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