Loading...
1/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
19#include <linux/module.h>
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>
28#include <linux/pagemap.h>
29#include <linux/gfp.h>
30#include <linux/swap.h>
31
32#include <linux/uaccess.h>
33
34#include "delegation.h"
35#include "internal.h"
36#include "iostat.h"
37#include "fscache.h"
38#include "pnfs.h"
39
40#include "nfstrace.h"
41
42#define NFSDBG_FACILITY NFSDBG_FILE
43
44static const struct vm_operations_struct nfs_file_vm_ops;
45
46/* Hack for future NFS swap support */
47#ifndef IS_SWAPFILE
48# define IS_SWAPFILE(inode) (0)
49#endif
50
51int nfs_check_flags(int flags)
52{
53 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
54 return -EINVAL;
55
56 return 0;
57}
58EXPORT_SYMBOL_GPL(nfs_check_flags);
59
60/*
61 * Open file
62 */
63static int
64nfs_file_open(struct inode *inode, struct file *filp)
65{
66 int res;
67
68 dprintk("NFS: open file(%pD2)\n", filp);
69
70 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
71 res = nfs_check_flags(filp->f_flags);
72 if (res)
73 return res;
74
75 res = nfs_open(inode, filp);
76 return res;
77}
78
79int
80nfs_file_release(struct inode *inode, struct file *filp)
81{
82 dprintk("NFS: release(%pD2)\n", filp);
83
84 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
85 nfs_file_clear_open_context(filp);
86 return 0;
87}
88EXPORT_SYMBOL_GPL(nfs_file_release);
89
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
105 if (filp->f_flags & O_DIRECT)
106 goto force_reval;
107 if (nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE))
108 goto force_reval;
109 return 0;
110force_reval:
111 return __nfs_revalidate_inode(server, inode);
112}
113
114loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
115{
116 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
117 filp, offset, whence);
118
119 /*
120 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
121 * the cached file length
122 */
123 if (whence != SEEK_SET && whence != SEEK_CUR) {
124 struct inode *inode = filp->f_mapping->host;
125
126 int retval = nfs_revalidate_file_size(inode, filp);
127 if (retval < 0)
128 return (loff_t)retval;
129 }
130
131 return generic_file_llseek(filp, offset, whence);
132}
133EXPORT_SYMBOL_GPL(nfs_file_llseek);
134
135/*
136 * Flush all dirty pages, and check for write errors.
137 */
138static int
139nfs_file_flush(struct file *file, fl_owner_t id)
140{
141 struct inode *inode = file_inode(file);
142
143 dprintk("NFS: flush(%pD2)\n", file);
144
145 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
146 if ((file->f_mode & FMODE_WRITE) == 0)
147 return 0;
148
149 /* Flush writes to the server and return any errors */
150 return vfs_fsync(file, 0);
151}
152
153ssize_t
154nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
155{
156 struct inode *inode = file_inode(iocb->ki_filp);
157 ssize_t result;
158
159 if (iocb->ki_flags & IOCB_DIRECT)
160 return nfs_file_direct_read(iocb, to);
161
162 dprintk("NFS: read(%pD2, %zu@%lu)\n",
163 iocb->ki_filp,
164 iov_iter_count(to), (unsigned long) iocb->ki_pos);
165
166 nfs_start_io_read(inode);
167 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
168 if (!result) {
169 result = generic_file_read_iter(iocb, to);
170 if (result > 0)
171 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
172 }
173 nfs_end_io_read(inode);
174 return result;
175}
176EXPORT_SYMBOL_GPL(nfs_file_read);
177
178int
179nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
180{
181 struct inode *inode = file_inode(file);
182 int status;
183
184 dprintk("NFS: mmap(%pD2)\n", file);
185
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);
190 if (!status) {
191 vma->vm_ops = &nfs_file_vm_ops;
192 status = nfs_revalidate_mapping(inode, file->f_mapping);
193 }
194 return status;
195}
196EXPORT_SYMBOL_GPL(nfs_file_mmap);
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.
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
207 * nfs_file_write() that a write error occurred, and hence cause it to
208 * fall back to doing a synchronous write.
209 */
210static int
211nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
212{
213 struct nfs_open_context *ctx = nfs_file_open_context(file);
214 struct inode *inode = file_inode(file);
215 int have_error, do_resend, status;
216 int ret = 0;
217
218 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
219
220 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
221 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
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);
225 if (have_error) {
226 ret = xchg(&ctx->error, 0);
227 if (ret)
228 goto out;
229 }
230 if (status < 0) {
231 ret = status;
232 goto out;
233 }
234 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
235 if (do_resend)
236 ret = -EAGAIN;
237out:
238 return ret;
239}
240
241int
242nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
243{
244 int ret;
245 struct inode *inode = file_inode(file);
246
247 trace_nfs_fsync_enter(inode);
248
249 do {
250 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
251 if (ret != 0)
252 break;
253 ret = nfs_file_fsync_commit(file, start, end, datasync);
254 if (!ret)
255 ret = pnfs_sync_inode(inode, !!datasync);
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;
263 } while (ret == -EAGAIN);
264
265 trace_nfs_fsync_exit(inode, ret);
266 return ret;
267}
268EXPORT_SYMBOL_GPL(nfs_file_fsync);
269
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);
294 unsigned int offset = pos & (PAGE_SIZE - 1);
295 unsigned int end = offset + len;
296
297 if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
298 if (!PageUptodate(page))
299 return 1;
300 return 0;
301 }
302
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
312/*
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.
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 */
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)
323{
324 int ret;
325 pgoff_t index = pos >> PAGE_SHIFT;
326 struct page *page;
327 int once_thru = 0;
328
329 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
330 file, mapping->host->i_ino, len, (long long) pos);
331
332start:
333 page = grab_cache_page_write_begin(mapping, index, flags);
334 if (!page)
335 return -ENOMEM;
336 *pagep = page;
337
338 ret = nfs_flush_incompatible(file, page);
339 if (ret) {
340 unlock_page(page);
341 put_page(page);
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);
346 put_page(page);
347 if (!ret)
348 goto start;
349 }
350 return ret;
351}
352
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)
356{
357 unsigned offset = pos & (PAGE_SIZE - 1);
358 struct nfs_open_context *ctx = nfs_file_open_context(file);
359 int status;
360
361 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
362 file, mapping->host->i_ino, len, (long long) pos);
363
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);
370 unsigned end = offset + copied;
371
372 if (pglen == 0) {
373 zero_user_segments(page, 0, offset,
374 end, PAGE_SIZE);
375 SetPageUptodate(page);
376 } else if (end >= pglen) {
377 zero_user_segment(page, end, PAGE_SIZE);
378 if (offset == 0)
379 SetPageUptodate(page);
380 } else
381 zero_user_segment(page, pglen, PAGE_SIZE);
382 }
383
384 status = nfs_updatepage(file, page, offset, copied);
385
386 unlock_page(page);
387 put_page(page);
388
389 if (status < 0)
390 return status;
391 NFS_I(mapping->host)->write_io += copied;
392
393 if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
394 status = nfs_wb_all(mapping->host);
395 if (status < 0)
396 return status;
397 }
398
399 return copied;
400}
401
402/*
403 * Partially or wholly invalidate a page
404 * - Release the private state associated with a page if undergoing complete
405 * page invalidation
406 * - Called if either PG_private or PG_fscache is set on the page
407 * - Caller holds page lock
408 */
409static void nfs_invalidate_page(struct page *page, unsigned int offset,
410 unsigned int length)
411{
412 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
413 page, offset, length);
414
415 if (offset != 0 || length < PAGE_SIZE)
416 return;
417 /* Cancel any unstarted writes on this page */
418 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
419
420 nfs_fscache_invalidate_page(page, page->mapping->host);
421}
422
423/*
424 * Attempt to release the private state associated with a page
425 * - Called if either PG_private or PG_fscache is set on the page
426 * - Caller holds page lock
427 * - Return true (may release page) or false (may not)
428 */
429static int nfs_release_page(struct page *page, gfp_t gfp)
430{
431 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
432
433 /* If PagePrivate() is set, then the page is not freeable */
434 if (PagePrivate(page))
435 return 0;
436 return nfs_fscache_release_page(page, gfp);
437}
438
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);
454 if (atomic_read(&nfsi->commit_info.rpcs_out)) {
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
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
472 * - Called if either PG_private or fscache is set on the page
473 * - Caller holds page lock
474 * - Return 0 if successful, -error otherwise
475 */
476static int nfs_launder_page(struct page *page)
477{
478 struct inode *inode = page_file_mapping(page)->host;
479 struct nfs_inode *nfsi = NFS_I(inode);
480
481 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
482 inode->i_ino, (long long)page_offset(page));
483
484 nfs_fscache_wait_on_page_write(nfsi, page);
485 return nfs_wb_launder_page(inode, page);
486}
487
488static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
489 sector_t *span)
490{
491 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
492
493 *span = sis->pages;
494
495 return rpc_clnt_swap_activate(clnt);
496}
497
498static void nfs_swap_deactivate(struct file *file)
499{
500 struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
501
502 rpc_clnt_swap_deactivate(clnt);
503}
504
505const struct address_space_operations nfs_file_aops = {
506 .readpage = nfs_readpage,
507 .readpages = nfs_readpages,
508 .set_page_dirty = __set_page_dirty_nobuffers,
509 .writepage = nfs_writepage,
510 .writepages = nfs_writepages,
511 .write_begin = nfs_write_begin,
512 .write_end = nfs_write_end,
513 .invalidatepage = nfs_invalidate_page,
514 .releasepage = nfs_release_page,
515 .direct_IO = nfs_direct_IO,
516#ifdef CONFIG_MIGRATION
517 .migratepage = nfs_migrate_page,
518#endif
519 .launder_page = nfs_launder_page,
520 .is_dirty_writeback = nfs_check_dirty_writeback,
521 .error_remove_page = generic_error_remove_page,
522 .swap_activate = nfs_swap_activate,
523 .swap_deactivate = nfs_swap_deactivate,
524};
525
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 */
531static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
532{
533 struct page *page = vmf->page;
534 struct file *filp = vma->vm_file;
535 struct inode *inode = file_inode(filp);
536 unsigned pagelen;
537 int ret = VM_FAULT_NOPAGE;
538 struct address_space *mapping;
539
540 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
541 filp, filp->f_mapping->host->i_ino,
542 (long long)page_offset(page));
543
544 sb_start_pagefault(inode->i_sb);
545
546 /* make sure the cache has finished storing the page */
547 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
548
549 wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
550 nfs_wait_bit_killable, TASK_KILLABLE);
551
552 lock_page(page);
553 mapping = page_file_mapping(page);
554 if (mapping != inode->i_mapping)
555 goto out_unlock;
556
557 wait_on_page_writeback(page);
558
559 pagelen = nfs_page_length(page);
560 if (pagelen == 0)
561 goto out_unlock;
562
563 ret = VM_FAULT_LOCKED;
564 if (nfs_flush_incompatible(filp, page) == 0 &&
565 nfs_updatepage(filp, page, 0, pagelen) == 0)
566 goto out;
567
568 ret = VM_FAULT_SIGBUS;
569out_unlock:
570 unlock_page(page);
571out:
572 sb_end_pagefault(inode->i_sb);
573 return ret;
574}
575
576static const struct vm_operations_struct nfs_file_vm_ops = {
577 .fault = filemap_fault,
578 .map_pages = filemap_map_pages,
579 .page_mkwrite = nfs_vm_page_mkwrite,
580};
581
582static int nfs_need_check_write(struct file *filp, struct inode *inode)
583{
584 struct nfs_open_context *ctx;
585
586 ctx = nfs_file_open_context(filp);
587 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
588 nfs_ctx_key_to_expire(ctx, inode))
589 return 1;
590 return 0;
591}
592
593ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
594{
595 struct file *file = iocb->ki_filp;
596 struct inode *inode = file_inode(file);
597 unsigned long written = 0;
598 ssize_t result;
599
600 result = nfs_key_timeout_notify(file, inode);
601 if (result)
602 return result;
603
604 if (iocb->ki_flags & IOCB_DIRECT)
605 return nfs_file_direct_write(iocb, from);
606
607 dprintk("NFS: write(%pD2, %zu@%Ld)\n",
608 file, iov_iter_count(from), (long long) iocb->ki_pos);
609
610 if (IS_SWAPFILE(inode))
611 goto out_swapfile;
612 /*
613 * O_APPEND implies that we must revalidate the file length.
614 */
615 if (iocb->ki_flags & IOCB_APPEND) {
616 result = nfs_revalidate_file_size(inode, file);
617 if (result)
618 goto out;
619 }
620
621 nfs_start_io_write(inode);
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 }
628 nfs_end_io_write(inode);
629 if (result <= 0)
630 goto out;
631
632 result = generic_write_sync(iocb, result);
633 if (result < 0)
634 goto out;
635 written = result;
636 iocb->ki_pos += written;
637
638 /* Return error values */
639 if (nfs_need_check_write(file, inode)) {
640 int err = vfs_fsync(file, 0);
641 if (err < 0)
642 result = err;
643 }
644 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
645out:
646 return result;
647
648out_swapfile:
649 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
650 return -EBUSY;
651}
652EXPORT_SYMBOL_GPL(nfs_file_write);
653
654static int
655do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
656{
657 struct inode *inode = filp->f_mapping->host;
658 int status = 0;
659 unsigned int saved_type = fl->fl_type;
660
661 /* Try local locking first */
662 posix_test_lock(filp, fl);
663 if (fl->fl_type != F_UNLCK) {
664 /* found a conflict */
665 goto out;
666 }
667 fl->fl_type = saved_type;
668
669 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
670 goto out_noconflict;
671
672 if (is_local)
673 goto out_noconflict;
674
675 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
676out:
677 return status;
678out_noconflict:
679 fl->fl_type = F_UNLCK;
680 goto out;
681}
682
683static int
684do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
685{
686 struct inode *inode = filp->f_mapping->host;
687 struct nfs_lock_context *l_ctx;
688 int status;
689
690 /*
691 * Flush all pending writes before doing anything
692 * with locks..
693 */
694 vfs_fsync(filp, 0);
695
696 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
697 if (!IS_ERR(l_ctx)) {
698 status = nfs_iocounter_wait(l_ctx);
699 nfs_put_lock_context(l_ctx);
700 if (status < 0)
701 return status;
702 }
703
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 */
708 /*
709 * Use local locking if mounted with "-onolock" or with appropriate
710 * "-olocal_lock="
711 */
712 if (!is_local)
713 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
714 else
715 status = locks_lock_file_wait(filp, fl);
716 return status;
717}
718
719static int
720do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
721{
722 struct inode *inode = filp->f_mapping->host;
723 int status;
724
725 /*
726 * Flush all pending writes before doing anything
727 * with locks..
728 */
729 status = nfs_sync_mapping(filp->f_mapping);
730 if (status != 0)
731 goto out;
732
733 /*
734 * Use local locking if mounted with "-onolock" or with appropriate
735 * "-olocal_lock="
736 */
737 if (!is_local)
738 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
739 else
740 status = locks_lock_file_wait(filp, fl);
741 if (status < 0)
742 goto out;
743
744 /*
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 *
749 * This makes locking act as a cache coherency point.
750 */
751 nfs_sync_mapping(filp->f_mapping);
752 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
753 nfs_zap_mapping(inode, filp->f_mapping);
754out:
755 return status;
756}
757
758/*
759 * Lock a (portion of) a file
760 */
761int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
762{
763 struct inode *inode = filp->f_mapping->host;
764 int ret = -ENOLCK;
765 int is_local = 0;
766
767 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
768 filp, fl->fl_type, fl->fl_flags,
769 (long long)fl->fl_start, (long long)fl->fl_end);
770
771 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
772
773 /* No mandatory locks over NFS */
774 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
775 goto out_err;
776
777 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
778 is_local = 1;
779
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 }
785
786 if (IS_GETLK(cmd))
787 ret = do_getlk(filp, cmd, fl, is_local);
788 else if (fl->fl_type == F_UNLCK)
789 ret = do_unlk(filp, cmd, fl, is_local);
790 else
791 ret = do_setlk(filp, cmd, fl, is_local);
792out_err:
793 return ret;
794}
795EXPORT_SYMBOL_GPL(nfs_lock);
796
797/*
798 * Lock a (portion of) a file
799 */
800int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
801{
802 struct inode *inode = filp->f_mapping->host;
803 int is_local = 0;
804
805 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
806 filp, fl->fl_type, fl->fl_flags);
807
808 if (!(fl->fl_flags & FL_FLOCK))
809 return -ENOLCK;
810
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
820 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
821 is_local = 1;
822
823 /* We're simulating flock() locks using posix locks on the server */
824 if (fl->fl_type == F_UNLCK)
825 return do_unlk(filp, cmd, fl, is_local);
826 return do_setlk(filp, cmd, fl, is_local);
827}
828EXPORT_SYMBOL_GPL(nfs_flock);
829
830const struct file_operations nfs_file_operations = {
831 .llseek = nfs_file_llseek,
832 .read_iter = nfs_file_read,
833 .write_iter = nfs_file_write,
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,
841 .splice_read = generic_file_splice_read,
842 .splice_write = iter_file_splice_write,
843 .check_flags = nfs_check_flags,
844 .setlease = simple_nosetlease,
845};
846EXPORT_SYMBOL_GPL(nfs_file_operations);
1/*
2 * linux/fs/nfs/file.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * Changes Copyright (C) 1994 by Florian La Roche
7 * - Do not copy data too often around in the kernel.
8 * - In nfs_file_read the return value of kmalloc wasn't checked.
9 * - Put in a better version of read look-ahead buffering. Original idea
10 * and implementation by Wai S Kok elekokws@ee.nus.sg.
11 *
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
13 *
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
15 *
16 * nfs regular file handling functions
17 */
18
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/errno.h>
22#include <linux/fcntl.h>
23#include <linux/stat.h>
24#include <linux/nfs_fs.h>
25#include <linux/nfs_mount.h>
26#include <linux/mm.h>
27#include <linux/pagemap.h>
28#include <linux/aio.h>
29#include <linux/gfp.h>
30#include <linux/swap.h>
31
32#include <asm/uaccess.h>
33
34#include "delegation.h"
35#include "internal.h"
36#include "iostat.h"
37#include "fscache.h"
38#include "pnfs.h"
39
40#define NFSDBG_FACILITY NFSDBG_FILE
41
42static const struct vm_operations_struct nfs_file_vm_ops;
43
44const struct inode_operations nfs_file_inode_operations = {
45 .permission = nfs_permission,
46 .getattr = nfs_getattr,
47 .setattr = nfs_setattr,
48};
49
50#ifdef CONFIG_NFS_V3
51const struct inode_operations nfs3_file_inode_operations = {
52 .permission = nfs_permission,
53 .getattr = nfs_getattr,
54 .setattr = nfs_setattr,
55 .listxattr = nfs3_listxattr,
56 .getxattr = nfs3_getxattr,
57 .setxattr = nfs3_setxattr,
58 .removexattr = nfs3_removexattr,
59};
60#endif /* CONFIG_NFS_v3 */
61
62/* Hack for future NFS swap support */
63#ifndef IS_SWAPFILE
64# define IS_SWAPFILE(inode) (0)
65#endif
66
67static int nfs_check_flags(int flags)
68{
69 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
70 return -EINVAL;
71
72 return 0;
73}
74
75/*
76 * Open file
77 */
78static int
79nfs_file_open(struct inode *inode, struct file *filp)
80{
81 int res;
82
83 dprintk("NFS: open file(%s/%s)\n",
84 filp->f_path.dentry->d_parent->d_name.name,
85 filp->f_path.dentry->d_name.name);
86
87 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
88 res = nfs_check_flags(filp->f_flags);
89 if (res)
90 return res;
91
92 res = nfs_open(inode, filp);
93 return res;
94}
95
96static int
97nfs_file_release(struct inode *inode, struct file *filp)
98{
99 dprintk("NFS: release(%s/%s)\n",
100 filp->f_path.dentry->d_parent->d_name.name,
101 filp->f_path.dentry->d_name.name);
102
103 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
104 return nfs_release(inode, filp);
105}
106
107/**
108 * nfs_revalidate_size - Revalidate the file size
109 * @inode - pointer to inode struct
110 * @file - pointer to struct file
111 *
112 * Revalidates the file length. This is basically a wrapper around
113 * nfs_revalidate_inode() that takes into account the fact that we may
114 * have cached writes (in which case we don't care about the server's
115 * idea of what the file length is), or O_DIRECT (in which case we
116 * shouldn't trust the cache).
117 */
118static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
119{
120 struct nfs_server *server = NFS_SERVER(inode);
121 struct nfs_inode *nfsi = NFS_I(inode);
122
123 if (nfs_have_delegated_attributes(inode))
124 goto out_noreval;
125
126 if (filp->f_flags & O_DIRECT)
127 goto force_reval;
128 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
129 goto force_reval;
130 if (nfs_attribute_timeout(inode))
131 goto force_reval;
132out_noreval:
133 return 0;
134force_reval:
135 return __nfs_revalidate_inode(server, inode);
136}
137
138static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
139{
140 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
141 filp->f_path.dentry->d_parent->d_name.name,
142 filp->f_path.dentry->d_name.name,
143 offset, origin);
144
145 /*
146 * origin == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
147 * the cached file length
148 */
149 if (origin != SEEK_SET && origin != SEEK_CUR) {
150 struct inode *inode = filp->f_mapping->host;
151
152 int retval = nfs_revalidate_file_size(inode, filp);
153 if (retval < 0)
154 return (loff_t)retval;
155 }
156
157 return generic_file_llseek(filp, offset, origin);
158}
159
160/*
161 * Flush all dirty pages, and check for write errors.
162 */
163static int
164nfs_file_flush(struct file *file, fl_owner_t id)
165{
166 struct dentry *dentry = file->f_path.dentry;
167 struct inode *inode = dentry->d_inode;
168
169 dprintk("NFS: flush(%s/%s)\n",
170 dentry->d_parent->d_name.name,
171 dentry->d_name.name);
172
173 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
174 if ((file->f_mode & FMODE_WRITE) == 0)
175 return 0;
176
177 /*
178 * If we're holding a write delegation, then just start the i/o
179 * but don't wait for completion (or send a commit).
180 */
181 if (nfs_have_delegation(inode, FMODE_WRITE))
182 return filemap_fdatawrite(file->f_mapping);
183
184 /* Flush writes to the server and return any errors */
185 return vfs_fsync(file, 0);
186}
187
188static ssize_t
189nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
190 unsigned long nr_segs, loff_t pos)
191{
192 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
193 struct inode * inode = dentry->d_inode;
194 ssize_t result;
195
196 if (iocb->ki_filp->f_flags & O_DIRECT)
197 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
198
199 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
200 dentry->d_parent->d_name.name, dentry->d_name.name,
201 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
202
203 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
204 if (!result) {
205 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
206 if (result > 0)
207 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
208 }
209 return result;
210}
211
212static ssize_t
213nfs_file_splice_read(struct file *filp, loff_t *ppos,
214 struct pipe_inode_info *pipe, size_t count,
215 unsigned int flags)
216{
217 struct dentry *dentry = filp->f_path.dentry;
218 struct inode *inode = dentry->d_inode;
219 ssize_t res;
220
221 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
222 dentry->d_parent->d_name.name, dentry->d_name.name,
223 (unsigned long) count, (unsigned long long) *ppos);
224
225 res = nfs_revalidate_mapping(inode, filp->f_mapping);
226 if (!res) {
227 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
228 if (res > 0)
229 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
230 }
231 return res;
232}
233
234static int
235nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
236{
237 struct dentry *dentry = file->f_path.dentry;
238 struct inode *inode = dentry->d_inode;
239 int status;
240
241 dprintk("NFS: mmap(%s/%s)\n",
242 dentry->d_parent->d_name.name, dentry->d_name.name);
243
244 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
245 * so we call that before revalidating the mapping
246 */
247 status = generic_file_mmap(file, vma);
248 if (!status) {
249 vma->vm_ops = &nfs_file_vm_ops;
250 status = nfs_revalidate_mapping(inode, file->f_mapping);
251 }
252 return status;
253}
254
255/*
256 * Flush any dirty pages for this process, and check for write errors.
257 * The return status from this call provides a reliable indication of
258 * whether any write errors occurred for this process.
259 *
260 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
261 * disk, but it retrieves and clears ctx->error after synching, despite
262 * the two being set at the same time in nfs_context_set_write_error().
263 * This is because the former is used to notify the _next_ call to
264 * nfs_file_write() that a write error occurred, and hence cause it to
265 * fall back to doing a synchronous write.
266 */
267static int
268nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
269{
270 struct dentry *dentry = file->f_path.dentry;
271 struct nfs_open_context *ctx = nfs_file_open_context(file);
272 struct inode *inode = dentry->d_inode;
273 int have_error, status;
274 int ret = 0;
275
276 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
277 dentry->d_parent->d_name.name, dentry->d_name.name,
278 datasync);
279
280 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
281 mutex_lock(&inode->i_mutex);
282
283 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
284 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
285 status = nfs_commit_inode(inode, FLUSH_SYNC);
286 if (status >= 0 && ret < 0)
287 status = ret;
288 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
289 if (have_error)
290 ret = xchg(&ctx->error, 0);
291 if (!ret && status < 0)
292 ret = status;
293 if (!ret && !datasync)
294 /* application has asked for meta-data sync */
295 ret = pnfs_layoutcommit_inode(inode, true);
296 mutex_unlock(&inode->i_mutex);
297 return ret;
298}
299
300/*
301 * Decide whether a read/modify/write cycle may be more efficient
302 * then a modify/write/read cycle when writing to a page in the
303 * page cache.
304 *
305 * The modify/write/read cycle may occur if a page is read before
306 * being completely filled by the writer. In this situation, the
307 * page must be completely written to stable storage on the server
308 * before it can be refilled by reading in the page from the server.
309 * This can lead to expensive, small, FILE_SYNC mode writes being
310 * done.
311 *
312 * It may be more efficient to read the page first if the file is
313 * open for reading in addition to writing, the page is not marked
314 * as Uptodate, it is not dirty or waiting to be committed,
315 * indicating that it was previously allocated and then modified,
316 * that there were valid bytes of data in that range of the file,
317 * and that the new data won't completely replace the old data in
318 * that range of the file.
319 */
320static int nfs_want_read_modify_write(struct file *file, struct page *page,
321 loff_t pos, unsigned len)
322{
323 unsigned int pglen = nfs_page_length(page);
324 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
325 unsigned int end = offset + len;
326
327 if ((file->f_mode & FMODE_READ) && /* open for read? */
328 !PageUptodate(page) && /* Uptodate? */
329 !PagePrivate(page) && /* i/o request already? */
330 pglen && /* valid bytes of file? */
331 (end < pglen || offset)) /* replace all valid bytes? */
332 return 1;
333 return 0;
334}
335
336/*
337 * This does the "real" work of the write. We must allocate and lock the
338 * page to be sent back to the generic routine, which then copies the
339 * data from user space.
340 *
341 * If the writer ends up delaying the write, the writer needs to
342 * increment the page use counts until he is done with the page.
343 */
344static int nfs_write_begin(struct file *file, struct address_space *mapping,
345 loff_t pos, unsigned len, unsigned flags,
346 struct page **pagep, void **fsdata)
347{
348 int ret;
349 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
350 struct page *page;
351 int once_thru = 0;
352
353 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
354 file->f_path.dentry->d_parent->d_name.name,
355 file->f_path.dentry->d_name.name,
356 mapping->host->i_ino, len, (long long) pos);
357
358start:
359 /*
360 * Prevent starvation issues if someone is doing a consistency
361 * sync-to-disk
362 */
363 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
364 nfs_wait_bit_killable, TASK_KILLABLE);
365 if (ret)
366 return ret;
367
368 page = grab_cache_page_write_begin(mapping, index, flags);
369 if (!page)
370 return -ENOMEM;
371 *pagep = page;
372
373 ret = nfs_flush_incompatible(file, page);
374 if (ret) {
375 unlock_page(page);
376 page_cache_release(page);
377 } else if (!once_thru &&
378 nfs_want_read_modify_write(file, page, pos, len)) {
379 once_thru = 1;
380 ret = nfs_readpage(file, page);
381 page_cache_release(page);
382 if (!ret)
383 goto start;
384 }
385 return ret;
386}
387
388static int nfs_write_end(struct file *file, struct address_space *mapping,
389 loff_t pos, unsigned len, unsigned copied,
390 struct page *page, void *fsdata)
391{
392 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
393 int status;
394
395 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
396 file->f_path.dentry->d_parent->d_name.name,
397 file->f_path.dentry->d_name.name,
398 mapping->host->i_ino, len, (long long) pos);
399
400 /*
401 * Zero any uninitialised parts of the page, and then mark the page
402 * as up to date if it turns out that we're extending the file.
403 */
404 if (!PageUptodate(page)) {
405 unsigned pglen = nfs_page_length(page);
406 unsigned end = offset + len;
407
408 if (pglen == 0) {
409 zero_user_segments(page, 0, offset,
410 end, PAGE_CACHE_SIZE);
411 SetPageUptodate(page);
412 } else if (end >= pglen) {
413 zero_user_segment(page, end, PAGE_CACHE_SIZE);
414 if (offset == 0)
415 SetPageUptodate(page);
416 } else
417 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
418 }
419
420 status = nfs_updatepage(file, page, offset, copied);
421
422 unlock_page(page);
423 page_cache_release(page);
424
425 if (status < 0)
426 return status;
427 NFS_I(mapping->host)->write_io += copied;
428 return copied;
429}
430
431/*
432 * Partially or wholly invalidate a page
433 * - Release the private state associated with a page if undergoing complete
434 * page invalidation
435 * - Called if either PG_private or PG_fscache is set on the page
436 * - Caller holds page lock
437 */
438static void nfs_invalidate_page(struct page *page, unsigned long offset)
439{
440 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
441
442 if (offset != 0)
443 return;
444 /* Cancel any unstarted writes on this page */
445 nfs_wb_page_cancel(page->mapping->host, page);
446
447 nfs_fscache_invalidate_page(page, page->mapping->host);
448}
449
450/*
451 * Attempt to release the private state associated with a page
452 * - Called if either PG_private or PG_fscache is set on the page
453 * - Caller holds page lock
454 * - Return true (may release page) or false (may not)
455 */
456static int nfs_release_page(struct page *page, gfp_t gfp)
457{
458 struct address_space *mapping = page->mapping;
459
460 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
461
462 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
463 * doing this memory reclaim for a fs-related allocation.
464 */
465 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
466 !(current->flags & PF_FSTRANS)) {
467 int how = FLUSH_SYNC;
468
469 /* Don't let kswapd deadlock waiting for OOM RPC calls */
470 if (current_is_kswapd())
471 how = 0;
472 nfs_commit_inode(mapping->host, how);
473 }
474 /* If PagePrivate() is set, then the page is not freeable */
475 if (PagePrivate(page))
476 return 0;
477 return nfs_fscache_release_page(page, gfp);
478}
479
480/*
481 * Attempt to clear the private state associated with a page when an error
482 * occurs that requires the cached contents of an inode to be written back or
483 * destroyed
484 * - Called if either PG_private or fscache is set on the page
485 * - Caller holds page lock
486 * - Return 0 if successful, -error otherwise
487 */
488static int nfs_launder_page(struct page *page)
489{
490 struct inode *inode = page->mapping->host;
491 struct nfs_inode *nfsi = NFS_I(inode);
492
493 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
494 inode->i_ino, (long long)page_offset(page));
495
496 nfs_fscache_wait_on_page_write(nfsi, page);
497 return nfs_wb_page(inode, page);
498}
499
500const struct address_space_operations nfs_file_aops = {
501 .readpage = nfs_readpage,
502 .readpages = nfs_readpages,
503 .set_page_dirty = __set_page_dirty_nobuffers,
504 .writepage = nfs_writepage,
505 .writepages = nfs_writepages,
506 .write_begin = nfs_write_begin,
507 .write_end = nfs_write_end,
508 .invalidatepage = nfs_invalidate_page,
509 .releasepage = nfs_release_page,
510 .direct_IO = nfs_direct_IO,
511 .migratepage = nfs_migrate_page,
512 .launder_page = nfs_launder_page,
513 .error_remove_page = generic_error_remove_page,
514};
515
516/*
517 * Notification that a PTE pointing to an NFS page is about to be made
518 * writable, implying that someone is about to modify the page through a
519 * shared-writable mapping
520 */
521static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
522{
523 struct page *page = vmf->page;
524 struct file *filp = vma->vm_file;
525 struct dentry *dentry = filp->f_path.dentry;
526 unsigned pagelen;
527 int ret = VM_FAULT_NOPAGE;
528 struct address_space *mapping;
529
530 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
531 dentry->d_parent->d_name.name, dentry->d_name.name,
532 filp->f_mapping->host->i_ino,
533 (long long)page_offset(page));
534
535 /* make sure the cache has finished storing the page */
536 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
537
538 lock_page(page);
539 mapping = page->mapping;
540 if (mapping != dentry->d_inode->i_mapping)
541 goto out_unlock;
542
543 wait_on_page_writeback(page);
544
545 pagelen = nfs_page_length(page);
546 if (pagelen == 0)
547 goto out_unlock;
548
549 ret = VM_FAULT_LOCKED;
550 if (nfs_flush_incompatible(filp, page) == 0 &&
551 nfs_updatepage(filp, page, 0, pagelen) == 0)
552 goto out;
553
554 ret = VM_FAULT_SIGBUS;
555out_unlock:
556 unlock_page(page);
557out:
558 return ret;
559}
560
561static const struct vm_operations_struct nfs_file_vm_ops = {
562 .fault = filemap_fault,
563 .page_mkwrite = nfs_vm_page_mkwrite,
564};
565
566static int nfs_need_sync_write(struct file *filp, struct inode *inode)
567{
568 struct nfs_open_context *ctx;
569
570 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
571 return 1;
572 ctx = nfs_file_open_context(filp);
573 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
574 return 1;
575 return 0;
576}
577
578static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
579 unsigned long nr_segs, loff_t pos)
580{
581 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
582 struct inode * inode = dentry->d_inode;
583 unsigned long written = 0;
584 ssize_t result;
585 size_t count = iov_length(iov, nr_segs);
586
587 if (iocb->ki_filp->f_flags & O_DIRECT)
588 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
589
590 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
591 dentry->d_parent->d_name.name, dentry->d_name.name,
592 (unsigned long) count, (long long) pos);
593
594 result = -EBUSY;
595 if (IS_SWAPFILE(inode))
596 goto out_swapfile;
597 /*
598 * O_APPEND implies that we must revalidate the file length.
599 */
600 if (iocb->ki_filp->f_flags & O_APPEND) {
601 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
602 if (result)
603 goto out;
604 }
605
606 result = count;
607 if (!count)
608 goto out;
609
610 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
611 if (result > 0)
612 written = result;
613
614 /* Return error values for O_DSYNC and IS_SYNC() */
615 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
616 int err = vfs_fsync(iocb->ki_filp, 0);
617 if (err < 0)
618 result = err;
619 }
620 if (result > 0)
621 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
622out:
623 return result;
624
625out_swapfile:
626 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
627 goto out;
628}
629
630static ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
631 struct file *filp, loff_t *ppos,
632 size_t count, unsigned int flags)
633{
634 struct dentry *dentry = filp->f_path.dentry;
635 struct inode *inode = dentry->d_inode;
636 unsigned long written = 0;
637 ssize_t ret;
638
639 dprintk("NFS splice_write(%s/%s, %lu@%llu)\n",
640 dentry->d_parent->d_name.name, dentry->d_name.name,
641 (unsigned long) count, (unsigned long long) *ppos);
642
643 /*
644 * The combination of splice and an O_APPEND destination is disallowed.
645 */
646
647 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
648 if (ret > 0)
649 written = ret;
650
651 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
652 int err = vfs_fsync(filp, 0);
653 if (err < 0)
654 ret = err;
655 }
656 if (ret > 0)
657 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
658 return ret;
659}
660
661static int
662do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
663{
664 struct inode *inode = filp->f_mapping->host;
665 int status = 0;
666 unsigned int saved_type = fl->fl_type;
667
668 /* Try local locking first */
669 posix_test_lock(filp, fl);
670 if (fl->fl_type != F_UNLCK) {
671 /* found a conflict */
672 goto out;
673 }
674 fl->fl_type = saved_type;
675
676 if (nfs_have_delegation(inode, FMODE_READ))
677 goto out_noconflict;
678
679 if (is_local)
680 goto out_noconflict;
681
682 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
683out:
684 return status;
685out_noconflict:
686 fl->fl_type = F_UNLCK;
687 goto out;
688}
689
690static int do_vfs_lock(struct file *file, struct file_lock *fl)
691{
692 int res = 0;
693 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
694 case FL_POSIX:
695 res = posix_lock_file_wait(file, fl);
696 break;
697 case FL_FLOCK:
698 res = flock_lock_file_wait(file, fl);
699 break;
700 default:
701 BUG();
702 }
703 return res;
704}
705
706static int
707do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
708{
709 struct inode *inode = filp->f_mapping->host;
710 int status;
711
712 /*
713 * Flush all pending writes before doing anything
714 * with locks..
715 */
716 nfs_sync_mapping(filp->f_mapping);
717
718 /* NOTE: special case
719 * If we're signalled while cleaning up locks on process exit, we
720 * still need to complete the unlock.
721 */
722 /*
723 * Use local locking if mounted with "-onolock" or with appropriate
724 * "-olocal_lock="
725 */
726 if (!is_local)
727 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
728 else
729 status = do_vfs_lock(filp, fl);
730 return status;
731}
732
733static int
734is_time_granular(struct timespec *ts) {
735 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
736}
737
738static int
739do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
740{
741 struct inode *inode = filp->f_mapping->host;
742 int status;
743
744 /*
745 * Flush all pending writes before doing anything
746 * with locks..
747 */
748 status = nfs_sync_mapping(filp->f_mapping);
749 if (status != 0)
750 goto out;
751
752 /*
753 * Use local locking if mounted with "-onolock" or with appropriate
754 * "-olocal_lock="
755 */
756 if (!is_local)
757 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
758 else
759 status = do_vfs_lock(filp, fl);
760 if (status < 0)
761 goto out;
762
763 /*
764 * Revalidate the cache if the server has time stamps granular
765 * enough to detect subsecond changes. Otherwise, clear the
766 * cache to prevent missing any changes.
767 *
768 * This makes locking act as a cache coherency point.
769 */
770 nfs_sync_mapping(filp->f_mapping);
771 if (!nfs_have_delegation(inode, FMODE_READ)) {
772 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
773 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
774 else
775 nfs_zap_caches(inode);
776 }
777out:
778 return status;
779}
780
781/*
782 * Lock a (portion of) a file
783 */
784static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
785{
786 struct inode *inode = filp->f_mapping->host;
787 int ret = -ENOLCK;
788 int is_local = 0;
789
790 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
791 filp->f_path.dentry->d_parent->d_name.name,
792 filp->f_path.dentry->d_name.name,
793 fl->fl_type, fl->fl_flags,
794 (long long)fl->fl_start, (long long)fl->fl_end);
795
796 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
797
798 /* No mandatory locks over NFS */
799 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
800 goto out_err;
801
802 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
803 is_local = 1;
804
805 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
806 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
807 if (ret < 0)
808 goto out_err;
809 }
810
811 if (IS_GETLK(cmd))
812 ret = do_getlk(filp, cmd, fl, is_local);
813 else if (fl->fl_type == F_UNLCK)
814 ret = do_unlk(filp, cmd, fl, is_local);
815 else
816 ret = do_setlk(filp, cmd, fl, is_local);
817out_err:
818 return ret;
819}
820
821/*
822 * Lock a (portion of) a file
823 */
824static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
825{
826 struct inode *inode = filp->f_mapping->host;
827 int is_local = 0;
828
829 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
830 filp->f_path.dentry->d_parent->d_name.name,
831 filp->f_path.dentry->d_name.name,
832 fl->fl_type, fl->fl_flags);
833
834 if (!(fl->fl_flags & FL_FLOCK))
835 return -ENOLCK;
836
837 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
838 is_local = 1;
839
840 /* We're simulating flock() locks using posix locks on the server */
841 fl->fl_owner = (fl_owner_t)filp;
842 fl->fl_start = 0;
843 fl->fl_end = OFFSET_MAX;
844
845 if (fl->fl_type == F_UNLCK)
846 return do_unlk(filp, cmd, fl, is_local);
847 return do_setlk(filp, cmd, fl, is_local);
848}
849
850/*
851 * There is no protocol support for leases, so we have no way to implement
852 * them correctly in the face of opens by other clients.
853 */
854static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
855{
856 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
857 file->f_path.dentry->d_parent->d_name.name,
858 file->f_path.dentry->d_name.name, arg);
859 return -EINVAL;
860}
861
862const struct file_operations nfs_file_operations = {
863 .llseek = nfs_file_llseek,
864 .read = do_sync_read,
865 .write = do_sync_write,
866 .aio_read = nfs_file_read,
867 .aio_write = nfs_file_write,
868 .mmap = nfs_file_mmap,
869 .open = nfs_file_open,
870 .flush = nfs_file_flush,
871 .release = nfs_file_release,
872 .fsync = nfs_file_fsync,
873 .lock = nfs_lock,
874 .flock = nfs_flock,
875 .splice_read = nfs_file_splice_read,
876 .splice_write = nfs_file_splice_write,
877 .check_flags = nfs_check_flags,
878 .setlease = nfs_setlease,
879};
880
881#ifdef CONFIG_NFS_V4
882static int
883nfs4_file_open(struct inode *inode, struct file *filp)
884{
885 struct nfs_open_context *ctx;
886 struct dentry *dentry = filp->f_path.dentry;
887 struct dentry *parent = NULL;
888 struct inode *dir;
889 unsigned openflags = filp->f_flags;
890 struct iattr attr;
891 int err;
892
893 BUG_ON(inode != dentry->d_inode);
894 /*
895 * If no cached dentry exists or if it's negative, NFSv4 handled the
896 * opens in ->lookup() or ->create().
897 *
898 * We only get this far for a cached positive dentry. We skipped
899 * revalidation, so handle it here by dropping the dentry and returning
900 * -EOPENSTALE. The VFS will retry the lookup/create/open.
901 */
902
903 dprintk("NFS: open file(%s/%s)\n",
904 dentry->d_parent->d_name.name,
905 dentry->d_name.name);
906
907 if ((openflags & O_ACCMODE) == 3)
908 openflags--;
909
910 /* We can't create new files here */
911 openflags &= ~(O_CREAT|O_EXCL);
912
913 parent = dget_parent(dentry);
914 dir = parent->d_inode;
915
916 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
917 err = PTR_ERR(ctx);
918 if (IS_ERR(ctx))
919 goto out;
920
921 attr.ia_valid = ATTR_OPEN;
922 if (openflags & O_TRUNC) {
923 attr.ia_valid |= ATTR_SIZE;
924 attr.ia_size = 0;
925 nfs_wb_all(inode);
926 }
927
928 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr);
929 if (IS_ERR(inode)) {
930 err = PTR_ERR(inode);
931 switch (err) {
932 case -EPERM:
933 case -EACCES:
934 case -EDQUOT:
935 case -ENOSPC:
936 case -EROFS:
937 goto out_put_ctx;
938 default:
939 goto out_drop;
940 }
941 }
942 iput(inode);
943 if (inode != dentry->d_inode)
944 goto out_drop;
945
946 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
947 nfs_file_set_open_context(filp, ctx);
948 err = 0;
949
950out_put_ctx:
951 put_nfs_open_context(ctx);
952out:
953 dput(parent);
954 return err;
955
956out_drop:
957 d_drop(dentry);
958 err = -EOPENSTALE;
959 goto out_put_ctx;
960}
961
962const struct file_operations nfs4_file_operations = {
963 .llseek = nfs_file_llseek,
964 .read = do_sync_read,
965 .write = do_sync_write,
966 .aio_read = nfs_file_read,
967 .aio_write = nfs_file_write,
968 .mmap = nfs_file_mmap,
969 .open = nfs4_file_open,
970 .flush = nfs_file_flush,
971 .release = nfs_file_release,
972 .fsync = nfs_file_fsync,
973 .lock = nfs_lock,
974 .flock = nfs_flock,
975 .splice_read = nfs_file_splice_read,
976 .splice_write = nfs_file_splice_write,
977 .check_flags = nfs_check_flags,
978 .setlease = nfs_setlease,
979};
980#endif /* CONFIG_NFS_V4 */