]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/nfs/file.c
pass iov_iter to ->direct_IO()
[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>
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 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);
46cb650c 85 return nfs_release(inode, filp);
1da177e4 86}
89d77c8f 87EXPORT_SYMBOL_GPL(nfs_file_release);
1da177e4 88
980802e3
TM
89/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
d7cf8dd0
TM
105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
980802e3
TM
108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
d7cf8dd0
TM
110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
980802e3
TM
116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
965c8e59 120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
980802e3 121{
6de1472f
AV
122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
b84e06c5 124
06222e49 125 /*
965c8e59 126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
06222e49
JB
127 * the cached file length
128 */
965c8e59 129 if (whence != SEEK_SET && whence != SEEK_CUR) {
980802e3 130 struct inode *inode = filp->f_mapping->host;
d5e66348 131
980802e3
TM
132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
79835a71 135 }
d5e66348 136
965c8e59 137 return generic_file_llseek(filp, offset, whence);
980802e3 138}
89d77c8f 139EXPORT_SYMBOL_GPL(nfs_file_llseek);
980802e3 140
1da177e4
LT
141/*
142 * Flush all dirty pages, and check for write errors.
1da177e4 143 */
ce4ef7c0 144int
75e1fcc0 145nfs_file_flush(struct file *file, fl_owner_t id)
1da177e4 146{
6de1472f 147 struct inode *inode = file_inode(file);
1da177e4 148
6de1472f 149 dprintk("NFS: flush(%pD2)\n", file);
1da177e4 150
c2459dc4 151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
1da177e4
LT
152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
7b159fc1 154
14546c33
TM
155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
011e2a7f 159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
14546c33
TM
160 return filemap_fdatawrite(file->f_mapping);
161
7fe5c398 162 /* Flush writes to the server and return any errors */
af7fa165 163 return vfs_fsync(file, 0);
1da177e4 164}
89d77c8f 165EXPORT_SYMBOL_GPL(nfs_file_flush);
1da177e4 166
ce4ef7c0 167ssize_t
027445c3
BP
168nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
1da177e4 170{
6de1472f 171 struct inode *inode = file_inode(iocb->ki_filp);
1da177e4
LT
172 ssize_t result;
173
1da177e4 174 if (iocb->ki_filp->f_flags & O_DIRECT)
a564b8f0 175 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
1da177e4 176
6de1472f
AV
177 dprintk("NFS: read(%pD2, %lu@%lu)\n",
178 iocb->ki_filp,
6f276e49 179 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
1da177e4 180
44b11874 181 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
4184dcf2 182 if (!result) {
027445c3 183 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
4184dcf2
CL
184 if (result > 0)
185 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 }
1da177e4
LT
187 return result;
188}
89d77c8f 189EXPORT_SYMBOL_GPL(nfs_file_read);
1da177e4 190
ce4ef7c0 191ssize_t
f0930fff
JA
192nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 struct pipe_inode_info *pipe, size_t count,
194 unsigned int flags)
1da177e4 195{
6de1472f 196 struct inode *inode = file_inode(filp);
1da177e4
LT
197 ssize_t res;
198
6de1472f
AV
199 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 filp, (unsigned long) count, (unsigned long long) *ppos);
1da177e4 201
44b11874 202 res = nfs_revalidate_mapping(inode, filp->f_mapping);
aa2f1ef1 203 if (!res) {
f0930fff 204 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
aa2f1ef1
CL
205 if (res > 0)
206 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 }
1da177e4
LT
208 return res;
209}
89d77c8f 210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
1da177e4 211
ce4ef7c0 212int
1da177e4
LT
213nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214{
6de1472f 215 struct inode *inode = file_inode(file);
1da177e4
LT
216 int status;
217
6de1472f 218 dprintk("NFS: mmap(%pD2)\n", file);
1da177e4 219
e1ebfd33
TM
220 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 * so we call that before revalidating the mapping
222 */
223 status = generic_file_mmap(file, vma);
94387fb1
TM
224 if (!status) {
225 vma->vm_ops = &nfs_file_vm_ops;
e1ebfd33 226 status = nfs_revalidate_mapping(inode, file->f_mapping);
94387fb1 227 }
1da177e4
LT
228 return status;
229}
89d77c8f 230EXPORT_SYMBOL_GPL(nfs_file_mmap);
1da177e4
LT
231
232/*
233 * Flush any dirty pages for this process, and check for write errors.
234 * The return status from this call provides a reliable indication of
235 * whether any write errors occurred for this process.
af7fa165
TM
236 *
237 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238 * disk, but it retrieves and clears ctx->error after synching, despite
239 * the two being set at the same time in nfs_context_set_write_error().
240 * This is because the former is used to notify the _next_ call to
25985edc 241 * nfs_file_write() that a write error occurred, and hence cause it to
af7fa165 242 * fall back to doing a synchronous write.
1da177e4 243 */
ce4ef7c0 244int
a5c58892 245nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
1da177e4 246{
cd3758e3 247 struct nfs_open_context *ctx = nfs_file_open_context(file);
6de1472f 248 struct inode *inode = file_inode(file);
05990d1b 249 int have_error, do_resend, status;
af7fa165
TM
250 int ret = 0;
251
6de1472f 252 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
1da177e4 253
91d5b470 254 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
05990d1b 255 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
af7fa165
TM
256 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 status = nfs_commit_inode(inode, FLUSH_SYNC);
258 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
05990d1b 259 if (have_error) {
af7fa165 260 ret = xchg(&ctx->error, 0);
05990d1b
TM
261 if (ret)
262 goto out;
263 }
264 if (status < 0) {
af7fa165 265 ret = status;
05990d1b
TM
266 goto out;
267 }
268 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 if (do_resend)
270 ret = -EAGAIN;
271out:
a5c58892
BS
272 return ret;
273}
89d77c8f 274EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
a5c58892
BS
275
276static int
277nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278{
279 int ret;
496ad9aa 280 struct inode *inode = file_inode(file);
a5c58892 281
f4ce1299
TM
282 trace_nfs_fsync_enter(inode);
283
05990d1b
TM
284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
dcfc4f25
TM
291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
05990d1b
TM
298 } while (ret == -EAGAIN);
299
f4ce1299 300 trace_nfs_fsync_exit(inode, ret);
af7fa165 301 return ret;
1da177e4
LT
302}
303
38c73044
PS
304/*
305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
331 if ((file->f_mode & FMODE_READ) && /* open for read? */
332 !PageUptodate(page) && /* Uptodate? */
333 !PagePrivate(page) && /* i/o request already? */
334 pglen && /* valid bytes of file? */
335 (end < pglen || offset)) /* replace all valid bytes? */
336 return 1;
337 return 0;
338}
339
1da177e4 340/*
4899f9c8
NP
341 * This does the "real" work of the write. We must allocate and lock the
342 * page to be sent back to the generic routine, which then copies the
343 * data from user space.
1da177e4
LT
344 *
345 * If the writer ends up delaying the write, the writer needs to
346 * increment the page use counts until he is done with the page.
347 */
4899f9c8
NP
348static int nfs_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
1da177e4 351{
4899f9c8 352 int ret;
38c73044 353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
4899f9c8 354 struct page *page;
38c73044 355 int once_thru = 0;
4899f9c8 356
1e8968c5 357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
6de1472f 358 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 359
38c73044 360start:
72cb77f4
TM
361 /*
362 * Prevent starvation issues if someone is doing a consistency
363 * sync-to-disk
364 */
365 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
366 nfs_wait_bit_killable, TASK_KILLABLE);
367 if (ret)
368 return ret;
369
54566b2c 370 page = grab_cache_page_write_begin(mapping, index, flags);
4899f9c8
NP
371 if (!page)
372 return -ENOMEM;
373 *pagep = page;
374
375 ret = nfs_flush_incompatible(file, page);
376 if (ret) {
377 unlock_page(page);
378 page_cache_release(page);
38c73044
PS
379 } else if (!once_thru &&
380 nfs_want_read_modify_write(file, page, pos, len)) {
381 once_thru = 1;
382 ret = nfs_readpage(file, page);
383 page_cache_release(page);
384 if (!ret)
385 goto start;
4899f9c8
NP
386 }
387 return ret;
1da177e4
LT
388}
389
4899f9c8
NP
390static int nfs_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
1da177e4 393{
4899f9c8 394 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
dc24826b 395 struct nfs_open_context *ctx = nfs_file_open_context(file);
4899f9c8 396 int status;
1da177e4 397
1e8968c5 398 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
6de1472f 399 file, mapping->host->i_ino, len, (long long) pos);
b7eaefaa 400
efc91ed0
TM
401 /*
402 * Zero any uninitialised parts of the page, and then mark the page
403 * as up to date if it turns out that we're extending the file.
404 */
405 if (!PageUptodate(page)) {
406 unsigned pglen = nfs_page_length(page);
407 unsigned end = offset + len;
408
409 if (pglen == 0) {
410 zero_user_segments(page, 0, offset,
411 end, PAGE_CACHE_SIZE);
412 SetPageUptodate(page);
413 } else if (end >= pglen) {
414 zero_user_segment(page, end, PAGE_CACHE_SIZE);
415 if (offset == 0)
416 SetPageUptodate(page);
417 } else
418 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
419 }
420
4899f9c8 421 status = nfs_updatepage(file, page, offset, copied);
4899f9c8
NP
422
423 unlock_page(page);
424 page_cache_release(page);
425
3d509e54
CL
426 if (status < 0)
427 return status;
2701d086 428 NFS_I(mapping->host)->write_io += copied;
dc24826b
AA
429
430 if (nfs_ctx_key_to_expire(ctx)) {
431 status = nfs_wb_all(mapping->host);
432 if (status < 0)
433 return status;
434 }
435
3d509e54 436 return copied;
1da177e4
LT
437}
438
6b9b3514
DH
439/*
440 * Partially or wholly invalidate a page
441 * - Release the private state associated with a page if undergoing complete
442 * page invalidation
545db45f 443 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
444 * - Caller holds page lock
445 */
d47992f8
LC
446static void nfs_invalidate_page(struct page *page, unsigned int offset,
447 unsigned int length)
cd52ed35 448{
d47992f8
LC
449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
450 page, offset, length);
b7eaefaa 451
d47992f8 452 if (offset != 0 || length < PAGE_CACHE_SIZE)
1c75950b 453 return;
d2ccddf0 454 /* Cancel any unstarted writes on this page */
d56b4ddf 455 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
545db45f
DH
456
457 nfs_fscache_invalidate_page(page, page->mapping->host);
cd52ed35
TM
458}
459
6b9b3514
DH
460/*
461 * Attempt to release the private state associated with a page
545db45f 462 * - Called if either PG_private or PG_fscache is set on the page
6b9b3514
DH
463 * - Caller holds page lock
464 * - Return true (may release page) or false (may not)
465 */
cd52ed35
TM
466static int nfs_release_page(struct page *page, gfp_t gfp)
467{
b608b283
TM
468 struct address_space *mapping = page->mapping;
469
b7eaefaa
CL
470 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
471
5cf02d09
JL
472 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
473 * doing this memory reclaim for a fs-related allocation.
474 */
475 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
476 !(current->flags & PF_FSTRANS)) {
b608b283
TM
477 int how = FLUSH_SYNC;
478
479 /* Don't let kswapd deadlock waiting for OOM RPC calls */
480 if (current_is_kswapd())
481 how = 0;
482 nfs_commit_inode(mapping->host, how);
483 }
e3db7691 484 /* If PagePrivate() is set, then the page is not freeable */
545db45f
DH
485 if (PagePrivate(page))
486 return 0;
487 return nfs_fscache_release_page(page, gfp);
e3db7691
TM
488}
489
f919b196
MG
490static void nfs_check_dirty_writeback(struct page *page,
491 bool *dirty, bool *writeback)
492{
493 struct nfs_inode *nfsi;
494 struct address_space *mapping = page_file_mapping(page);
495
496 if (!mapping || PageSwapCache(page))
497 return;
498
499 /*
500 * Check if an unstable page is currently being committed and
501 * if so, have the VM treat it as if the page is under writeback
502 * so it will not block due to pages that will shortly be freeable.
503 */
504 nfsi = NFS_I(mapping->host);
505 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
506 *writeback = true;
507 return;
508 }
509
510 /*
511 * If PagePrivate() is set, then the page is not freeable and as the
512 * inode is not being committed, it's not going to be cleaned in the
513 * near future so treat it as dirty
514 */
515 if (PagePrivate(page))
516 *dirty = true;
517}
518
6b9b3514
DH
519/*
520 * Attempt to clear the private state associated with a page when an error
521 * occurs that requires the cached contents of an inode to be written back or
522 * destroyed
545db45f 523 * - Called if either PG_private or fscache is set on the page
6b9b3514
DH
524 * - Caller holds page lock
525 * - Return 0 if successful, -error otherwise
526 */
e3db7691
TM
527static int nfs_launder_page(struct page *page)
528{
d56b4ddf 529 struct inode *inode = page_file_mapping(page)->host;
545db45f 530 struct nfs_inode *nfsi = NFS_I(inode);
b7eaefaa
CL
531
532 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
533 inode->i_ino, (long long)page_offset(page));
534
545db45f 535 nfs_fscache_wait_on_page_write(nfsi, page);
b7eaefaa 536 return nfs_wb_page(inode, page);
cd52ed35
TM
537}
538
a564b8f0
MG
539#ifdef CONFIG_NFS_SWAP
540static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
541 sector_t *span)
542{
543 *span = sis->pages;
544 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
545}
546
547static void nfs_swap_deactivate(struct file *file)
548{
549 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
550}
551#endif
552
f5e54d6e 553const struct address_space_operations nfs_file_aops = {
1da177e4
LT
554 .readpage = nfs_readpage,
555 .readpages = nfs_readpages,
9cccef95 556 .set_page_dirty = __set_page_dirty_nobuffers,
1da177e4
LT
557 .writepage = nfs_writepage,
558 .writepages = nfs_writepages,
4899f9c8
NP
559 .write_begin = nfs_write_begin,
560 .write_end = nfs_write_end,
cd52ed35
TM
561 .invalidatepage = nfs_invalidate_page,
562 .releasepage = nfs_release_page,
1da177e4 563 .direct_IO = nfs_direct_IO,
074cc1de 564 .migratepage = nfs_migrate_page,
e3db7691 565 .launder_page = nfs_launder_page,
f919b196 566 .is_dirty_writeback = nfs_check_dirty_writeback,
f590f333 567 .error_remove_page = generic_error_remove_page,
a564b8f0
MG
568#ifdef CONFIG_NFS_SWAP
569 .swap_activate = nfs_swap_activate,
570 .swap_deactivate = nfs_swap_deactivate,
571#endif
1da177e4
LT
572};
573
6b9b3514
DH
574/*
575 * Notification that a PTE pointing to an NFS page is about to be made
576 * writable, implying that someone is about to modify the page through a
577 * shared-writable mapping
578 */
c2ec175c 579static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
94387fb1 580{
c2ec175c 581 struct page *page = vmf->page;
94387fb1 582 struct file *filp = vma->vm_file;
6de1472f 583 struct inode *inode = file_inode(filp);
94387fb1 584 unsigned pagelen;
bc4866b6 585 int ret = VM_FAULT_NOPAGE;
4899f9c8 586 struct address_space *mapping;
94387fb1 587
1e8968c5 588 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
6de1472f 589 filp, filp->f_mapping->host->i_ino,
b7eaefaa
CL
590 (long long)page_offset(page));
591
545db45f 592 /* make sure the cache has finished storing the page */
6de1472f 593 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
545db45f 594
94387fb1 595 lock_page(page);
d56b4ddf 596 mapping = page_file_mapping(page);
6de1472f 597 if (mapping != inode->i_mapping)
8b1f9ee5
TM
598 goto out_unlock;
599
2aeb98f4
TM
600 wait_on_page_writeback(page);
601
94387fb1 602 pagelen = nfs_page_length(page);
8b1f9ee5
TM
603 if (pagelen == 0)
604 goto out_unlock;
4899f9c8 605
bc4866b6
TM
606 ret = VM_FAULT_LOCKED;
607 if (nfs_flush_incompatible(filp, page) == 0 &&
608 nfs_updatepage(filp, page, 0, pagelen) == 0)
609 goto out;
8b1f9ee5 610
bc4866b6 611 ret = VM_FAULT_SIGBUS;
8b1f9ee5
TM
612out_unlock:
613 unlock_page(page);
bc4866b6
TM
614out:
615 return ret;
94387fb1
TM
616}
617
f0f37e2f 618static const struct vm_operations_struct nfs_file_vm_ops = {
94387fb1 619 .fault = filemap_fault,
f1820361 620 .map_pages = filemap_map_pages,
94387fb1 621 .page_mkwrite = nfs_vm_page_mkwrite,
0b173bc4 622 .remap_pages = generic_file_remap_pages,
94387fb1
TM
623};
624
7b159fc1
TM
625static int nfs_need_sync_write(struct file *filp, struct inode *inode)
626{
627 struct nfs_open_context *ctx;
628
6b2f3d1f 629 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
7b159fc1 630 return 1;
cd3758e3 631 ctx = nfs_file_open_context(filp);
dc24826b
AA
632 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
633 nfs_ctx_key_to_expire(ctx))
7b159fc1
TM
634 return 1;
635 return 0;
636}
637
ce4ef7c0
BS
638ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
639 unsigned long nr_segs, loff_t pos)
1da177e4 640{
6de1472f
AV
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
7e381172 643 unsigned long written = 0;
1da177e4 644 ssize_t result;
027445c3 645 size_t count = iov_length(iov, nr_segs);
1da177e4 646
6de1472f 647 result = nfs_key_timeout_notify(file, inode);
dc24826b
AA
648 if (result)
649 return result;
650
6de1472f 651 if (file->f_flags & O_DIRECT)
a564b8f0 652 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
1da177e4 653
6de1472f
AV
654 dprintk("NFS: write(%pD2, %lu@%Ld)\n",
655 file, (unsigned long) count, (long long) pos);
1da177e4
LT
656
657 result = -EBUSY;
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
7d52e862
TM
660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
6de1472f
AV
663 if (file->f_flags & O_APPEND) {
664 result = nfs_revalidate_file_size(inode, file);
7d52e862
TM
665 if (result)
666 goto out;
fe51beec 667 }
1da177e4
LT
668
669 result = count;
670 if (!count)
671 goto out;
672
027445c3 673 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
7e381172
CL
674 if (result > 0)
675 written = result;
676
6b2f3d1f 677 /* Return error values for O_DSYNC and IS_SYNC() */
6de1472f
AV
678 if (result >= 0 && nfs_need_sync_write(file, inode)) {
679 int err = vfs_fsync(file, 0);
200baa21
TM
680 if (err < 0)
681 result = err;
682 }
7e381172
CL
683 if (result > 0)
684 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
1da177e4
LT
685out:
686 return result;
687
688out_swapfile:
689 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 goto out;
691}
89d77c8f 692EXPORT_SYMBOL_GPL(nfs_file_write);
1da177e4 693
ce4ef7c0
BS
694ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
695 struct file *filp, loff_t *ppos,
696 size_t count, unsigned int flags)
bf40d343 697{
6de1472f 698 struct inode *inode = file_inode(filp);
7e381172 699 unsigned long written = 0;
bf40d343
SJ
700 ssize_t ret;
701
6de1472f
AV
702 dprintk("NFS splice_write(%pD2, %lu@%llu)\n",
703 filp, (unsigned long) count, (unsigned long long) *ppos);
bf40d343
SJ
704
705 /*
706 * The combination of splice and an O_APPEND destination is disallowed.
707 */
708
bf40d343 709 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
7e381172
CL
710 if (ret > 0)
711 written = ret;
712
bf40d343 713 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
af7fa165 714 int err = vfs_fsync(filp, 0);
bf40d343
SJ
715 if (err < 0)
716 ret = err;
717 }
7e381172
CL
718 if (ret > 0)
719 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
bf40d343
SJ
720 return ret;
721}
89d77c8f 722EXPORT_SYMBOL_GPL(nfs_file_splice_write);
bf40d343 723
5eebde23
SJ
724static int
725do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
726{
727 struct inode *inode = filp->f_mapping->host;
728 int status = 0;
21ac19d4 729 unsigned int saved_type = fl->fl_type;
1da177e4 730
039c4d7a 731 /* Try local locking first */
6d34ac19
BF
732 posix_test_lock(filp, fl);
733 if (fl->fl_type != F_UNLCK) {
734 /* found a conflict */
039c4d7a 735 goto out;
1da177e4 736 }
21ac19d4 737 fl->fl_type = saved_type;
039c4d7a 738
011e2a7f 739 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
039c4d7a
TM
740 goto out_noconflict;
741
5eebde23 742 if (is_local)
039c4d7a
TM
743 goto out_noconflict;
744
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746out:
1da177e4 747 return status;
039c4d7a
TM
748out_noconflict:
749 fl->fl_type = F_UNLCK;
750 goto out;
1da177e4
LT
751}
752
753static int do_vfs_lock(struct file *file, struct file_lock *fl)
754{
755 int res = 0;
756 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
757 case FL_POSIX:
758 res = posix_lock_file_wait(file, fl);
759 break;
760 case FL_FLOCK:
761 res = flock_lock_file_wait(file, fl);
762 break;
763 default:
764 BUG();
765 }
1da177e4
LT
766 return res;
767}
768
5eebde23
SJ
769static int
770do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
771{
772 struct inode *inode = filp->f_mapping->host;
7a8203d8 773 struct nfs_lock_context *l_ctx;
1da177e4
LT
774 int status;
775
1da177e4
LT
776 /*
777 * Flush all pending writes before doing anything
778 * with locks..
779 */
29884df0 780 nfs_sync_mapping(filp->f_mapping);
1da177e4 781
7a8203d8
TM
782 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
783 if (!IS_ERR(l_ctx)) {
784 status = nfs_iocounter_wait(&l_ctx->io_count);
785 nfs_put_lock_context(l_ctx);
786 if (status < 0)
787 return status;
788 }
789
1da177e4
LT
790 /* NOTE: special case
791 * If we're signalled while cleaning up locks on process exit, we
792 * still need to complete the unlock.
793 */
5eebde23
SJ
794 /*
795 * Use local locking if mounted with "-onolock" or with appropriate
796 * "-olocal_lock="
797 */
798 if (!is_local)
1da177e4
LT
799 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
800 else
801 status = do_vfs_lock(filp, fl);
1da177e4
LT
802 return status;
803}
804
6b96724e
RL
805static int
806is_time_granular(struct timespec *ts) {
807 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
808}
809
5eebde23
SJ
810static int
811do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
1da177e4
LT
812{
813 struct inode *inode = filp->f_mapping->host;
1da177e4
LT
814 int status;
815
1da177e4
LT
816 /*
817 * Flush all pending writes before doing anything
818 * with locks..
819 */
29884df0
TM
820 status = nfs_sync_mapping(filp->f_mapping);
821 if (status != 0)
1da177e4
LT
822 goto out;
823
5eebde23
SJ
824 /*
825 * Use local locking if mounted with "-onolock" or with appropriate
826 * "-olocal_lock="
827 */
828 if (!is_local)
1da177e4 829 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
c4d7c402 830 else
1da177e4 831 status = do_vfs_lock(filp, fl);
1da177e4
LT
832 if (status < 0)
833 goto out;
6b96724e 834
1da177e4 835 /*
6b96724e
RL
836 * Revalidate the cache if the server has time stamps granular
837 * enough to detect subsecond changes. Otherwise, clear the
838 * cache to prevent missing any changes.
839 *
1da177e4
LT
840 * This makes locking act as a cache coherency point.
841 */
29884df0 842 nfs_sync_mapping(filp->f_mapping);
011e2a7f 843 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
6b96724e
RL
844 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
845 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
846 else
847 nfs_zap_caches(inode);
848 }
1da177e4 849out:
1da177e4
LT
850 return status;
851}
852
853/*
854 * Lock a (portion of) a file
855 */
ce4ef7c0 856int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 857{
6da24bc9 858 struct inode *inode = filp->f_mapping->host;
2116271a 859 int ret = -ENOLCK;
5eebde23 860 int is_local = 0;
1da177e4 861
6de1472f
AV
862 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
863 filp, fl->fl_type, fl->fl_flags,
1da177e4 864 (long long)fl->fl_start, (long long)fl->fl_end);
6da24bc9 865
91d5b470 866 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
1da177e4
LT
867
868 /* No mandatory locks over NFS */
dfad9441 869 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
2116271a
TM
870 goto out_err;
871
5eebde23
SJ
872 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
873 is_local = 1;
874
2116271a
TM
875 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
876 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
877 if (ret < 0)
878 goto out_err;
879 }
1da177e4
LT
880
881 if (IS_GETLK(cmd))
5eebde23 882 ret = do_getlk(filp, cmd, fl, is_local);
2116271a 883 else if (fl->fl_type == F_UNLCK)
5eebde23 884 ret = do_unlk(filp, cmd, fl, is_local);
2116271a 885 else
5eebde23 886 ret = do_setlk(filp, cmd, fl, is_local);
2116271a
TM
887out_err:
888 return ret;
1da177e4 889}
89d77c8f 890EXPORT_SYMBOL_GPL(nfs_lock);
1da177e4
LT
891
892/*
893 * Lock a (portion of) a file
894 */
ce4ef7c0 895int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
1da177e4 896{
5eebde23
SJ
897 struct inode *inode = filp->f_mapping->host;
898 int is_local = 0;
899
6de1472f
AV
900 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
901 filp, fl->fl_type, fl->fl_flags);
1da177e4 902
1da177e4
LT
903 if (!(fl->fl_flags & FL_FLOCK))
904 return -ENOLCK;
905
ad0fcd4e
JL
906 /*
907 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
908 * any standard. In principle we might be able to support LOCK_MAND
909 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
910 * NFS code is not set up for it.
911 */
912 if (fl->fl_type & LOCK_MAND)
913 return -EINVAL;
914
5eebde23
SJ
915 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
916 is_local = 1;
917
1da177e4
LT
918 /* We're simulating flock() locks using posix locks on the server */
919 fl->fl_owner = (fl_owner_t)filp;
920 fl->fl_start = 0;
921 fl->fl_end = OFFSET_MAX;
922
923 if (fl->fl_type == F_UNLCK)
5eebde23
SJ
924 return do_unlk(filp, cmd, fl, is_local);
925 return do_setlk(filp, cmd, fl, is_local);
1da177e4 926}
89d77c8f 927EXPORT_SYMBOL_GPL(nfs_flock);
370f6599 928
6da24bc9
CL
929/*
930 * There is no protocol support for leases, so we have no way to implement
931 * them correctly in the face of opens by other clients.
932 */
ce4ef7c0 933int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
370f6599 934{
6de1472f 935 dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg);
370f6599
BF
936 return -EINVAL;
937}
89d77c8f 938EXPORT_SYMBOL_GPL(nfs_setlease);
1788ea6e 939
0486958f
JL
940const struct file_operations nfs_file_operations = {
941 .llseek = nfs_file_llseek,
942 .read = do_sync_read,
943 .write = do_sync_write,
944 .aio_read = nfs_file_read,
945 .aio_write = nfs_file_write,
946 .mmap = nfs_file_mmap,
947 .open = nfs_file_open,
948 .flush = nfs_file_flush,
949 .release = nfs_file_release,
950 .fsync = nfs_file_fsync,
951 .lock = nfs_lock,
952 .flock = nfs_flock,
953 .splice_read = nfs_file_splice_read,
954 .splice_write = nfs_file_splice_write,
955 .check_flags = nfs_check_flags,
956 .setlease = nfs_setlease,
957};
ddda8e0a 958EXPORT_SYMBOL_GPL(nfs_file_operations);