
366{
367 unsigned int pglen = nfs_page_length(page);
368 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
369 unsigned int end = offset + len;
370
371 if ((file->f_mode & FMODE_READ) && /* open for read? */
372 !PageUptodate(page) && /* Uptodate? */
373 !PagePrivate(page) && /* i/o request already? */
374 pglen && /* valid bytes of file? */
375 (end < pglen || offset)) /* replace all valid bytes? */
376 return 1;
377 return 0;
378}
379
1da177e4 380/*
4899f9c8
NP
381 * This does the "real" work of the write. We must allocate and lock the
382 * page to be sent back to the generic routine, which then copies the
383 * data from user space.
1da177e4
LT
384 *
385 * If the writer ends up delaying the write, the writer needs to
386 * increment the page use counts until he is done with the page.
387 */
4899f9c8
NP
388static int nfs_write_begin(struct file *file, struct address_space *mapping,
389 loff_t pos, unsigned len, unsigned flags,
390 struct page **pagep, void **fsdata)
1da177e4 391{
4899f9c8 392 int ret;
38c73044 393 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
4899f9c8 394 struct page *page;
38c73044 395 int once_thru = 0;
4899f9c8 396
b7eaefaa
CL
397 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
398 file->f_path.dentry->d_parent->d_name.name,
399 file->f_path.dentry->d_name.name,
400 mapping->host->i_ino, len, (long long) pos);
401
38c73044 402start:
72cb77f4
TM
403 /*
404 * Prevent starvation issues if someone is doing a consistency
405 * sync-to-disk
406 */
407 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
408 nfs_wait_bit_killable, TASK_KILLABLE);
409 if (ret)
410 return ret;
411
54566b2c 412 page = grab_cache_page_write_begin(mapping, index, flags);
4899f9c8
NP
413 if (!page)
414 return -ENOMEM;
415 *pagep = page;
416
417 ret = nfs_flush_incompatible(file, page);
418 if (ret) {
419 unlock_page(page);
420 page_cache_release(page);
38c73044
PS
421 } else if (!once_thru &&
422 nfs_want_read_modify_write(file, page, pos, len)) {
423 once_thru = 1;
424 ret = nfs_readpage(file, page);
425 page_cache_release(page);
426 if (!ret)
427 goto start;
4899f9c8
NP
428 }
429 return ret;
1da177e4
LT
430}
431
4899f9c8
NP
432static int nfs_write_end(struct file *file, struct address_space *mapping,
433 loff_t pos, unsigned len, unsigned copied,
434 struct page *page, void *fsdata)
1da177e4 435{
4899f9c8
NP
436 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
437 int status;
1da177e4 438
b7eaefaa
CL
439 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
440 file->f_path.dentry->d_parent->d_name.name,
441 file->f_path.dentry->d_name.name,
442 mapping->host->i_ino, len, (long long) pos);
443
efc91ed0
TM
444 /*
445 * Zero any uninitialised parts of the page, and then mark the page
446 * as up to date if it turns out that we're extending the file.
447 */
448 if (!PageUptodate(page)) {
449 unsigned pglen = nfs_page_length(page);
450 unsigned end = offset + len;
451
452 if (pglen == 0) {
453 zero_user_segments(page, 0, offset,
454 end, PAGE_CACHE_SIZE);
455 SetPageUptodate(page);
456 } else if (end >= pglen) {
457 zero_user_segment(page, end, PAGE_CACHE_SIZE);
458 if (offset == 0)
459 SetPageUptodate(page);
460 } else
461 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
462 }
463
4899f9c8 464 status = nfs_updatepage(file, page, offset, copied);
4899f9c8
NP
465
466 unlock_page(page);
467 page_cache_release(page);
468
3d509e54
CL
469 if (status < 0)
470 return status;
471 return copied;
1da177e4
LT
472}
473
6b9b3514
DH
474/*
475 * Partially or wholly invalidate a page
476 * - Release the private state associated with a page if undergoing complete
477 * page invalidation
545db45f 478 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
479 * - Caller holds page lock
480 */
2ff28e22 481static void nfs_invalidate_page(struct page *page, unsigned long offset)
cd52ed35 482{
b7eaefaa
CL
483 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
484
1c75950b
TM
485 if (offset != 0)
486 return;
d2ccddf0 487 /* Cancel any unstarted writes on this page */
1b3b4a1a 488 nfs_wb_page_cancel(page->mapping->host, page);
545db45f
DH
489
490 nfs_fscache_invalidate_page(page, page->mapping->host);
cd52ed35
TM
491}
492
6b9b3514
DH
493/*
494 * Attempt to release the private state associated with a page
545db45f 495 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
496 * - Caller holds page lock
497 * - Return true (may release page) or false (may not)
498 */
cd52ed35
TM
499static int nfs_release_page(struct page *page, gfp_t gfp)
500{
b608b283
TM
501 struct address_space *mapping = page->mapping;
502
b7eaefaa
CL
503 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
504
d812e575 505 /* Only do I/O if gfp is a superset of GFP_KERNEL */
b608b283
TM
506 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL) {
507 int how = FLUSH_SYNC;
508
509 /* Don't let kswapd deadlock waiting for OOM RPC calls */
510 if (current_is_kswapd())
511 how = 0;
512 nfs_commit_inode(mapping->host, how);
513 }
e3db7691 514 /* If PagePrivate() is set, then the page is not freeable */
545db45f
DH
515 if (PagePrivate(page))
516 return 0;
517 return nfs_fscache_release_page(page, gfp);
e3db7691
TM
518}
519
6b9b3514
DH
520/*
521 * Attempt to clear the private state associated with a page when an error
522 * occurs that requires the cached contents of an inode to be written back or
523 * destroyed
545db45f 524 * - Called if either PG_private or fscache is set on the page
6b9b3514
DH
525 * - Caller holds page lock
526 * - Return 0 if successful, -error otherwise
527 */
e3db7691
TM
528static int nfs_launder_page(struct page *page)
529{
b7eaefaa 530 struct inode *inode = page->mapping->host;
545db45f 531 struct nfs_inode *nfsi = NFS_I(inode);
b7eaefaa
CL
532
533 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
534 inode->i_ino, (long long)page_offset(page));
535
545db45f 536 nfs_fscache_wait_on_page_write(nfsi, page);
b7eaefaa 537 return nfs_wb_page(inode, page);
cd52ed35
TM
538}
539
f5e54d6e 540const struct address_space_operations nfs_file_aops = {
1da177e4
LT
541 .readpage = nfs_readpage,
542 .readpages = nfs_readpages,
9cccef95 543 .set_page_dirty = __set_page_dirty_nobuffers,
1da177e4
LT
544 .writepage = nfs_writepage,
545 .writepages = nfs_writepages,
4899f9c8
NP
546 .write_begin = nfs_write_begin,
547 .write_end = nfs_write_end,
cd52ed35
TM
548 .invalidatepage = nfs_invalidate_page,
549 .releasepage = nfs_release_page,
1da177e4 550 .direct_IO = nfs_direct_IO,
074cc1de 551 .migratepage = nfs_migrate_page,
e3db7691 552 .launder_page = nfs_launder_page,
f590f333 553 .error_remove_page = generic_error_remove_page,
1da177e4
LT
554};
555
6b9b3514
DH
556/*
557 * Notification that a PTE pointing to an NFS page is about to be made
558 * writable, implying that someone is about to modify the page through a
559 * shared-writable mapping
560 */
c2ec175c 561static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
94387fb1 562{
c2ec175c 563 struct page *page = vmf->page;
94387fb1 564 struct file *filp = vma->vm_file;
b7eaefaa 565 struct dentry *dentry = filp->f_path.dentry;
94387fb1 566 unsigned pagelen;
bc4866b6 567 int ret = VM_FAULT_NOPAGE;
4899f9c8 568 struct address_space *mapping;
94387fb1 569
b7eaefaa
CL
570 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
571 dentry->d_parent->d_name.name, dentry->d_name.name,
572 filp->f_mapping->host->i_ino,
573 (long long)page_offset(page));
574
545db45f
DH
575 /* make sure the cache has finished storing the page */
576 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
577
94387fb1 578 lock_page(page);
4899f9c8 579 mapping = page->mapping;
b7eaefaa 580 if (mapping != dentry->d_inode->i_mapping)
8b1f9ee5
TM
581 goto out_unlock;
582
94387fb1 583 pagelen = nfs_page_length(page);
8b1f9ee5
TM
584 if (pagelen == 0)
585 goto out_unlock;
4899f9c8 586
bc4866b6
TM
587 ret = VM_FAULT_LOCKED;
588 if (nfs_flush_incompatible(filp, page) == 0 &&
589 nfs_updatepage(filp, page, 0, pagelen) == 0)
590 goto out;
8b1f9ee5 591
bc4866b6 592 ret = VM_FAULT_SIGBUS;
8b1f9ee5
TM
593out_unlock:
594 unlock_page(page);
bc4866b6
TM
595out:
596 return ret;
94387fb1
TM
597}
598
f0f37e2f 599static const struct vm_operations_struct nfs_file_vm_ops = {
94387fb1
TM
600 .fault = filemap_fault,
601 .page_mkwrite = nfs_vm_page_mkwrite,
602};
603
7b159fc1
TM
604static int nfs_need_sync_write(struct file *filp, struct inode *inode)
605{
606 struct nfs_open_context *ctx;
607
6b2f3d1f 608 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
7b159fc1 609 return 1;
cd3758e3 610 ctx = nfs_file_open_context(filp);
7b159fc1
TM
611 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
612 return 1;
613 return 0;
614}
615
027445c3
BP
616static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
617 unsigned long nr_segs, loff_t pos)
1da177e4 618{
01cce933 619 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
1da177e4 620 struct inode * inode = dentry->d_inode;
7e381172 621 unsigned long written = 0;
1da177e4 622 ssize_t result;
027445c3 623 size_t count = iov_length(iov, nr_segs);
1da177e4 624
1da177e4 625 if (iocb->ki_filp->f_flags & O_DIRECT)
027445c3 626 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
1da177e4 627
6da24bc9 628 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
1da177e4 629 dentry->d_parent->d_name.name, dentry->d_name.name,
6da24bc9 630 (unsigned long) count, (long long) pos);
1da177e4
LT
631
632 result = -EBUSY;
633 if (IS_SWAPFILE(inode))
634 goto out_swapfile;
7d52e862
TM
635 /*
636 * O_APPEND implies that we must revalidate the file length.
637 */
638 if (iocb->ki_filp->f_flags & O_APPEND) {
639 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
640 if (result)
641 goto out;
fe51beec 642 }
1da177e4
LT
643
644 result = count;
645 if (!count)
646 goto out;
647
027445c3 648 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
7e381172
CL
649 if (result > 0)
650 written = result;
651
6b2f3d1f 652 /* Return error values for O_DSYNC and IS_SYNC() */
7b159fc1 653 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
af7fa165 654 int err = vfs_fsync(iocb->ki_filp, 0);
200baa21
TM
655 if (err < 0)
656 result = err;
657 }
7e381172
CL
658 if (result > 0)
659 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
1da177e4
LT
660out:
661 return result;
662
663out_swapfile:
664 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
665 goto out;
666}
667
bf40d343
SJ
668static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
669 struct file *filp, loff_t *ppos,
670 size_t count, unsigned int flags)
671{
672 struct dentry *dentry = filp->f_path.dentry;
673 struct inode *inode = dentry->d_inode;
7e381172 674 unsigned long written = 0;
bf40d343
SJ
675 ssize_t ret;
676
677 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
678 dentry->d_parent->d_name.name, dentry->d_name.name,
679 (unsigned long) count, (unsigned long long) *ppos);
680
681 /*
682 * The combination of splice and an O_APPEND destination is disallowed.
683 */
684
bf40d343 685 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
7e381172
CL
686 if (ret > 0)
687 written = ret;
688
bf40d343 689 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
af7fa165 690 int err = vfs_fsync(filp, 0);
bf40d343
SJ
691 if (err < 0)
692 ret = err;
693 }
7e381172
CL
694 if (ret > 0)
695 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
bf40d343
SJ
696 return ret;
697}
698
5eebde23
SJ
699static int
700do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
701{
702 struct inode *inode = filp->f_mapping->host;
703 int status = 0;
21ac19d4 704 unsigned int saved_type = fl->fl_type;
1da177e4 705
039c4d7a 706 /* Try local locking first */
6d34ac19
BF
707 posix_test_lock(filp, fl);
708 if (fl->fl_type != F_UNLCK) {
709 /* found a conflict */
039c4d7a 710 goto out;
1da177e4 711 }
21ac19d4 712 fl->fl_type = saved_type;
039c4d7a
TM
713
714 if (nfs_have_delegation(inode, FMODE_READ))
715 goto out_noconflict;
716
5eebde23 717 if (is_local)
039c4d7a
TM
718 goto out_noconflict;
719
720 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
721out:
1da177e4 722 return status;
039c4d7a
TM
723out_noconflict:
724 fl->fl_type = F_UNLCK;
725 goto out;
1da177e4
LT
726}
727
728static int do_vfs_lock(struct file *file, struct file_lock *fl)
729{
730 int res = 0;
731 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
732 case FL_POSIX:
733 res = posix_lock_file_wait(file, fl);
734 break;
735 case FL_FLOCK:
736 res = flock_lock_file_wait(file, fl);
737 break;
738 default:
739 BUG();
740 }
1da177e4
LT
741 return res;
742}
743
5eebde23
SJ
744static int
745do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
746{
747 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
748 int status;
749
1da177e4
LT
750 /*
751 * Flush all pending writes before doing anything
752 * with locks..
753 */
29884df0 754 nfs_sync_mapping(filp->f_mapping);
1da177e4
LT
755
756 /* NOTE: special case
757 * If we're signalled while cleaning up locks on process exit, we
758 * still need to complete the unlock.
759 */
5eebde23
SJ
760 /*
761 * Use local locking if mounted with "-onolock" or with appropriate
762 * "-olocal_lock="
763 */
764 if (!is_local)
1da177e4
LT
765 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
766 else
767 status = do_vfs_lock(filp, fl);
1da177e4
LT
768 return status;
769}
770
6b96724e
RL
771static int
772is_time_granular(struct timespec *ts) {
773 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
774}
775
5eebde23
SJ
776static int
777do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
778{
779 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
780 int status;
781
1da177e4
LT
782 /*
783 * Flush all pending writes before doing anything
784 * with locks..
785 */
29884df0
TM
786 status = nfs_sync_mapping(filp->f_mapping);
787 if (status != 0)
1da177e4
LT
788 goto out;
789
5eebde23
SJ
790 /*
791 * Use local locking if mounted with "-onolock" or with appropriate
792 * "-olocal_lock="
793 */
794 if (!is_local)
1da177e4 795 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
c4d7c402 796 else
1da177e4 797 status = do_vfs_lock(filp, fl);
1da177e4
LT
798 if (status < 0)
799 goto out;
6b96724e 800
1da177e4 801 /*
6b96724e
RL
802 * Revalidate the cache if the server has time stamps granular
803 * enough to detect subsecond changes. Otherwise, clear the
804 * cache to prevent missing any changes.
805 *
1da177e4
LT
806 * This makes locking act as a cache coherency point.
807 */
29884df0 808 nfs_sync_mapping(filp->f_mapping);
6b96724e
RL
809 if (!nfs_have_delegation(inode, FMODE_READ)) {
810 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
811 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
812 else
813 nfs_zap_caches(inode);
814 }
1da177e4 815out:
1da177e4
LT
816 return status;
817}
818
819/*
820 * Lock a (portion of) a file
821 */
822static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
823{
6da24bc9 824 struct inode *inode = filp->f_mapping->host;
2116271a 825 int ret = -ENOLCK;
5eebde23 826 int is_local = 0;
1da177e4 827
6da24bc9
CL
828 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
829 filp->f_path.dentry->d_parent->d_name.name,
830 filp->f_path.dentry->d_name.name,
1da177e4
LT
831 fl->fl_type, fl->fl_flags,
832 (long long)fl->fl_start, (long long)fl->fl_end);
6da24bc9 833
91d5b470 834 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
1da177e4
LT
835
836 /* No mandatory locks over NFS */
dfad9441 837 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
2116271a
TM
838 goto out_err;
839
5eebde23
SJ
840 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
841 is_local = 1;
842
2116271a
TM
843 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
844 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
845 if (ret < 0)
846 goto out_err;
847 }
1da177e4
LT
848
849 if (IS_GETLK(cmd))
5eebde23 850 ret = do_getlk(filp, cmd, fl, is_local);
2116271a 851 else if (fl->fl_type == F_UNLCK)
5eebde23 852 ret = do_unlk(filp, cmd, fl, is_local);
2116271a 853 else
5eebde23 854 ret = do_setlk(filp, cmd, fl, is_local);
2116271a
TM
855out_err:
856 return ret;
1da177e4
LT
857}
858
859/*
860 * Lock a (portion of) a file
861 */
862static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
863{
5eebde23
SJ
864 struct inode *inode = filp->f_mapping->host;
865 int is_local = 0;
866
6da24bc9
CL
867 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
868 filp->f_path.dentry->d_parent->d_name.name,
869 filp->f_path.dentry->d_name.name,
1da177e4
LT
870 fl->fl_type, fl->fl_flags);
871
1da177e4
LT
872 if (!(fl->fl_flags & FL_FLOCK))
873 return -ENOLCK;
874
5eebde23
SJ
875 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
876 is_local = 1;
877
1da177e4
LT
878 /* We're simulating flock() locks using posix locks on the server */
879 fl->fl_owner = (fl_owner_t)filp;
880 fl->fl_start = 0;
881 fl->fl_end = OFFSET_MAX;
882
883 if (fl->fl_type == F_UNLCK)
5eebde23
SJ
884 return do_unlk(filp, cmd, fl, is_local);
885 return do_setlk(filp, cmd, fl, is_local);
1da177e4 886}
370f6599 887
6da24bc9
CL
888/*
889 * There is no protocol support for leases, so we have no way to implement
890 * them correctly in the face of opens by other clients.
891 */
370f6599
BF
892static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
893{
6da24bc9
CL
894 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
895 file->f_path.dentry->d_parent->d_name.name,
896 file->f_path.dentry->d_name.name, arg);
370f6599
BF
897 return -EINVAL;
898}