]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/nfs/file.c
UBUNTU: Ubuntu-4.10.0-37.41
[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 31
7c0f6ba6 32#include <linux/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);
d7cf8dd0 104
980802e3
TM
105 if (filp->f_flags & O_DIRECT)
106 goto force_reval;
61540bf6 107 if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
d7cf8dd0 108 goto force_reval;
d7cf8dd0 109 return 0;
980802e3
TM
110force_reval:
111 return __nfs_revalidate_inode(server, inode);
112}
113
965c8e59 114loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
980802e3 115{
6de1472f
AV
116 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
117 filp, offset, whence);
b84e06c5 118
06222e49 119 /*
965c8e59 120 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
06222e49
JB
121 * the cached file length
122 */
965c8e59 123 if (whence != SEEK_SET && whence != SEEK_CUR) {
980802e3 124 struct inode *inode = filp->f_mapping->host;
d5e66348 125
980802e3
TM
126 int retval = nfs_revalidate_file_size(inode, filp);
127 if (retval < 0)
128 return (loff_t)retval;
79835a71 129 }
d5e66348 130
965c8e59 131 return generic_file_llseek(filp, offset, whence);
980802e3 132}
89d77c8f 133EXPORT_SYMBOL_GPL(nfs_file_llseek);
980802e3 134
1da177e4
LT
135/*
136 * Flush all dirty pages, and check for write errors.
1da177e4 137 */
5445b1fb 138static int
75e1fcc0 139nfs_file_flush(struct file *file, fl_owner_t id)
1da177e4 140{
6de1472f 141 struct inode *inode = file_inode(file);
1da177e4 142
6de1472f 143 dprintk("NFS: flush(%pD2)\n", file);
1da177e4 144
c2459dc4 145 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
1da177e4
LT
146 if ((file->f_mode & FMODE_WRITE) == 0)
147 return 0;
7b159fc1 148
7fe5c398 149 /* Flush writes to the server and return any errors */
af7fa165 150 return vfs_fsync(file, 0);
1da177e4
LT
151}
152
ce4ef7c0 153ssize_t
3aa2d199 154nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
1da177e4 155{
6de1472f 156 struct inode *inode = file_inode(iocb->ki_filp);
1da177e4
LT
157 ssize_t result;
158
2ba48ce5 159 if (iocb->ki_flags & IOCB_DIRECT)
c8b8e32d 160 return nfs_file_direct_read(iocb, to);
1da177e4 161
619d30b4 162 dprintk("NFS: read(%pD2, %zu@%lu)\n",
6de1472f 163 iocb->ki_filp,
3aa2d199 164 iov_iter_count(to), (unsigned long) iocb->ki_pos);
1da177e4 165
a5864c99
TM
166 nfs_start_io_read(inode);
167 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
4184dcf2 168 if (!result) {
3aa2d199 169 result = generic_file_read_iter(iocb, to);
4184dcf2
CL
170 if (result > 0)
171 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
172 }
a5864c99 173 nfs_end_io_read(inode);
1da177e4
LT
174 return result;
175}
89d77c8f 176EXPORT_SYMBOL_GPL(nfs_file_read);
1da177e4 177
ce4ef7c0 178int
1da177e4
LT
179nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
180{
6de1472f 181 struct inode *inode = file_inode(file);
1da177e4
LT
182 int status;
183
6de1472f 184 dprintk("NFS: mmap(%pD2)\n", file);
1da177e4 185
e1ebfd33
TM
186 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
187 * so we call that before revalidating the mapping
188 */
189 status = generic_file_mmap(file, vma);
94387fb1
TM
190 if (!status) {
191 vma->vm_ops = &nfs_file_vm_ops;
e1ebfd33 192 status = nfs_revalidate_mapping(inode, file->f_mapping);
94387fb1 193 }
1da177e4
LT
194 return status;
195}
89d77c8f 196EXPORT_SYMBOL_GPL(nfs_file_mmap);
1da177e4
LT
197
198/*
199 * Flush any dirty pages for this process, and check for write errors.
200 * The return status from this call provides a reliable indication of
201 * whether any write errors occurred for this process.
af7fa165
TM
202 *
203 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
204 * disk, but it retrieves and clears ctx->error after synching, despite
205 * the two being set at the same time in nfs_context_set_write_error().
206 * This is because the former is used to notify the _next_ call to
25985edc 207 * nfs_file_write() that a write error occurred, and hence cause it to
af7fa165 208 * fall back to doing a synchronous write.
1da177e4 209 */
4ff79bc7 210static int
a5c58892 211nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
1da177e4 212{
cd3758e3 213 struct nfs_open_context *ctx = nfs_file_open_context(file);
6de1472f 214 struct inode *inode = file_inode(file);
05990d1b 215 int have_error, do_resend, status;
af7fa165
TM
216 int ret = 0;
217
6de1472f 218 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
1da177e4 219
91d5b470 220 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
05990d1b 221 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
af7fa165
TM
222 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
223 status = nfs_commit_inode(inode, FLUSH_SYNC);
224 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
05990d1b 225 if (have_error) {
af7fa165 226 ret = xchg(&ctx->error, 0);
05990d1b
TM
227 if (ret)
228 goto out;
229 }
230 if (status < 0) {
af7fa165 231 ret = status;
05990d1b
TM
232 goto out;
233 }
234 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
235 if (do_resend)
236 ret = -EAGAIN;
237out:
a5c58892
BS
238 return ret;
239}
240
4ff79bc7 241int
a5c58892
BS
242nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
243{
244 int ret;
496ad9aa 245 struct inode *inode = file_inode(file);
a5c58892 246
f4ce1299
TM
247 trace_nfs_fsync_enter(inode);
248
05990d1b
TM
249 do {
250 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
251 if (ret != 0)
252 break;
05990d1b 253 ret = nfs_file_fsync_commit(file, start, end, datasync);
4ff79bc7
CH
254 if (!ret)
255 ret = pnfs_sync_inode(inode, !!datasync);
dcfc4f25
TM
256 /*
257 * If nfs_file_fsync_commit detected a server reboot, then
258 * resend all dirty pages that might have been covered by
259 * the NFS_CONTEXT_RESEND_WRITES flag
260 */
261 start = 0;
262 end = LLONG_MAX;
05990d1b
TM
263 } while (ret == -EAGAIN);
264
f4ce1299 265 trace_nfs_fsync_exit(inode, ret);
af7fa165 266 return ret;
1da177e4 267}
4ff79bc7 268EXPORT_SYMBOL_GPL(nfs_file_fsync);
1da177e4 269
38c73044
PS
270/*
271 * Decide whether a read/modify/write cycle may be more efficient
272 * then a modify/write/read cycle when writing to a page in the
273 * page cache.
274 *
275 * The modify/write/read cycle may occur if a page is read before
276 * being completely filled by the writer. In this situation, the
277 * page must be completely written to stable storage on the server
278 * before it can be refilled by reading in the page from the server.
279 * This can lead to expensive, small, FILE_SYNC mode writes being
280 * done.
281 *
282 * It may be more efficient to read the page first if the file is
283 * open for reading in addition to writing, the page is not marked
284 * as Uptodate, it is not dirty or waiting to be committed,
285 * indicating that it was previously allocated and then modified,
286 * that there were valid bytes of data in that range of the file,
287 * and that the new data won't completely replace the old data in
288 * that range of the file.
289 */
290static int nfs_want_read_modify_write(struct file *file, struct page *page,
291 loff_t pos, unsigned len)
292{
293 unsigned int pglen = nfs_page_length(page);
09cbfeaf 294 unsigned int offset = pos & (PAGE_SIZE - 1);
38c73044
PS
295 unsigned int end = offset + len;
296
612aa983
CH
297 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
298 if (!PageUptodate(page))
299 return 1;
300 return 0;
301 }
302
38c73044
PS
303 if ((file->f_mode & FMODE_READ) && /* open for read? */
304 !PageUptodate(page) && /* Uptodate? */
305 !PagePrivate(page) && /* i/o request already? */
306 pglen && /* valid bytes of file? */
307 (end < pglen || offset)) /* replace all valid bytes? */
308 return 1;
309 return 0;
310}
311
1da177e4 312/*
4899f9c8
NP
313 * This does the "real" work of the write. We must allocate and lock the
314 * page to be sent back to the generic routine, which then copies the
315 * data from user space.
1da177e4
LT
316 *
317 * If the writer ends up delaying the write, the writer needs to
318 * increment the page use counts until he is done with the page.
319 */
4899f9c8
NP
320static int nfs_write_begin(struct file *file, struct address_space *mapping,
321 loff_t pos, unsigned len, unsigned flags,
322 struct page **pagep, void **fsdata)
1da177e4 323{
4899f9c8 324 int ret;
09cbfeaf 325 pgoff_t index = pos >> PAGE_SHIFT;
4899f9c8 326 struct page *page;
38c73044 327 int once_thru = 0;
4899f9c8 328
1e8968c5 329 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
6de1472f 330 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 331
38c73044 332start:
54566b2c 333 page = grab_cache_page_write_begin(mapping, index, flags);
4899f9c8
NP
334 if (!page)
335 return -ENOMEM;
336 *pagep = page;
337
338 ret = nfs_flush_incompatible(file, page);
339 if (ret) {
340 unlock_page(page);
09cbfeaf 341 put_page(page);
38c73044
PS
342 } else if (!once_thru &&
343 nfs_want_read_modify_write(file, page, pos, len)) {
344 once_thru = 1;
345 ret = nfs_readpage(file, page);
09cbfeaf 346 put_page(page);
38c73044
PS
347 if (!ret)
348 goto start;
4899f9c8
NP
349 }
350 return ret;
1da177e4
LT
351}
352
4899f9c8
NP
353static int nfs_write_end(struct file *file, struct address_space *mapping,
354 loff_t pos, unsigned len, unsigned copied,
355 struct page *page, void *fsdata)
1da177e4 356{
09cbfeaf 357 unsigned offset = pos & (PAGE_SIZE - 1);
dc24826b 358 struct nfs_open_context *ctx = nfs_file_open_context(file);
4899f9c8 359 int status;
1da177e4 360
1e8968c5 361 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
6de1472f 362 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 363
efc91ed0
TM
364 /*
365 * Zero any uninitialised parts of the page, and then mark the page
366 * as up to date if it turns out that we're extending the file.
367 */
368 if (!PageUptodate(page)) {
369 unsigned pglen = nfs_page_length(page);
c0cf3ef5 370 unsigned end = offset + copied;
efc91ed0
TM
371
372 if (pglen == 0) {
373 zero_user_segments(page, 0, offset,
09cbfeaf 374 end, PAGE_SIZE);
efc91ed0
TM
375 SetPageUptodate(page);
376 } else if (end >= pglen) {
09cbfeaf 377 zero_user_segment(page, end, PAGE_SIZE);
efc91ed0
TM
378 if (offset == 0)
379 SetPageUptodate(page);
380 } else
09cbfeaf 381 zero_user_segment(page, pglen, PAGE_SIZE);
efc91ed0
TM
382 }
383
4899f9c8 384 status = nfs_updatepage(file, page, offset, copied);
4899f9c8
NP
385
386 unlock_page(page);
09cbfeaf 387 put_page(page);
4899f9c8 388
3d509e54
CL
389 if (status < 0)
390 return status;
2701d086 391 NFS_I(mapping->host)->write_io += copied;
dc24826b 392
ce52914e 393 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
dc24826b
AA
394 status = nfs_wb_all(mapping->host);
395 if (status < 0)
396 return status;
397 }
398
3d509e54 399 return copied;
1da177e4
LT
400}
401
6b9b3514
DH
402/*
403 * Partially or wholly invalidate a page
404 * - Release the private state associated with a page if undergoing complete
405 * page invalidation
545db45f 406 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
407 * - Caller holds page lock
408 */
d47992f8
LC
409static void nfs_invalidate_page(struct page *page, unsigned int offset,
410 unsigned int length)
cd52ed35 411{
d47992f8
LC
412 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
413 page, offset, length);
b7eaefaa 414
09cbfeaf 415 if (offset != 0 || length < PAGE_SIZE)
1c75950b 416 return;
d2ccddf0 417 /* Cancel any unstarted writes on this page */
d56b4ddf 418 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
545db45f
DH
419
420 nfs_fscache_invalidate_page(page, page->mapping->host);
cd52ed35
TM
421}
422
6b9b3514
DH
423/*
424 * Attempt to release the private state associated with a page
545db45f 425 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
426 * - Caller holds page lock
427 * - Return true (may release page) or false (may not)
428 */
cd52ed35
TM
429static int nfs_release_page(struct page *page, gfp_t gfp)
430{
b7eaefaa
CL
431 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
432
e3db7691 433 /* If PagePrivate() is set, then the page is not freeable */
545db45f
DH
434 if (PagePrivate(page))
435 return 0;
436 return nfs_fscache_release_page(page, gfp);
e3db7691
TM
437}
438
f919b196
MG
439static void nfs_check_dirty_writeback(struct page *page,
440 bool *dirty, bool *writeback)
441{
442 struct nfs_inode *nfsi;
443 struct address_space *mapping = page_file_mapping(page);
444
445 if (!mapping || PageSwapCache(page))
446 return;
447
448 /*
449 * Check if an unstable page is currently being committed and
450 * if so, have the VM treat it as if the page is under writeback
451 * so it will not block due to pages that will shortly be freeable.
452 */
453 nfsi = NFS_I(mapping->host);
af7cf057 454 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
f919b196
MG
455 *writeback = true;
456 return;
457 }
458
459 /*
460 * If PagePrivate() is set, then the page is not freeable and as the
461 * inode is not being committed, it's not going to be cleaned in the
462 * near future so treat it as dirty
463 */
464 if (PagePrivate(page))
465 *dirty = true;
466}
467
6b9b3514
DH
468/*
469 * Attempt to clear the private state associated with a page when an error
470 * occurs that requires the cached contents of an inode to be written back or
471 * destroyed
545db45f 472 * - Called if either PG_private or fscache is set on the page
6b9b3514
DH
473 * - Caller holds page lock
474 * - Return 0 if successful, -error otherwise
475 */
e3db7691
TM
476static int nfs_launder_page(struct page *page)
477{
d56b4ddf 478 struct inode *inode = page_file_mapping(page)->host;
545db45f 479 struct nfs_inode *nfsi = NFS_I(inode);
b7eaefaa
CL
480
481 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
482 inode->i_ino, (long long)page_offset(page));
483
545db45f 484 nfs_fscache_wait_on_page_write(nfsi, page);
d6c843b9 485 return nfs_wb_launder_page(inode, page);
cd52ed35
TM
486}
487
a564b8f0
MG
488static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
489 sector_t *span)
490{
dad2b015
JL
491 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
492
a564b8f0 493 *span = sis->pages;
dad2b015 494
3c87ef6e 495 return rpc_clnt_swap_activate(clnt);
a564b8f0
MG
496}
497
498static void nfs_swap_deactivate(struct file *file)
499{
dad2b015
JL
500 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
501
3c87ef6e 502 rpc_clnt_swap_deactivate(clnt);
a564b8f0 503}
a564b8f0 504
f5e54d6e 505const struct address_space_operations nfs_file_aops = {
1da177e4
LT
506 .readpage = nfs_readpage,
507 .readpages = nfs_readpages,
9cccef95 508 .set_page_dirty = __set_page_dirty_nobuffers,
1da177e4
LT
509 .writepage = nfs_writepage,
510 .writepages = nfs_writepages,
4899f9c8
NP
511 .write_begin = nfs_write_begin,
512 .write_end = nfs_write_end,
cd52ed35
TM
513 .invalidatepage = nfs_invalidate_page,
514 .releasepage = nfs_release_page,
1da177e4 515 .direct_IO = nfs_direct_IO,
f844cd0d 516#ifdef CONFIG_MIGRATION
074cc1de 517 .migratepage = nfs_migrate_page,
f844cd0d 518#endif
e3db7691 519 .launder_page = nfs_launder_page,
f919b196 520 .is_dirty_writeback = nfs_check_dirty_writeback,
f590f333 521 .error_remove_page = generic_error_remove_page,
a564b8f0
MG
522 .swap_activate = nfs_swap_activate,
523 .swap_deactivate = nfs_swap_deactivate,
1da177e4
LT
524};
525
6b9b3514
DH
526/*
527 * Notification that a PTE pointing to an NFS page is about to be made
528 * writable, implying that someone is about to modify the page through a
529 * shared-writable mapping
530 */
c2ec175c 531static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
94387fb1 532{
c2ec175c 533 struct page *page = vmf->page;
94387fb1 534 struct file *filp = vma->vm_file;
6de1472f 535 struct inode *inode = file_inode(filp);
94387fb1 536 unsigned pagelen;
bc4866b6 537 int ret = VM_FAULT_NOPAGE;
4899f9c8 538 struct address_space *mapping;
94387fb1 539
1e8968c5 540 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
6de1472f 541 filp, filp->f_mapping->host->i_ino,
b7eaefaa
CL
542 (long long)page_offset(page));
543
9a773e7c
TM
544 sb_start_pagefault(inode->i_sb);
545
545db45f 546 /* make sure the cache has finished storing the page */
6de1472f 547 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
545db45f 548
ef070dcb
TM
549 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
550 nfs_wait_bit_killable, TASK_KILLABLE);
551
94387fb1 552 lock_page(page);
d56b4ddf 553 mapping = page_file_mapping(page);
6de1472f 554 if (mapping != inode->i_mapping)
8b1f9ee5
TM
555 goto out_unlock;
556
2aeb98f4
TM
557 wait_on_page_writeback(page);
558
94387fb1 559 pagelen = nfs_page_length(page);
8b1f9ee5
TM
560 if (pagelen == 0)
561 goto out_unlock;
4899f9c8 562
bc4866b6
TM
563 ret = VM_FAULT_LOCKED;
564 if (nfs_flush_incompatible(filp, page) == 0 &&
565 nfs_updatepage(filp, page, 0, pagelen) == 0)
566 goto out;
8b1f9ee5 567
bc4866b6 568 ret = VM_FAULT_SIGBUS;
8b1f9ee5
TM
569out_unlock:
570 unlock_page(page);
bc4866b6 571out:
9a773e7c 572 sb_end_pagefault(inode->i_sb);
bc4866b6 573 return ret;
94387fb1
TM
574}
575
f0f37e2f 576static const struct vm_operations_struct nfs_file_vm_ops = {
94387fb1 577 .fault = filemap_fault,
f1820361 578 .map_pages = filemap_map_pages,
94387fb1
TM
579 .page_mkwrite = nfs_vm_page_mkwrite,
580};
581
7e94d6c4 582static int nfs_need_check_write(struct file *filp, struct inode *inode)
7b159fc1
TM
583{
584 struct nfs_open_context *ctx;
585
cd3758e3 586 ctx = nfs_file_open_context(filp);
dc24826b 587 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
ce52914e 588 nfs_ctx_key_to_expire(ctx, inode))
7b159fc1
TM
589 return 1;
590 return 0;
591}
592
edaf4369 593ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
1da177e4 594{
6de1472f
AV
595 struct file *file = iocb->ki_filp;
596 struct inode *inode = file_inode(file);
7e381172 597 unsigned long written = 0;
1da177e4
LT
598 ssize_t result;
599
6de1472f 600 result = nfs_key_timeout_notify(file, inode);
dc24826b
AA
601 if (result)
602 return result;
603
89698b24 604 if (iocb->ki_flags & IOCB_DIRECT)
65a4a1ca 605 return nfs_file_direct_write(iocb, from);
1da177e4 606
619d30b4 607 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
18290650 608 file, iov_iter_count(from), (long long) iocb->ki_pos);
1da177e4 609
1da177e4
LT
610 if (IS_SWAPFILE(inode))
611 goto out_swapfile;
7d52e862
TM
612 /*
613 * O_APPEND implies that we must revalidate the file length.
614 */
2ba48ce5 615 if (iocb->ki_flags & IOCB_APPEND) {
6de1472f 616 result = nfs_revalidate_file_size(inode, file);
7d52e862
TM
617 if (result)
618 goto out;
fe51beec 619 }
1da177e4 620
a5864c99 621 nfs_start_io_write(inode);
18290650
TM
622 result = generic_write_checks(iocb, from);
623 if (result > 0) {
624 current->backing_dev_info = inode_to_bdi(inode);
625 result = generic_perform_write(file, from, iocb->ki_pos);
626 current->backing_dev_info = NULL;
627 }
a5864c99 628 nfs_end_io_write(inode);
18290650 629 if (result <= 0)
1da177e4
LT
630 goto out;
631
c49edecd
TM
632 result = generic_write_sync(iocb, result);
633 if (result < 0)
634 goto out;
635 written = result;
18290650 636 iocb->ki_pos += written;
7e381172 637
7e94d6c4 638 /* Return error values */
18290650 639 if (nfs_need_check_write(file, inode)) {
6de1472f 640 int err = vfs_fsync(file, 0);
200baa21
TM
641 if (err < 0)
642 result = err;
643 }
18290650 644 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
1da177e4
LT
645out:
646 return result;
647
648out_swapfile:
649 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
18290650 650 return -EBUSY;
1da177e4 651}
89d77c8f 652EXPORT_SYMBOL_GPL(nfs_file_write);
1da177e4 653
5eebde23
SJ
654static int
655do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
656{
657 struct inode *inode = filp->f_mapping->host;
658 int status = 0;
21ac19d4 659 unsigned int saved_type = fl->fl_type;
1da177e4 660
039c4d7a 661 /* Try local locking first */
6d34ac19
BF
662 posix_test_lock(filp, fl);
663 if (fl->fl_type != F_UNLCK) {
664 /* found a conflict */
039c4d7a 665 goto out;
1da177e4 666 }
21ac19d4 667 fl->fl_type = saved_type;
039c4d7a 668
011e2a7f 669 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
039c4d7a
TM
670 goto out_noconflict;
671
5eebde23 672 if (is_local)
039c4d7a
TM
673 goto out_noconflict;
674
675 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
676out:
1da177e4 677 return status;
039c4d7a
TM
678out_noconflict:
679 fl->fl_type = F_UNLCK;
680 goto out;
1da177e4
LT
681}
682
5eebde23
SJ
683static int
684do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
685{
686 struct inode *inode = filp->f_mapping->host;
7a8203d8 687 struct nfs_lock_context *l_ctx;
1da177e4
LT
688 int status;
689
1da177e4
LT
690 /*
691 * Flush all pending writes before doing anything
692 * with locks..
693 */
d9dabc1a 694 vfs_fsync(filp, 0);
1da177e4 695
7a8203d8
TM
696 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
697 if (!IS_ERR(l_ctx)) {
210c7c17 698 status = nfs_iocounter_wait(l_ctx);
7a8203d8
TM
699 nfs_put_lock_context(l_ctx);
700 if (status < 0)
701 return status;
702 }
703
1da177e4
LT
704 /* NOTE: special case
705 * If we're signalled while cleaning up locks on process exit, we
706 * still need to complete the unlock.
707 */
5eebde23
SJ
708 /*
709 * Use local locking if mounted with "-onolock" or with appropriate
710 * "-olocal_lock="
711 */
712 if (!is_local)
1da177e4
LT
713 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
714 else
75575ddf 715 status = locks_lock_file_wait(filp, fl);
1da177e4
LT
716 return status;
717}
718
5eebde23
SJ
719static int
720do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
721{
722 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
723 int status;
724
1da177e4
LT
725 /*
726 * Flush all pending writes before doing anything
727 * with locks..
728 */
29884df0
TM
729 status = nfs_sync_mapping(filp->f_mapping);
730 if (status != 0)
1da177e4
LT
731 goto out;
732
5eebde23
SJ
733 /*
734 * Use local locking if mounted with "-onolock" or with appropriate
735 * "-olocal_lock="
736 */
737 if (!is_local)
1da177e4 738 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
c4d7c402 739 else
75575ddf 740 status = locks_lock_file_wait(filp, fl);
1da177e4
LT
741 if (status < 0)
742 goto out;
6b96724e 743
1da177e4 744 /*
6b96724e
RL
745 * Revalidate the cache if the server has time stamps granular
746 * enough to detect subsecond changes. Otherwise, clear the
747 * cache to prevent missing any changes.
748 *
1da177e4
LT
749 * This makes locking act as a cache coherency point.
750 */
29884df0 751 nfs_sync_mapping(filp->f_mapping);
ca0daa27
TM
752 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
753 nfs_zap_mapping(inode, filp->f_mapping);
1da177e4 754out:
1da177e4
LT
755 return status;
756}
757
758/*
759 * Lock a (portion of) a file
760 */
ce4ef7c0 761int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 762{
6da24bc9 763 struct inode *inode = filp->f_mapping->host;
2116271a 764 int ret = -ENOLCK;
5eebde23 765 int is_local = 0;
1da177e4 766
6de1472f
AV
767 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
768 filp, fl->fl_type, fl->fl_flags,
1da177e4 769 (long long)fl->fl_start, (long long)fl->fl_end);
6da24bc9 770
91d5b470 771 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
1da177e4
LT
772
773 /* No mandatory locks over NFS */
dfad9441 774 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
2116271a
TM
775 goto out_err;
776
5eebde23
SJ
777 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
778 is_local = 1;
779
2116271a
TM
780 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
781 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
782 if (ret < 0)
783 goto out_err;
784 }
1da177e4
LT
785
786 if (IS_GETLK(cmd))
5eebde23 787 ret = do_getlk(filp, cmd, fl, is_local);
2116271a 788 else if (fl->fl_type == F_UNLCK)
5eebde23 789 ret = do_unlk(filp, cmd, fl, is_local);
2116271a 790 else
5eebde23 791 ret = do_setlk(filp, cmd, fl, is_local);
2116271a
TM
792out_err:
793 return ret;
1da177e4 794}
89d77c8f 795EXPORT_SYMBOL_GPL(nfs_lock);
1da177e4
LT
796
797/*
798 * Lock a (portion of) a file
799 */
ce4ef7c0 800int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 801{
5eebde23
SJ
802 struct inode *inode = filp->f_mapping->host;
803 int is_local = 0;
804
6de1472f
AV
805 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
806 filp, fl->fl_type, fl->fl_flags);
1da177e4 807
1da177e4
LT
808 if (!(fl->fl_flags & FL_FLOCK))
809 return -ENOLCK;
810
ad0fcd4e
JL
811 /*
812 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
813 * any standard. In principle we might be able to support LOCK_MAND
814 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
815 * NFS code is not set up for it.
816 */
817 if (fl->fl_type & LOCK_MAND)
818 return -EINVAL;
819
5eebde23
SJ
820 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
821 is_local = 1;
822
1da177e4 823 /* We're simulating flock() locks using posix locks on the server */
1da177e4 824 if (fl->fl_type == F_UNLCK)
5eebde23
SJ
825 return do_unlk(filp, cmd, fl, is_local);
826 return do_setlk(filp, cmd, fl, is_local);
1da177e4 827}
89d77c8f 828EXPORT_SYMBOL_GPL(nfs_flock);
370f6599 829
0486958f
JL
830const struct file_operations nfs_file_operations = {
831 .llseek = nfs_file_llseek,
3aa2d199 832 .read_iter = nfs_file_read,
edaf4369 833 .write_iter = nfs_file_write,
0486958f
JL
834 .mmap = nfs_file_mmap,
835 .open = nfs_file_open,
836 .flush = nfs_file_flush,
837 .release = nfs_file_release,
838 .fsync = nfs_file_fsync,
839 .lock = nfs_lock,
840 .flock = nfs_flock,
82c156f8 841 .splice_read = generic_file_splice_read,
4da54c21 842 .splice_write = iter_file_splice_write,
0486958f 843 .check_flags = nfs_check_flags,
1c994a09 844 .setlease = simple_nosetlease,
0486958f 845};
ddda8e0a 846EXPORT_SYMBOL_GPL(nfs_file_operations);