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/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/aio.h>
30#include <linux/gfp.h>
31#include <linux/swap.h>
32
33#include <asm/uaccess.h>
34
35#include "delegation.h"
36#include "internal.h"
37#include "iostat.h"
38#include "fscache.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 return nfs_release(inode, filp);
86}
87EXPORT_SYMBOL_GPL(nfs_file_release);
88
89/**
90 * nfs_revalidate_size - Revalidate the file size
91 * @inode - pointer to inode struct
92 * @file - pointer to struct file
93 *
94 * Revalidates the file length. This is basically a wrapper around
95 * nfs_revalidate_inode() that takes into account the fact that we may
96 * have cached writes (in which case we don't care about the server's
97 * idea of what the file length is), or O_DIRECT (in which case we
98 * shouldn't trust the cache).
99 */
100static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
101{
102 struct nfs_server *server = NFS_SERVER(inode);
103 struct nfs_inode *nfsi = NFS_I(inode);
104
105 if (nfs_have_delegated_attributes(inode))
106 goto out_noreval;
107
108 if (filp->f_flags & O_DIRECT)
109 goto force_reval;
110 if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
111 goto force_reval;
112 if (nfs_attribute_timeout(inode))
113 goto force_reval;
114out_noreval:
115 return 0;
116force_reval:
117 return __nfs_revalidate_inode(server, inode);
118}
119
120loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
121{
122 dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
123 filp, offset, whence);
124
125 /*
126 * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
127 * the cached file length
128 */
129 if (whence != SEEK_SET && whence != SEEK_CUR) {
130 struct inode *inode = filp->f_mapping->host;
131
132 int retval = nfs_revalidate_file_size(inode, filp);
133 if (retval < 0)
134 return (loff_t)retval;
135 }
136
137 return generic_file_llseek(filp, offset, whence);
138}
139EXPORT_SYMBOL_GPL(nfs_file_llseek);
140
141/*
142 * Flush all dirty pages, and check for write errors.
143 */
144int
145nfs_file_flush(struct file *file, fl_owner_t id)
146{
147 struct inode *inode = file_inode(file);
148
149 dprintk("NFS: flush(%pD2)\n", file);
150
151 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
152 if ((file->f_mode & FMODE_WRITE) == 0)
153 return 0;
154
155 /*
156 * If we're holding a write delegation, then just start the i/o
157 * but don't wait for completion (or send a commit).
158 */
159 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
160 return filemap_fdatawrite(file->f_mapping);
161
162 /* Flush writes to the server and return any errors */
163 return vfs_fsync(file, 0);
164}
165EXPORT_SYMBOL_GPL(nfs_file_flush);
166
167ssize_t
168nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
169 unsigned long nr_segs, loff_t pos)
170{
171 struct inode *inode = file_inode(iocb->ki_filp);
172 ssize_t result;
173
174 if (iocb->ki_filp->f_flags & O_DIRECT)
175 return nfs_file_direct_read(iocb, iov, nr_segs, pos, true);
176
177 dprintk("NFS: read(%pD2, %lu@%lu)\n",
178 iocb->ki_filp,
179 (unsigned long) iov_length(iov, nr_segs), (unsigned long) pos);
180
181 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
182 if (!result) {
183 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
184 if (result > 0)
185 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
186 }
187 return result;
188}
189EXPORT_SYMBOL_GPL(nfs_file_read);
190
191ssize_t
192nfs_file_splice_read(struct file *filp, loff_t *ppos,
193 struct pipe_inode_info *pipe, size_t count,
194 unsigned int flags)
195{
196 struct inode *inode = file_inode(filp);
197 ssize_t res;
198
199 dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
200 filp, (unsigned long) count, (unsigned long long) *ppos);
201
202 res = nfs_revalidate_mapping(inode, filp->f_mapping);
203 if (!res) {
204 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
205 if (res > 0)
206 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
207 }
208 return res;
209}
210EXPORT_SYMBOL_GPL(nfs_file_splice_read);
211
212int
213nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
214{
215 struct inode *inode = file_inode(file);
216 int status;
217
218 dprintk("NFS: mmap(%pD2)\n", file);
219
220 /* Note: generic_file_mmap() returns ENOSYS on nommu systems
221 * so we call that before revalidating the mapping
222 */
223 status = generic_file_mmap(file, vma);
224 if (!status) {
225 vma->vm_ops = &nfs_file_vm_ops;
226 status = nfs_revalidate_mapping(inode, file->f_mapping);
227 }
228 return status;
229}
230EXPORT_SYMBOL_GPL(nfs_file_mmap);
231
232/*
233 * Flush any dirty pages for this process, and check for write errors.
234 * The return status from this call provides a reliable indication of
235 * whether any write errors occurred for this process.
236 *
237 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
238 * disk, but it retrieves and clears ctx->error after synching, despite
239 * the two being set at the same time in nfs_context_set_write_error().
240 * This is because the former is used to notify the _next_ call to
241 * nfs_file_write() that a write error occurred, and hence cause it to
242 * fall back to doing a synchronous write.
243 */
244int
245nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
246{
247 struct nfs_open_context *ctx = nfs_file_open_context(file);
248 struct inode *inode = file_inode(file);
249 int have_error, do_resend, status;
250 int ret = 0;
251
252 dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
253
254 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
255 do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
256 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
257 status = nfs_commit_inode(inode, FLUSH_SYNC);
258 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
259 if (have_error) {
260 ret = xchg(&ctx->error, 0);
261 if (ret)
262 goto out;
263 }
264 if (status < 0) {
265 ret = status;
266 goto out;
267 }
268 do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
269 if (do_resend)
270 ret = -EAGAIN;
271out:
272 return ret;
273}
274EXPORT_SYMBOL_GPL(nfs_file_fsync_commit);
275
276static int
277nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
278{
279 int ret;
280 struct inode *inode = file_inode(file);
281
282 trace_nfs_fsync_enter(inode);
283
284 do {
285 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
286 if (ret != 0)
287 break;
288 mutex_lock(&inode->i_mutex);
289 ret = nfs_file_fsync_commit(file, start, end, datasync);
290 mutex_unlock(&inode->i_mutex);
291 /*
292 * If nfs_file_fsync_commit detected a server reboot, then
293 * resend all dirty pages that might have been covered by
294 * the NFS_CONTEXT_RESEND_WRITES flag
295 */
296 start = 0;
297 end = LLONG_MAX;
298 } while (ret == -EAGAIN);
299
300 trace_nfs_fsync_exit(inode, ret);
301 return ret;
302}
303
304/*
305 * Decide whether a read/modify/write cycle may be more efficient
306 * then a modify/write/read cycle when writing to a page in the
307 * page cache.
308 *
309 * The modify/write/read cycle may occur if a page is read before
310 * being completely filled by the writer. In this situation, the
311 * page must be completely written to stable storage on the server
312 * before it can be refilled by reading in the page from the server.
313 * This can lead to expensive, small, FILE_SYNC mode writes being
314 * done.
315 *
316 * It may be more efficient to read the page first if the file is
317 * open for reading in addition to writing, the page is not marked
318 * as Uptodate, it is not dirty or waiting to be committed,
319 * indicating that it was previously allocated and then modified,
320 * that there were valid bytes of data in that range of the file,
321 * and that the new data won't completely replace the old data in
322 * that range of the file.
323 */
324static int nfs_want_read_modify_write(struct file *file, struct page *page,
325 loff_t pos, unsigned len)
326{
327 unsigned int pglen = nfs_page_length(page);
328 unsigned int offset = pos & (PAGE_CACHE_SIZE - 1);
329 unsigned int end = offset + len;
330
331 if ((file->f_mode & FMODE_READ) && /* open for read? */
332 !PageUptodate(page) && /* Uptodate? */
333 !PagePrivate(page) && /* i/o request already? */
334 pglen && /* valid bytes of file? */
335 (end < pglen || offset)) /* replace all valid bytes? */
336 return 1;
337 return 0;
338}
339
340/*
341 * This does the "real" work of the write. We must allocate and lock the
342 * page to be sent back to the generic routine, which then copies the
343 * data from user space.
344 *
345 * If the writer ends up delaying the write, the writer needs to
346 * increment the page use counts until he is done with the page.
347 */
348static int nfs_write_begin(struct file *file, struct address_space *mapping,
349 loff_t pos, unsigned len, unsigned flags,
350 struct page **pagep, void **fsdata)
351{
352 int ret;
353 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
354 struct page *page;
355 int once_thru = 0;
356
357 dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
358 file, mapping->host->i_ino, len, (long long) pos);
359
360start:
361 /*
362 * Prevent starvation issues if someone is doing a consistency
363 * sync-to-disk
364 */
365 ret = wait_on_bit(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING,
366 nfs_wait_bit_killable, TASK_KILLABLE);
367 if (ret)
368 return ret;
369
370 page = grab_cache_page_write_begin(mapping, index, flags);
371 if (!page)
372 return -ENOMEM;
373 *pagep = page;
374
375 ret = nfs_flush_incompatible(file, page);
376 if (ret) {
377 unlock_page(page);
378 page_cache_release(page);
379 } else if (!once_thru &&
380 nfs_want_read_modify_write(file, page, pos, len)) {
381 once_thru = 1;
382 ret = nfs_readpage(file, page);
383 page_cache_release(page);
384 if (!ret)
385 goto start;
386 }
387 return ret;
388}
389
390static int nfs_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
393{
394 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
395 struct nfs_open_context *ctx = nfs_file_open_context(file);
396 int status;
397
398 dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
399 file, mapping->host->i_ino, len, (long long) pos);
400
401 /*
402 * Zero any uninitialised parts of the page, and then mark the page
403 * as up to date if it turns out that we're extending the file.
404 */
405 if (!PageUptodate(page)) {
406 unsigned pglen = nfs_page_length(page);
407 unsigned end = offset + len;
408
409 if (pglen == 0) {
410 zero_user_segments(page, 0, offset,
411 end, PAGE_CACHE_SIZE);
412 SetPageUptodate(page);
413 } else if (end >= pglen) {
414 zero_user_segment(page, end, PAGE_CACHE_SIZE);
415 if (offset == 0)
416 SetPageUptodate(page);
417 } else
418 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
419 }
420
421 status = nfs_updatepage(file, page, offset, copied);
422
423 unlock_page(page);
424 page_cache_release(page);
425
426 if (status < 0)
427 return status;
428 NFS_I(mapping->host)->write_io += copied;
429
430 if (nfs_ctx_key_to_expire(ctx)) {
431 status = nfs_wb_all(mapping->host);
432 if (status < 0)
433 return status;
434 }
435
436 return copied;
437}
438
439/*
440 * Partially or wholly invalidate a page
441 * - Release the private state associated with a page if undergoing complete
442 * page invalidation
443 * - Called if either PG_private or PG_fscache is set on the page
444 * - Caller holds page lock
445 */
446static void nfs_invalidate_page(struct page *page, unsigned int offset,
447 unsigned int length)
448{
449 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
450 page, offset, length);
451
452 if (offset != 0 || length < PAGE_CACHE_SIZE)
453 return;
454 /* Cancel any unstarted writes on this page */
455 nfs_wb_page_cancel(page_file_mapping(page)->host, page);
456
457 nfs_fscache_invalidate_page(page, page->mapping->host);
458}
459
460/*
461 * Attempt to release the private state associated with a page
462 * - Called if either PG_private or PG_fscache is set on the page
463 * - Caller holds page lock
464 * - Return true (may release page) or false (may not)
465 */
466static int nfs_release_page(struct page *page, gfp_t gfp)
467{
468 struct address_space *mapping = page->mapping;
469
470 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
471
472 /* Only do I/O if gfp is a superset of GFP_KERNEL, and we're not
473 * doing this memory reclaim for a fs-related allocation.
474 */
475 if (mapping && (gfp & GFP_KERNEL) == GFP_KERNEL &&
476 !(current->flags & PF_FSTRANS)) {
477 int how = FLUSH_SYNC;
478
479 /* Don't let kswapd deadlock waiting for OOM RPC calls */
480 if (current_is_kswapd())
481 how = 0;
482 nfs_commit_inode(mapping->host, how);
483 }
484 /* If PagePrivate() is set, then the page is not freeable */
485 if (PagePrivate(page))
486 return 0;
487 return nfs_fscache_release_page(page, gfp);
488}
489
490static void nfs_check_dirty_writeback(struct page *page,
491 bool *dirty, bool *writeback)
492{
493 struct nfs_inode *nfsi;
494 struct address_space *mapping = page_file_mapping(page);
495
496 if (!mapping || PageSwapCache(page))
497 return;
498
499 /*
500 * Check if an unstable page is currently being committed and
501 * if so, have the VM treat it as if the page is under writeback
502 * so it will not block due to pages that will shortly be freeable.
503 */
504 nfsi = NFS_I(mapping->host);
505 if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) {
506 *writeback = true;
507 return;
508 }
509
510 /*
511 * If PagePrivate() is set, then the page is not freeable and as the
512 * inode is not being committed, it's not going to be cleaned in the
513 * near future so treat it as dirty
514 */
515 if (PagePrivate(page))
516 *dirty = true;
517}
518
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_file_mapping(page)->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
539#ifdef CONFIG_NFS_SWAP
540static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
541 sector_t *span)
542{
543 *span = sis->pages;
544 return xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 1);
545}
546
547static void nfs_swap_deactivate(struct file *file)
548{
549 xs_swapper(NFS_CLIENT(file->f_mapping->host)->cl_xprt, 0);
550}
551#endif
552
553const struct address_space_operations nfs_file_aops = {
554 .readpage = nfs_readpage,
555 .readpages = nfs_readpages,
556 .set_page_dirty = __set_page_dirty_nobuffers,
557 .writepage = nfs_writepage,
558 .writepages = nfs_writepages,
559 .write_begin = nfs_write_begin,
560 .write_end = nfs_write_end,
561 .invalidatepage = nfs_invalidate_page,
562 .releasepage = nfs_release_page,
563 .direct_IO = nfs_direct_IO,
564 .migratepage = nfs_migrate_page,
565 .launder_page = nfs_launder_page,
566 .is_dirty_writeback = nfs_check_dirty_writeback,
567 .error_remove_page = generic_error_remove_page,
568#ifdef CONFIG_NFS_SWAP
569 .swap_activate = nfs_swap_activate,
570 .swap_deactivate = nfs_swap_deactivate,
571#endif
572};
573
574/*
575 * Notification that a PTE pointing to an NFS page is about to be made
576 * writable, implying that someone is about to modify the page through a
577 * shared-writable mapping
578 */
579static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
580{
581 struct page *page = vmf->page;
582 struct file *filp = vma->vm_file;
583 struct inode *inode = file_inode(filp);
584 unsigned pagelen;
585 int ret = VM_FAULT_NOPAGE;
586 struct address_space *mapping;
587
588 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
589 filp, filp->f_mapping->host->i_ino,
590 (long long)page_offset(page));
591
592 /* make sure the cache has finished storing the page */
593 nfs_fscache_wait_on_page_write(NFS_I(inode), page);
594
595 lock_page(page);
596 mapping = page_file_mapping(page);
597 if (mapping != inode->i_mapping)
598 goto out_unlock;
599
600 wait_on_page_writeback(page);
601
602 pagelen = nfs_page_length(page);
603 if (pagelen == 0)
604 goto out_unlock;
605
606 ret = VM_FAULT_LOCKED;
607 if (nfs_flush_incompatible(filp, page) == 0 &&
608 nfs_updatepage(filp, page, 0, pagelen) == 0)
609 goto out;
610
611 ret = VM_FAULT_SIGBUS;
612out_unlock:
613 unlock_page(page);
614out:
615 return ret;
616}
617
618static const struct vm_operations_struct nfs_file_vm_ops = {
619 .fault = filemap_fault,
620 .map_pages = filemap_map_pages,
621 .page_mkwrite = nfs_vm_page_mkwrite,
622 .remap_pages = generic_file_remap_pages,
623};
624
625static int nfs_need_sync_write(struct file *filp, struct inode *inode)
626{
627 struct nfs_open_context *ctx;
628
629 if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC))
630 return 1;
631 ctx = nfs_file_open_context(filp);
632 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
633 nfs_ctx_key_to_expire(ctx))
634 return 1;
635 return 0;
636}
637
638ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
639 unsigned long nr_segs, loff_t pos)
640{
641 struct file *file = iocb->ki_filp;
642 struct inode *inode = file_inode(file);
643 unsigned long written = 0;
644 ssize_t result;
645 size_t count = iov_length(iov, nr_segs);
646
647 result = nfs_key_timeout_notify(file, inode);
648 if (result)
649 return result;
650
651 if (file->f_flags & O_DIRECT)
652 return nfs_file_direct_write(iocb, iov, nr_segs, pos, true);
653
654 dprintk("NFS: write(%pD2, %lu@%Ld)\n",
655 file, (unsigned long) count, (long long) pos);
656
657 result = -EBUSY;
658 if (IS_SWAPFILE(inode))
659 goto out_swapfile;
660 /*
661 * O_APPEND implies that we must revalidate the file length.
662 */
663 if (file->f_flags & O_APPEND) {
664 result = nfs_revalidate_file_size(inode, file);
665 if (result)
666 goto out;
667 }
668
669 result = count;
670 if (!count)
671 goto out;
672
673 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
674 if (result > 0)
675 written = result;
676
677 /* Return error values for O_DSYNC and IS_SYNC() */
678 if (result >= 0 && nfs_need_sync_write(file, inode)) {
679 int err = vfs_fsync(file, 0);
680 if (err < 0)
681 result = err;
682 }
683 if (result > 0)
684 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
685out:
686 return result;
687
688out_swapfile:
689 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
690 goto out;
691}
692EXPORT_SYMBOL_GPL(nfs_file_write);
693
694ssize_t nfs_file_splice_write(struct pipe_inode_info *pipe,
695 struct file *filp, loff_t *ppos,
696 size_t count, unsigned int flags)
697{
698 struct inode *inode = file_inode(filp);
699 unsigned long written = 0;
700 ssize_t ret;
701
702 dprintk("NFS splice_write(%pD2, %lu@%llu)\n",
703 filp, (unsigned long) count, (unsigned long long) *ppos);
704
705 /*
706 * The combination of splice and an O_APPEND destination is disallowed.
707 */
708
709 ret = generic_file_splice_write(pipe, filp, ppos, count, flags);
710 if (ret > 0)
711 written = ret;
712
713 if (ret >= 0 && nfs_need_sync_write(filp, inode)) {
714 int err = vfs_fsync(filp, 0);
715 if (err < 0)
716 ret = err;
717 }
718 if (ret > 0)
719 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
720 return ret;
721}
722EXPORT_SYMBOL_GPL(nfs_file_splice_write);
723
724static int
725do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
726{
727 struct inode *inode = filp->f_mapping->host;
728 int status = 0;
729 unsigned int saved_type = fl->fl_type;
730
731 /* Try local locking first */
732 posix_test_lock(filp, fl);
733 if (fl->fl_type != F_UNLCK) {
734 /* found a conflict */
735 goto out;
736 }
737 fl->fl_type = saved_type;
738
739 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
740 goto out_noconflict;
741
742 if (is_local)
743 goto out_noconflict;
744
745 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
746out:
747 return status;
748out_noconflict:
749 fl->fl_type = F_UNLCK;
750 goto out;
751}
752
753static int do_vfs_lock(struct file *file, struct file_lock *fl)
754{
755 int res = 0;
756 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
757 case FL_POSIX:
758 res = posix_lock_file_wait(file, fl);
759 break;
760 case FL_FLOCK:
761 res = flock_lock_file_wait(file, fl);
762 break;
763 default:
764 BUG();
765 }
766 return res;
767}
768
769static int
770do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
771{
772 struct inode *inode = filp->f_mapping->host;
773 struct nfs_lock_context *l_ctx;
774 int status;
775
776 /*
777 * Flush all pending writes before doing anything
778 * with locks..
779 */
780 nfs_sync_mapping(filp->f_mapping);
781
782 l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
783 if (!IS_ERR(l_ctx)) {
784 status = nfs_iocounter_wait(&l_ctx->io_count);
785 nfs_put_lock_context(l_ctx);
786 if (status < 0)
787 return status;
788 }
789
790 /* NOTE: special case
791 * If we're signalled while cleaning up locks on process exit, we
792 * still need to complete the unlock.
793 */
794 /*
795 * Use local locking if mounted with "-onolock" or with appropriate
796 * "-olocal_lock="
797 */
798 if (!is_local)
799 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
800 else
801 status = do_vfs_lock(filp, fl);
802 return status;
803}
804
805static int
806is_time_granular(struct timespec *ts) {
807 return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000));
808}
809
810static int
811do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
812{
813 struct inode *inode = filp->f_mapping->host;
814 int status;
815
816 /*
817 * Flush all pending writes before doing anything
818 * with locks..
819 */
820 status = nfs_sync_mapping(filp->f_mapping);
821 if (status != 0)
822 goto out;
823
824 /*
825 * Use local locking if mounted with "-onolock" or with appropriate
826 * "-olocal_lock="
827 */
828 if (!is_local)
829 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
830 else
831 status = do_vfs_lock(filp, fl);
832 if (status < 0)
833 goto out;
834
835 /*
836 * Revalidate the cache if the server has time stamps granular
837 * enough to detect subsecond changes. Otherwise, clear the
838 * cache to prevent missing any changes.
839 *
840 * This makes locking act as a cache coherency point.
841 */
842 nfs_sync_mapping(filp->f_mapping);
843 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) {
844 if (is_time_granular(&NFS_SERVER(inode)->time_delta))
845 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
846 else
847 nfs_zap_caches(inode);
848 }
849out:
850 return status;
851}
852
853/*
854 * Lock a (portion of) a file
855 */
856int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
857{
858 struct inode *inode = filp->f_mapping->host;
859 int ret = -ENOLCK;
860 int is_local = 0;
861
862 dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
863 filp, fl->fl_type, fl->fl_flags,
864 (long long)fl->fl_start, (long long)fl->fl_end);
865
866 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
867
868 /* No mandatory locks over NFS */
869 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
870 goto out_err;
871
872 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
873 is_local = 1;
874
875 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
876 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
877 if (ret < 0)
878 goto out_err;
879 }
880
881 if (IS_GETLK(cmd))
882 ret = do_getlk(filp, cmd, fl, is_local);
883 else if (fl->fl_type == F_UNLCK)
884 ret = do_unlk(filp, cmd, fl, is_local);
885 else
886 ret = do_setlk(filp, cmd, fl, is_local);
887out_err:
888 return ret;
889}
890EXPORT_SYMBOL_GPL(nfs_lock);
891
892/*
893 * Lock a (portion of) a file
894 */
895int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
896{
897 struct inode *inode = filp->f_mapping->host;
898 int is_local = 0;
899
900 dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
901 filp, fl->fl_type, fl->fl_flags);
902
903 if (!(fl->fl_flags & FL_FLOCK))
904 return -ENOLCK;
905
906 /*
907 * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
908 * any standard. In principle we might be able to support LOCK_MAND
909 * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
910 * NFS code is not set up for it.
911 */
912 if (fl->fl_type & LOCK_MAND)
913 return -EINVAL;
914
915 if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
916 is_local = 1;
917
918 /* We're simulating flock() locks using posix locks on the server */
919 fl->fl_owner = (fl_owner_t)filp;
920 fl->fl_start = 0;
921 fl->fl_end = OFFSET_MAX;
922
923 if (fl->fl_type == F_UNLCK)
924 return do_unlk(filp, cmd, fl, is_local);
925 return do_setlk(filp, cmd, fl, is_local);
926}
927EXPORT_SYMBOL_GPL(nfs_flock);
928
929/*
930 * There is no protocol support for leases, so we have no way to implement
931 * them correctly in the face of opens by other clients.
932 */
933int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
934{
935 dprintk("NFS: setlease(%pD2, arg=%ld)\n", file, arg);
936 return -EINVAL;
937}
938EXPORT_SYMBOL_GPL(nfs_setlease);
939
940const struct file_operations nfs_file_operations = {
941 .llseek = nfs_file_llseek,
942 .read = do_sync_read,
943 .write = do_sync_write,
944 .aio_read = nfs_file_read,
945 .aio_write = nfs_file_write,
946 .mmap = nfs_file_mmap,
947 .open = nfs_file_open,
948 .flush = nfs_file_flush,
949 .release = nfs_file_release,
950 .fsync = nfs_file_fsync,
951 .lock = nfs_lock,
952 .flock = nfs_flock,
953 .splice_read = nfs_file_splice_read,
954 .splice_write = nfs_file_splice_write,
955 .check_flags = nfs_check_flags,
956 .setlease = nfs_setlease,
957};
958EXPORT_SYMBOL_GPL(nfs_file_operations);