]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/nfs/file.c
Merge branch 'multipath'
[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
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>
5a0e3ad6 29#include <linux/gfp.h>
b608b283 30#include <linux/swap.h>
1da177e4
LT
31
32#include <asm/uaccess.h>
1da177e4
LT
33
34#include "delegation.h"
94387fb1 35#include "internal.h"
91d5b470 36#include "iostat.h"
545db45f 37#include "fscache.h"
612aa983 38#include "pnfs.h"
1da177e4 39
f4ce1299
TM
40#include "nfstrace.h"
41
1da177e4
LT
42#define NFSDBG_FACILITY NFSDBG_FILE
43
f0f37e2f 44static const struct vm_operations_struct nfs_file_vm_ops;
94387fb1 45
1da177e4
LT
46/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
ce4ef7c0 51int nfs_check_flags(int flags)
1da177e4
LT
52{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
89d77c8f 58EXPORT_SYMBOL_GPL(nfs_check_flags);
1da177e4
LT
59
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
1da177e4
LT
66 int res;
67
6de1472f 68 dprintk("NFS: open file(%pD2)\n", filp);
cc0dd2d1 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{
6de1472f 82 dprintk("NFS: release(%pD2)\n", filp);
6da24bc9 83
91d5b470 84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
aff8d8dc
AS
85 nfs_file_clear_open_context(filp);
86 return 0;
1da177e4 87}
89d77c8f 88EXPORT_SYMBOL_GPL(nfs_file_release);
1da177e4 89
980802e3
TM
90/**
91 * nfs_revalidate_size - Revalidate the file size
92 * @inode - pointer to inode struct
93 * @file - pointer to struct file
94 *
95 * Revalidates the file length. This is basically a wrapper around
96 * nfs_revalidate_inode() that takes into account the fact that we may
97 * have cached writes (in which case we don't care about the server's
98 * idea of what the file length is), or O_DIRECT (in which case we
99 * shouldn't trust the cache).
100 */
101static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102{
103 struct nfs_server *server = NFS_SERVER(inode);
104 struct nfs_inode *nfsi = NFS_I(inode);
105
d7cf8dd0
TM
106 if (nfs_have_delegated_attributes(inode))
107 goto out_noreval;
108
980802e3
TM
109 if (filp->f_flags & O_DIRECT)
110 goto force_reval;
d7cf8dd0
TM
111 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112 goto force_reval;
113 if (nfs_attribute_timeout(inode))
114 goto force_reval;
115out_noreval:
116 return 0;
980802e3
TM
117force_reval:
118 return __nfs_revalidate_inode(server, inode);
119}
120
965c8e59 121loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
980802e3 122{
6de1472f
AV
123 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124 filp, offset, whence);
b84e06c5 125
06222e49 126 /*
965c8e59 127 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
06222e49
JB
128 * the cached file length
129 */
965c8e59 130 if (whence != SEEK_SET && whence != SEEK_CUR) {
980802e3 131 struct inode *inode = filp->f_mapping->host;
d5e66348 132
980802e3
TM
133 int retval = nfs_revalidate_file_size(inode, filp);
134 if (retval < 0)
135 return (loff_t)retval;
79835a71 136 }
d5e66348 137
965c8e59 138 return generic_file_llseek(filp, offset, whence);
980802e3 139}
89d77c8f 140EXPORT_SYMBOL_GPL(nfs_file_llseek);
980802e3 141
1da177e4
LT
142/*
143 * Flush all dirty pages, and check for write errors.
1da177e4 144 */
5445b1fb 145static int
75e1fcc0 146nfs_file_flush(struct file *file, fl_owner_t id)
1da177e4 147{
6de1472f 148 struct inode *inode = file_inode(file);
1da177e4 149
6de1472f 150 dprintk("NFS: flush(%pD2)\n", file);
1da177e4 151
c2459dc4 152 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
1da177e4
LT
153 if ((file->f_mode & FMODE_WRITE) == 0)
154 return 0;
7b159fc1 155
7fe5c398 156 /* Flush writes to the server and return any errors */
af7fa165 157 return vfs_fsync(file, 0);
1da177e4
LT
158}
159
ce4ef7c0 160ssize_t
3aa2d199 161nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
1da177e4 162{
6de1472f 163 struct inode *inode = file_inode(iocb->ki_filp);
1da177e4
LT
164 ssize_t result;
165
2ba48ce5 166 if (iocb->ki_flags & IOCB_DIRECT)
e19a8a0a 167 return nfs_file_direct_read(iocb, to, iocb->ki_pos);
1da177e4 168
619d30b4 169 dprintk("NFS: read(%pD2, %zu@%lu)\n",
6de1472f 170 iocb->ki_filp,
3aa2d199 171 iov_iter_count(to), (unsigned long) iocb->ki_pos);
1da177e4 172
874f9463 173 result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping);
4184dcf2 174 if (!result) {
3aa2d199 175 result = generic_file_read_iter(iocb, to);
4184dcf2
CL
176 if (result > 0)
177 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
178 }
1da177e4
LT
179 return result;
180}
89d77c8f 181EXPORT_SYMBOL_GPL(nfs_file_read);
1da177e4 182
ce4ef7c0 183ssize_t
f0930fff
JA
184nfs_file_splice_read(struct file *filp, loff_t *ppos,
185 struct pipe_inode_info *pipe, size_t count,
186 unsigned int flags)
1da177e4 187{
6de1472f 188 struct inode *inode = file_inode(filp);
1da177e4
LT
189 ssize_t res;
190
6de1472f
AV
191 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
192 filp, (unsigned long) count, (unsigned long long) *ppos);
1da177e4 193
874f9463 194 res = nfs_revalidate_mapping_protected(inode, filp->f_mapping);
aa2f1ef1 195 if (!res) {
f0930fff 196 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
aa2f1ef1
CL
197 if (res > 0)
198 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
199 }
1da177e4
LT
200 return res;
201}
89d77c8f 202EXPORT_SYMBOL_GPL(nfs_file_splice_read);
1da177e4 203
ce4ef7c0 204int
1da177e4
LT
205nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
206{
6de1472f 207 struct inode *inode = file_inode(file);
1da177e4
LT
208 int status;
209
6de1472f 210 dprintk("NFS: mmap(%pD2)\n", file);
1da177e4 211
e1ebfd33
TM
212 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
213 * so we call that before revalidating the mapping
214 */
215 status = generic_file_mmap(file, vma);
94387fb1
TM
216 if (!status) {
217 vma->vm_ops = &nfs_file_vm_ops;
e1ebfd33 218 status = nfs_revalidate_mapping(inode, file->f_mapping);
94387fb1 219 }
1da177e4
LT
220 return status;
221}
89d77c8f 222EXPORT_SYMBOL_GPL(nfs_file_mmap);
1da177e4
LT
223
224/*
225 * Flush any dirty pages for this process, and check for write errors.
226 * The return status from this call provides a reliable indication of
227 * whether any write errors occurred for this process.
af7fa165
TM
228 *
229 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
230 * disk, but it retrieves and clears ctx->error after synching, despite
231 * the two being set at the same time in nfs_context_set_write_error().
232 * This is because the former is used to notify the _next_ call to
25985edc 233 * nfs_file_write() that a write error occurred, and hence cause it to
af7fa165 234 * fall back to doing a synchronous write.
1da177e4 235 */
ce4ef7c0 236int
a5c58892 237nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
1da177e4 238{
cd3758e3 239 struct nfs_open_context *ctx = nfs_file_open_context(file);
6de1472f 240 struct inode *inode = file_inode(file);
05990d1b 241 int have_error, do_resend, status;
af7fa165
TM
242 int ret = 0;
243
6de1472f 244 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
1da177e4 245
91d5b470 246 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
05990d1b 247 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
af7fa165
TM
248 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
249 status = nfs_commit_inode(inode, FLUSH_SYNC);
250 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
05990d1b 251 if (have_error) {
af7fa165 252 ret = xchg(&ctx->error, 0);
05990d1b
TM
253 if (ret)
254 goto out;
255 }
256 if (status < 0) {
af7fa165 257 ret = status;
05990d1b
TM
258 goto out;
259 }
260 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
261 if (do_resend)
262 ret = -EAGAIN;
263out:
a5c58892
BS
264 return ret;
265}
89d77c8f 266EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
a5c58892
BS
267
268static int
269nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
270{
271 int ret;
496ad9aa 272 struct inode *inode = file_inode(file);
a5c58892 273
f4ce1299
TM
274 trace_nfs_fsync_enter(inode);
275
a0815d55 276 nfs_inode_dio_wait(inode);
05990d1b
TM
277 do {
278 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
279 if (ret != 0)
280 break;
5955102c 281 inode_lock(inode);
05990d1b 282 ret = nfs_file_fsync_commit(file, start, end, datasync);
5955102c 283 inode_unlock(inode);
dcfc4f25
TM
284 /*
285 * If nfs_file_fsync_commit detected a server reboot, then
286 * resend all dirty pages that might have been covered by
287 * the NFS_CONTEXT_RESEND_WRITES flag
288 */
289 start = 0;
290 end = LLONG_MAX;
05990d1b
TM
291 } while (ret == -EAGAIN);
292
f4ce1299 293 trace_nfs_fsync_exit(inode, ret);
af7fa165 294 return ret;
1da177e4
LT
295}
296
38c73044
PS
297/*
298 * Decide whether a read/modify/write cycle may be more efficient
299 * then a modify/write/read cycle when writing to a page in the
300 * page cache.
301 *
302 * The modify/write/read cycle may occur if a page is read before
303 * being completely filled by the writer. In this situation, the
304 * page must be completely written to stable storage on the server
305 * before it can be refilled by reading in the page from the server.
306 * This can lead to expensive, small, FILE_SYNC mode writes being
307 * done.
308 *
309 * It may be more efficient to read the page first if the file is
310 * open for reading in addition to writing, the page is not marked
311 * as Uptodate, it is not dirty or waiting to be committed,
312 * indicating that it was previously allocated and then modified,
313 * that there were valid bytes of data in that range of the file,
314 * and that the new data won't completely replace the old data in
315 * that range of the file.
316 */
317static int nfs_want_read_modify_write(struct file *file, struct page *page,
318 loff_t pos, unsigned len)
319{
320 unsigned int pglen = nfs_page_length(page);
321 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
322 unsigned int end = offset + len;
323
612aa983
CH
324 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
325 if (!PageUptodate(page))
326 return 1;
327 return 0;
328 }
329
38c73044
PS
330 if ((file->f_mode & FMODE_READ) && /* open for read? */
331 !PageUptodate(page) && /* Uptodate? */
332 !PagePrivate(page) && /* i/o request already? */
333 pglen && /* valid bytes of file? */
334 (end < pglen || offset)) /* replace all valid bytes? */
335 return 1;
336 return 0;
337}
338
1da177e4 339/*
4899f9c8
NP
340 * This does the "real" work of the write. We must allocate and lock the
341 * page to be sent back to the generic routine, which then copies the
342 * data from user space.
1da177e4
LT
343 *
344 * If the writer ends up delaying the write, the writer needs to
345 * increment the page use counts until he is done with the page.
346 */
4899f9c8
NP
347static int nfs_write_begin(struct file *file, struct address_space *mapping,
348 loff_t pos, unsigned len, unsigned flags,
349 struct page **pagep, void **fsdata)
1da177e4 350{
4899f9c8 351 int ret;
38c73044 352 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
4899f9c8 353 struct page *page;
38c73044 354 int once_thru = 0;
4899f9c8 355
1e8968c5 356 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
6de1472f 357 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 358
38c73044 359start:
72cb77f4
TM
360 /*
361 * Prevent starvation issues if someone is doing a consistency
362 * sync-to-disk
363 */
74316201
N
364 ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
365 nfs_wait_bit_killable, TASK_KILLABLE);
72cb77f4
TM
366 if (ret)
367 return ret;
aa5accea
TM
368 /*
369 * Wait for O_DIRECT to complete
370 */
371 nfs_inode_dio_wait(mapping->host);
72cb77f4 372
54566b2c 373 page = grab_cache_page_write_begin(mapping, index, flags);
4899f9c8
NP
374 if (!page)
375 return -ENOMEM;
376 *pagep = page;
377
378 ret = nfs_flush_incompatible(file, page);
379 if (ret) {
380 unlock_page(page);
381 page_cache_release(page);
38c73044
PS
382 } else if (!once_thru &&
383 nfs_want_read_modify_write(file, page, pos, len)) {
384 once_thru = 1;
385 ret = nfs_readpage(file, page);
386 page_cache_release(page);
387 if (!ret)
388 goto start;
4899f9c8
NP
389 }
390 return ret;
1da177e4
LT
391}
392
4899f9c8
NP
393static int nfs_write_end(struct file *file, struct address_space *mapping,
394 loff_t pos, unsigned len, unsigned copied,
395 struct page *page, void *fsdata)
1da177e4 396{
4899f9c8 397 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
dc24826b 398 struct nfs_open_context *ctx = nfs_file_open_context(file);
4899f9c8 399 int status;
1da177e4 400
1e8968c5 401 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
6de1472f 402 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 403
efc91ed0
TM
404 /*
405 * Zero any uninitialised parts of the page, and then mark the page
406 * as up to date if it turns out that we're extending the file.
407 */
408 if (!PageUptodate(page)) {
409 unsigned pglen = nfs_page_length(page);
410 unsigned end = offset + len;
411
412 if (pglen == 0) {
413 zero_user_segments(page, 0, offset,
414 end, PAGE_CACHE_SIZE);
415 SetPageUptodate(page);
416 } else if (end >= pglen) {
417 zero_user_segment(page, end, PAGE_CACHE_SIZE);
418 if (offset == 0)
419 SetPageUptodate(page);
420 } else
421 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
422 }
423
4899f9c8 424 status = nfs_updatepage(file, page, offset, copied);
4899f9c8
NP
425
426 unlock_page(page);
427 page_cache_release(page);
428
3d509e54
CL
429 if (status < 0)
430 return status;
2701d086 431 NFS_I(mapping->host)->write_io += copied;
dc24826b
AA
432
433 if (nfs_ctx_key_to_expire(ctx)) {
434 status = nfs_wb_all(mapping->host);
435 if (status < 0)
436 return status;
437 }
438
3d509e54 439 return copied;
1da177e4
LT
440}
441
6b9b3514
DH
442/*
443 * Partially or wholly invalidate a page
444 * - Release the private state associated with a page if undergoing complete
445 * page invalidation
545db45f 446 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
447 * - Caller holds page lock
448 */
d47992f8
LC
449static void nfs_invalidate_page(struct page *page, unsigned int offset,
450 unsigned int length)
cd52ed35 451{
d47992f8
LC
452 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
453 page, offset, length);
b7eaefaa 454
d47992f8 455 if (offset != 0 || length < PAGE_CACHE_SIZE)
1c75950b 456 return;
d2ccddf0 457 /* Cancel any unstarted writes on this page */
d56b4ddf 458 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
545db45f
DH
459
460 nfs_fscache_invalidate_page(page, page->mapping->host);
cd52ed35
TM
461}
462
6b9b3514
DH
463/*
464 * Attempt to release the private state associated with a page
545db45f 465 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
466 * - Caller holds page lock
467 * - Return true (may release page) or false (may not)
468 */
cd52ed35
TM
469static int nfs_release_page(struct page *page, gfp_t gfp)
470{
b608b283
TM
471 struct address_space *mapping = page->mapping;
472
b7eaefaa
CL
473 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
474
95905446 475 /* Always try to initiate a 'commit' if relevant, but only
d0164adc
MG
476 * wait for it if the caller allows blocking. Even then,
477 * only wait 1 second and only if the 'bdi' is not congested.
95905446 478 * Waiting indefinitely can cause deadlocks when the NFS
1aff5256
N
479 * server is on this machine, when a new TCP connection is
480 * needed and in other rare cases. There is no particular
481 * need to wait extensively here. A short wait has the
482 * benefit that someone else can worry about the freezer.
5cf02d09 483 */
95905446
N
484 if (mapping) {
485 struct nfs_server *nfss = NFS_SERVER(mapping->host);
486 nfs_commit_inode(mapping->host, 0);
d0164adc 487 if (gfpflags_allow_blocking(gfp) &&
353db796 488 !bdi_write_congested(&nfss->backing_dev_info)) {
95905446
N
489 wait_on_page_bit_killable_timeout(page, PG_private,
490 HZ);
353db796
N
491 if (PagePrivate(page))
492 set_bdi_congested(&nfss->backing_dev_info,
493 BLK_RW_ASYNC);
95905446 494 }
b608b283 495 }
e3db7691 496 /* If PagePrivate() is set, then the page is not freeable */
545db45f
DH
497 if (PagePrivate(page))
498 return 0;
499 return nfs_fscache_release_page(page, gfp);
e3db7691
TM
500}
501
f919b196
MG
502static void nfs_check_dirty_writeback(struct page *page,
503 bool *dirty, bool *writeback)
504{
505 struct nfs_inode *nfsi;
506 struct address_space *mapping = page_file_mapping(page);
507
508 if (!mapping || PageSwapCache(page))
509 return;
510
511 /*
512 * Check if an unstable page is currently being committed and
513 * if so, have the VM treat it as if the page is under writeback
514 * so it will not block due to pages that will shortly be freeable.
515 */
516 nfsi = NFS_I(mapping->host);
af7cf057 517 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
f919b196
MG
518 *writeback = true;
519 return;
520 }
521
522 /*
523 * If PagePrivate() is set, then the page is not freeable and as the
524 * inode is not being committed, it's not going to be cleaned in the
525 * near future so treat it as dirty
526 */
527 if (PagePrivate(page))
528 *dirty = true;
529}
530
6b9b3514
DH
531/*
532 * Attempt to clear the private state associated with a page when an error
533 * occurs that requires the cached contents of an inode to be written back or
534 * destroyed
545db45f 535 * - Called if either PG_private or fscache is set on the page
6b9b3514
DH
536 * - Caller holds page lock
537 * - Return 0 if successful, -error otherwise
538 */
e3db7691
TM
539static int nfs_launder_page(struct page *page)
540{
d56b4ddf 541 struct inode *inode = page_file_mapping(page)->host;
545db45f 542 struct nfs_inode *nfsi = NFS_I(inode);
b7eaefaa
CL
543
544 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
545 inode->i_ino, (long long)page_offset(page));
546
545db45f 547 nfs_fscache_wait_on_page_write(nfsi, page);
d6c843b9 548 return nfs_wb_launder_page(inode, page);
cd52ed35
TM
549}
550
a564b8f0
MG
551static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
552 sector_t *span)
553{
dad2b015
JL
554 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
555
a564b8f0 556 *span = sis->pages;
dad2b015 557
3c87ef6e 558 return rpc_clnt_swap_activate(clnt);
a564b8f0
MG
559}
560
561static void nfs_swap_deactivate(struct file *file)
562{
dad2b015
JL
563 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
564
3c87ef6e 565 rpc_clnt_swap_deactivate(clnt);
a564b8f0 566}
a564b8f0 567
f5e54d6e 568const struct address_space_operations nfs_file_aops = {
1da177e4
LT
569 .readpage = nfs_readpage,
570 .readpages = nfs_readpages,
9cccef95 571 .set_page_dirty = __set_page_dirty_nobuffers,
1da177e4
LT
572 .writepage = nfs_writepage,
573 .writepages = nfs_writepages,
4899f9c8
NP
574 .write_begin = nfs_write_begin,
575 .write_end = nfs_write_end,
cd52ed35
TM
576 .invalidatepage = nfs_invalidate_page,
577 .releasepage = nfs_release_page,
1da177e4 578 .direct_IO = nfs_direct_IO,
074cc1de 579 .migratepage = nfs_migrate_page,
e3db7691 580 .launder_page = nfs_launder_page,
f919b196 581 .is_dirty_writeback = nfs_check_dirty_writeback,
f590f333 582 .error_remove_page = generic_error_remove_page,
a564b8f0
MG
583 .swap_activate = nfs_swap_activate,
584 .swap_deactivate = nfs_swap_deactivate,
1da177e4
LT
585};
586
6b9b3514
DH
587/*
588 * Notification that a PTE pointing to an NFS page is about to be made
589 * writable, implying that someone is about to modify the page through a
590 * shared-writable mapping
591 */
c2ec175c 592static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
94387fb1 593{
c2ec175c 594 struct page *page = vmf->page;
94387fb1 595 struct file *filp = vma->vm_file;
6de1472f 596 struct inode *inode = file_inode(filp);
94387fb1 597 unsigned pagelen;
bc4866b6 598 int ret = VM_FAULT_NOPAGE;
4899f9c8 599 struct address_space *mapping;
94387fb1 600
1e8968c5 601 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
6de1472f 602 filp, filp->f_mapping->host->i_ino,
b7eaefaa
CL
603 (long long)page_offset(page));
604
545db45f 605 /* make sure the cache has finished storing the page */
6de1472f 606 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
545db45f 607
ef070dcb
TM
608 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
609 nfs_wait_bit_killable, TASK_KILLABLE);
610
94387fb1 611 lock_page(page);
d56b4ddf 612 mapping = page_file_mapping(page);
6de1472f 613 if (mapping != inode->i_mapping)
8b1f9ee5
TM
614 goto out_unlock;
615
2aeb98f4
TM
616 wait_on_page_writeback(page);
617
94387fb1 618 pagelen = nfs_page_length(page);
8b1f9ee5
TM
619 if (pagelen == 0)
620 goto out_unlock;
4899f9c8 621
bc4866b6
TM
622 ret = VM_FAULT_LOCKED;
623 if (nfs_flush_incompatible(filp, page) == 0 &&
624 nfs_updatepage(filp, page, 0, pagelen) == 0)
625 goto out;
8b1f9ee5 626
bc4866b6 627 ret = VM_FAULT_SIGBUS;
8b1f9ee5
TM
628out_unlock:
629 unlock_page(page);
bc4866b6
TM
630out:
631 return ret;
94387fb1
TM
632}
633
f0f37e2f 634static const struct vm_operations_struct nfs_file_vm_ops = {
94387fb1 635 .fault = filemap_fault,
f1820361 636 .map_pages = filemap_map_pages,
94387fb1
TM
637 .page_mkwrite = nfs_vm_page_mkwrite,
638};
639
7e94d6c4 640static int nfs_need_check_write(struct file *filp, struct inode *inode)
7b159fc1
TM
641{
642 struct nfs_open_context *ctx;
643
cd3758e3 644 ctx = nfs_file_open_context(filp);
dc24826b
AA
645 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
646 nfs_ctx_key_to_expire(ctx))
7b159fc1
TM
647 return 1;
648 return 0;
649}
650
edaf4369 651ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
1da177e4 652{
6de1472f
AV
653 struct file *file = iocb->ki_filp;
654 struct inode *inode = file_inode(file);
7e381172 655 unsigned long written = 0;
1da177e4 656 ssize_t result;
edaf4369 657 size_t count = iov_iter_count(from);
1da177e4 658
6de1472f 659 result = nfs_key_timeout_notify(file, inode);
dc24826b
AA
660 if (result)
661 return result;
662
65a4a1ca
AV
663 if (iocb->ki_flags & IOCB_DIRECT) {
664 result = generic_write_checks(iocb, from);
665 if (result <= 0)
666 return result;
667 return nfs_file_direct_write(iocb, from);
668 }
1da177e4 669
619d30b4 670 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
65a4a1ca 671 file, count, (long long) iocb->ki_pos);
1da177e4
LT
672
673 result = -EBUSY;
674 if (IS_SWAPFILE(inode))
675 goto out_swapfile;
7d52e862
TM
676 /*
677 * O_APPEND implies that we must revalidate the file length.
678 */
2ba48ce5 679 if (iocb->ki_flags & IOCB_APPEND) {
6de1472f 680 result = nfs_revalidate_file_size(inode, file);
7d52e862
TM
681 if (result)
682 goto out;
fe51beec 683 }
1da177e4
LT
684
685 result = count;
686 if (!count)
687 goto out;
688
edaf4369 689 result = generic_file_write_iter(iocb, from);
7e381172
CL
690 if (result > 0)
691 written = result;
692
7e94d6c4
TM
693 /* Return error values */
694 if (result >= 0 && nfs_need_check_write(file, inode)) {
6de1472f 695 int err = vfs_fsync(file, 0);
200baa21
TM
696 if (err < 0)
697 result = err;
698 }
7e381172
CL
699 if (result > 0)
700 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
1da177e4
LT
701out:
702 return result;
703
704out_swapfile:
705 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
706 goto out;
707}
89d77c8f 708EXPORT_SYMBOL_GPL(nfs_file_write);
1da177e4 709
5eebde23
SJ
710static int
711do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
712{
713 struct inode *inode = filp->f_mapping->host;
714 int status = 0;
21ac19d4 715 unsigned int saved_type = fl->fl_type;
1da177e4 716
039c4d7a 717 /* Try local locking first */
6d34ac19
BF
718 posix_test_lock(filp, fl);
719 if (fl->fl_type != F_UNLCK) {
720 /* found a conflict */
039c4d7a 721 goto out;
1da177e4 722 }
21ac19d4 723 fl->fl_type = saved_type;
039c4d7a 724
011e2a7f 725 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
039c4d7a
TM
726 goto out_noconflict;
727
5eebde23 728 if (is_local)
039c4d7a
TM
729 goto out_noconflict;
730
731 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
732out:
1da177e4 733 return status;
039c4d7a
TM
734out_noconflict:
735 fl->fl_type = F_UNLCK;
736 goto out;
1da177e4
LT
737}
738
739static int do_vfs_lock(struct file *file, struct file_lock *fl)
740{
4f656367 741 return locks_lock_file_wait(file, fl);
1da177e4
LT
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;
7a8203d8 748 struct nfs_lock_context *l_ctx;
1da177e4
LT
749 int status;
750
1da177e4
LT
751 /*
752 * Flush all pending writes before doing anything
753 * with locks..
754 */
d9dabc1a 755 vfs_fsync(filp, 0);
1da177e4 756
7a8203d8
TM
757 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
758 if (!IS_ERR(l_ctx)) {
210c7c17 759 status = nfs_iocounter_wait(l_ctx);
7a8203d8
TM
760 nfs_put_lock_context(l_ctx);
761 if (status < 0)
762 return status;
763 }
764
1da177e4
LT
765 /* NOTE: special case
766 * If we're signalled while cleaning up locks on process exit, we
767 * still need to complete the unlock.
768 */
5eebde23
SJ
769 /*
770 * Use local locking if mounted with "-onolock" or with appropriate
771 * "-olocal_lock="
772 */
773 if (!is_local)
1da177e4
LT
774 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
775 else
776 status = do_vfs_lock(filp, fl);
1da177e4
LT
777 return status;
778}
779
6b96724e
RL
780static int
781is_time_granular(struct timespec *ts) {
782 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
783}
784
5eebde23
SJ
785static int
786do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
787{
788 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
789 int status;
790
1da177e4
LT
791 /*
792 * Flush all pending writes before doing anything
793 * with locks..
794 */
29884df0
TM
795 status = nfs_sync_mapping(filp->f_mapping);
796 if (status != 0)
1da177e4
LT
797 goto out;
798
5eebde23
SJ
799 /*
800 * Use local locking if mounted with "-onolock" or with appropriate
801 * "-olocal_lock="
802 */
803 if (!is_local)
1da177e4 804 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
c4d7c402 805 else
1da177e4 806 status = do_vfs_lock(filp, fl);
1da177e4
LT
807 if (status < 0)
808 goto out;
6b96724e 809
1da177e4 810 /*
6b96724e
RL
811 * Revalidate the cache if the server has time stamps granular
812 * enough to detect subsecond changes. Otherwise, clear the
813 * cache to prevent missing any changes.
814 *
1da177e4
LT
815 * This makes locking act as a cache coherency point.
816 */
29884df0 817 nfs_sync_mapping(filp->f_mapping);
011e2a7f 818 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
6b96724e
RL
819 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
820 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
821 else
822 nfs_zap_caches(inode);
823 }
1da177e4 824out:
1da177e4
LT
825 return status;
826}
827
828/*
829 * Lock a (portion of) a file
830 */
ce4ef7c0 831int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 832{
6da24bc9 833 struct inode *inode = filp->f_mapping->host;
2116271a 834 int ret = -ENOLCK;
5eebde23 835 int is_local = 0;
1da177e4 836
6de1472f
AV
837 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
838 filp, fl->fl_type, fl->fl_flags,
1da177e4 839 (long long)fl->fl_start, (long long)fl->fl_end);
6da24bc9 840
91d5b470 841 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
1da177e4
LT
842
843 /* No mandatory locks over NFS */
dfad9441 844 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
2116271a
TM
845 goto out_err;
846
5eebde23
SJ
847 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
848 is_local = 1;
849
2116271a
TM
850 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
851 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
852 if (ret < 0)
853 goto out_err;
854 }
1da177e4
LT
855
856 if (IS_GETLK(cmd))
5eebde23 857 ret = do_getlk(filp, cmd, fl, is_local);
2116271a 858 else if (fl->fl_type == F_UNLCK)
5eebde23 859 ret = do_unlk(filp, cmd, fl, is_local);
2116271a 860 else
5eebde23 861 ret = do_setlk(filp, cmd, fl, is_local);
2116271a
TM
862out_err:
863 return ret;
1da177e4 864}
89d77c8f 865EXPORT_SYMBOL_GPL(nfs_lock);
1da177e4
LT
866
867/*
868 * Lock a (portion of) a file
869 */
ce4ef7c0 870int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 871{
5eebde23
SJ
872 struct inode *inode = filp->f_mapping->host;
873 int is_local = 0;
874
6de1472f
AV
875 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
876 filp, fl->fl_type, fl->fl_flags);
1da177e4 877
1da177e4
LT
878 if (!(fl->fl_flags & FL_FLOCK))
879 return -ENOLCK;
880
ad0fcd4e
JL
881 /*
882 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
883 * any standard. In principle we might be able to support LOCK_MAND
884 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
885 * NFS code is not set up for it.
886 */
887 if (fl->fl_type & LOCK_MAND)
888 return -EINVAL;
889
5eebde23
SJ
890 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
891 is_local = 1;
892
1da177e4 893 /* We're simulating flock() locks using posix locks on the server */
1da177e4 894 if (fl->fl_type == F_UNLCK)
5eebde23
SJ
895 return do_unlk(filp, cmd, fl, is_local);
896 return do_setlk(filp, cmd, fl, is_local);
1da177e4 897}
89d77c8f 898EXPORT_SYMBOL_GPL(nfs_flock);
370f6599 899
0486958f
JL
900const struct file_operations nfs_file_operations = {
901 .llseek = nfs_file_llseek,
3aa2d199 902 .read_iter = nfs_file_read,
edaf4369 903 .write_iter = nfs_file_write,
0486958f
JL
904 .mmap = nfs_file_mmap,
905 .open = nfs_file_open,
906 .flush = nfs_file_flush,
907 .release = nfs_file_release,
908 .fsync = nfs_file_fsync,
909 .lock = nfs_lock,
910 .flock = nfs_flock,
911 .splice_read = nfs_file_splice_read,
4da54c21 912 .splice_write = iter_file_splice_write,
0486958f 913 .check_flags = nfs_check_flags,
1c994a09 914 .setlease = simple_nosetlease,
0486958f 915};
ddda8e0a 916EXPORT_SYMBOL_GPL(nfs_file_operations);