Merge git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6/sactl.git] / fs / nfs / file.c
blob90f292b520d25eec248a9728b7d61cf20cd061f9
1 /*
2 * linux/fs/nfs/file.c
4 * Copyright (C) 1992 Rick Sladkey
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.
12 * Expire cache on write to a file by Wai S Kok (Oct 1994).
14 * Total rewrite of read side for new NFS buffer cache.. Linus.
16 * nfs regular file handling functions
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/slab.h>
28 #include <linux/pagemap.h>
29 #include <linux/smp_lock.h>
30 #include <linux/aio.h>
32 #include <asm/uaccess.h>
33 #include <asm/system.h>
35 #include "delegation.h"
36 #include "internal.h"
37 #include "iostat.h"
39 #define NFSDBG_FACILITY NFSDBG_FILE
41 static int nfs_file_open(struct inode *, struct file *);
42 static int nfs_file_release(struct inode *, struct file *);
43 static loff_t nfs_file_llseek(struct file *file, loff_t offset, int origin);
44 static int nfs_file_mmap(struct file *, struct vm_area_struct *);
45 static ssize_t nfs_file_splice_read(struct file *filp, loff_t *ppos,
46 struct pipe_inode_info *pipe,
47 size_t count, unsigned int flags);
48 static ssize_t nfs_file_read(struct kiocb *, const struct iovec *iov,
49 unsigned long nr_segs, loff_t pos);
50 static ssize_t nfs_file_write(struct kiocb *, const struct iovec *iov,
51 unsigned long nr_segs, loff_t pos);
52 static int nfs_file_flush(struct file *, fl_owner_t id);
53 static int nfs_file_fsync(struct file *, struct dentry *dentry, int datasync);
54 static int nfs_check_flags(int flags);
55 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl);
56 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl);
57 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl);
59 static struct vm_operations_struct nfs_file_vm_ops;
61 const struct file_operations nfs_file_operations = {
62 .llseek = nfs_file_llseek,
63 .read = do_sync_read,
64 .write = do_sync_write,
65 .aio_read = nfs_file_read,
66 .aio_write = nfs_file_write,
67 #ifdef CONFIG_MMU
68 .mmap = nfs_file_mmap,
69 #else
70 .mmap = generic_file_mmap,
71 #endif
72 .open = nfs_file_open,
73 .flush = nfs_file_flush,
74 .release = nfs_file_release,
75 .fsync = nfs_file_fsync,
76 .lock = nfs_lock,
77 .flock = nfs_flock,
78 .splice_read = nfs_file_splice_read,
79 .check_flags = nfs_check_flags,
80 .setlease = nfs_setlease,
83 const struct inode_operations nfs_file_inode_operations = {
84 .permission = nfs_permission,
85 .getattr = nfs_getattr,
86 .setattr = nfs_setattr,
89 #ifdef CONFIG_NFS_V3
90 const struct inode_operations nfs3_file_inode_operations = {
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
94 .listxattr = nfs3_listxattr,
95 .getxattr = nfs3_getxattr,
96 .setxattr = nfs3_setxattr,
97 .removexattr = nfs3_removexattr,
99 #endif /* CONFIG_NFS_v3 */
101 /* Hack for future NFS swap support */
102 #ifndef IS_SWAPFILE
103 # define IS_SWAPFILE(inode) (0)
104 #endif
106 static int nfs_check_flags(int flags)
108 if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
109 return -EINVAL;
111 return 0;
115 * Open file
117 static int
118 nfs_file_open(struct inode *inode, struct file *filp)
120 int res;
122 dprintk("NFS: open file(%s/%s)\n",
123 filp->f_path.dentry->d_parent->d_name.name,
124 filp->f_path.dentry->d_name.name);
126 res = nfs_check_flags(filp->f_flags);
127 if (res)
128 return res;
130 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
131 res = nfs_open(inode, filp);
132 return res;
135 static int
136 nfs_file_release(struct inode *inode, struct file *filp)
138 struct dentry *dentry = filp->f_path.dentry;
140 dprintk("NFS: release(%s/%s)\n",
141 dentry->d_parent->d_name.name,
142 dentry->d_name.name);
144 /* Ensure that dirty pages are flushed out with the right creds */
145 if (filp->f_mode & FMODE_WRITE)
146 nfs_wb_all(dentry->d_inode);
147 nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
148 return nfs_release(inode, filp);
152 * nfs_revalidate_size - Revalidate the file size
153 * @inode - pointer to inode struct
154 * @file - pointer to struct file
156 * Revalidates the file length. This is basically a wrapper around
157 * nfs_revalidate_inode() that takes into account the fact that we may
158 * have cached writes (in which case we don't care about the server's
159 * idea of what the file length is), or O_DIRECT (in which case we
160 * shouldn't trust the cache).
162 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
164 struct nfs_server *server = NFS_SERVER(inode);
165 struct nfs_inode *nfsi = NFS_I(inode);
167 if (server->flags & NFS_MOUNT_NOAC)
168 goto force_reval;
169 if (filp->f_flags & O_DIRECT)
170 goto force_reval;
171 if (nfsi->npages != 0)
172 return 0;
173 if (!(nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) && !nfs_attribute_timeout(inode))
174 return 0;
175 force_reval:
176 return __nfs_revalidate_inode(server, inode);
179 static loff_t nfs_file_llseek(struct file *filp, loff_t offset, int origin)
181 loff_t loff;
183 dprintk("NFS: llseek file(%s/%s, %lld, %d)\n",
184 filp->f_path.dentry->d_parent->d_name.name,
185 filp->f_path.dentry->d_name.name,
186 offset, origin);
188 /* origin == SEEK_END => we must revalidate the cached file length */
189 if (origin == SEEK_END) {
190 struct inode *inode = filp->f_mapping->host;
192 int retval = nfs_revalidate_file_size(inode, filp);
193 if (retval < 0)
194 return (loff_t)retval;
196 spin_lock(&inode->i_lock);
197 loff = generic_file_llseek_unlocked(filp, offset, origin);
198 spin_unlock(&inode->i_lock);
199 } else
200 loff = generic_file_llseek_unlocked(filp, offset, origin);
201 return loff;
205 * Helper for nfs_file_flush() and nfs_file_fsync()
207 * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
208 * disk, but it retrieves and clears ctx->error after synching, despite
209 * the two being set at the same time in nfs_context_set_write_error().
210 * This is because the former is used to notify the _next_ call to
211 * nfs_file_write() that a write error occured, and hence cause it to
212 * fall back to doing a synchronous write.
214 static int nfs_do_fsync(struct nfs_open_context *ctx, struct inode *inode)
216 int have_error, status;
217 int ret = 0;
219 have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
220 status = nfs_wb_all(inode);
221 have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
222 if (have_error)
223 ret = xchg(&ctx->error, 0);
224 if (!ret)
225 ret = status;
226 return ret;
230 * Flush all dirty pages, and check for write errors.
232 static int
233 nfs_file_flush(struct file *file, fl_owner_t id)
235 struct nfs_open_context *ctx = nfs_file_open_context(file);
236 struct dentry *dentry = file->f_path.dentry;
237 struct inode *inode = dentry->d_inode;
238 int status;
240 dprintk("NFS: flush(%s/%s)\n",
241 dentry->d_parent->d_name.name,
242 dentry->d_name.name);
244 if ((file->f_mode & FMODE_WRITE) == 0)
245 return 0;
246 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
248 /* Ensure that data+attribute caches are up to date after close() */
249 status = nfs_do_fsync(ctx, inode);
250 if (!status)
251 nfs_revalidate_inode(NFS_SERVER(inode), inode);
252 return status;
255 static ssize_t
256 nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
257 unsigned long nr_segs, loff_t pos)
259 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
260 struct inode * inode = dentry->d_inode;
261 ssize_t result;
262 size_t count = iov_length(iov, nr_segs);
264 if (iocb->ki_filp->f_flags & O_DIRECT)
265 return nfs_file_direct_read(iocb, iov, nr_segs, pos);
267 dprintk("NFS: read(%s/%s, %lu@%lu)\n",
268 dentry->d_parent->d_name.name, dentry->d_name.name,
269 (unsigned long) count, (unsigned long) pos);
271 result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
272 nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
273 if (!result)
274 result = generic_file_aio_read(iocb, iov, nr_segs, pos);
275 return result;
278 static ssize_t
279 nfs_file_splice_read(struct file *filp, loff_t *ppos,
280 struct pipe_inode_info *pipe, size_t count,
281 unsigned int flags)
283 struct dentry *dentry = filp->f_path.dentry;
284 struct inode *inode = dentry->d_inode;
285 ssize_t res;
287 dprintk("NFS: splice_read(%s/%s, %lu@%Lu)\n",
288 dentry->d_parent->d_name.name, dentry->d_name.name,
289 (unsigned long) count, (unsigned long long) *ppos);
291 res = nfs_revalidate_mapping(inode, filp->f_mapping);
292 if (!res)
293 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
294 return res;
297 static int
298 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
300 struct dentry *dentry = file->f_path.dentry;
301 struct inode *inode = dentry->d_inode;
302 int status;
304 dprintk("NFS: mmap(%s/%s)\n",
305 dentry->d_parent->d_name.name, dentry->d_name.name);
307 status = nfs_revalidate_mapping(inode, file->f_mapping);
308 if (!status) {
309 vma->vm_ops = &nfs_file_vm_ops;
310 vma->vm_flags |= VM_CAN_NONLINEAR;
311 file_accessed(file);
313 return status;
317 * Flush any dirty pages for this process, and check for write errors.
318 * The return status from this call provides a reliable indication of
319 * whether any write errors occurred for this process.
321 static int
322 nfs_file_fsync(struct file *file, struct dentry *dentry, int datasync)
324 struct nfs_open_context *ctx = nfs_file_open_context(file);
325 struct inode *inode = dentry->d_inode;
327 dprintk("NFS: fsync file(%s/%s) datasync %d\n",
328 dentry->d_parent->d_name.name, dentry->d_name.name,
329 datasync);
331 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
332 return nfs_do_fsync(ctx, inode);
336 * This does the "real" work of the write. We must allocate and lock the
337 * page to be sent back to the generic routine, which then copies the
338 * data from user space.
340 * If the writer ends up delaying the write, the writer needs to
341 * increment the page use counts until he is done with the page.
343 static int nfs_write_begin(struct file *file, struct address_space *mapping,
344 loff_t pos, unsigned len, unsigned flags,
345 struct page **pagep, void **fsdata)
347 int ret;
348 pgoff_t index;
349 struct page *page;
350 index = pos >> PAGE_CACHE_SHIFT;
352 dfprintk(PAGECACHE, "NFS: write_begin(%s/%s(%ld), %u@%lld)\n",
353 file->f_path.dentry->d_parent->d_name.name,
354 file->f_path.dentry->d_name.name,
355 mapping->host->i_ino, len, (long long) pos);
357 page = grab_cache_page_write_begin(mapping, index, flags);
358 if (!page)
359 return -ENOMEM;
360 *pagep = page;
362 ret = nfs_flush_incompatible(file, page);
363 if (ret) {
364 unlock_page(page);
365 page_cache_release(page);
367 return ret;
370 static int nfs_write_end(struct file *file, struct address_space *mapping,
371 loff_t pos, unsigned len, unsigned copied,
372 struct page *page, void *fsdata)
374 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
375 int status;
377 dfprintk(PAGECACHE, "NFS: write_end(%s/%s(%ld), %u@%lld)\n",
378 file->f_path.dentry->d_parent->d_name.name,
379 file->f_path.dentry->d_name.name,
380 mapping->host->i_ino, len, (long long) pos);
383 * Zero any uninitialised parts of the page, and then mark the page
384 * as up to date if it turns out that we're extending the file.
386 if (!PageUptodate(page)) {
387 unsigned pglen = nfs_page_length(page);
388 unsigned end = offset + len;
390 if (pglen == 0) {
391 zero_user_segments(page, 0, offset,
392 end, PAGE_CACHE_SIZE);
393 SetPageUptodate(page);
394 } else if (end >= pglen) {
395 zero_user_segment(page, end, PAGE_CACHE_SIZE);
396 if (offset == 0)
397 SetPageUptodate(page);
398 } else
399 zero_user_segment(page, pglen, PAGE_CACHE_SIZE);
402 status = nfs_updatepage(file, page, offset, copied);
404 unlock_page(page);
405 page_cache_release(page);
407 if (status < 0)
408 return status;
409 return copied;
412 static void nfs_invalidate_page(struct page *page, unsigned long offset)
414 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
416 if (offset != 0)
417 return;
418 /* Cancel any unstarted writes on this page */
419 nfs_wb_page_cancel(page->mapping->host, page);
422 static int nfs_release_page(struct page *page, gfp_t gfp)
424 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
426 /* If PagePrivate() is set, then the page is not freeable */
427 return 0;
430 static int nfs_launder_page(struct page *page)
432 struct inode *inode = page->mapping->host;
434 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
435 inode->i_ino, (long long)page_offset(page));
437 return nfs_wb_page(inode, page);
440 const struct address_space_operations nfs_file_aops = {
441 .readpage = nfs_readpage,
442 .readpages = nfs_readpages,
443 .set_page_dirty = __set_page_dirty_nobuffers,
444 .writepage = nfs_writepage,
445 .writepages = nfs_writepages,
446 .write_begin = nfs_write_begin,
447 .write_end = nfs_write_end,
448 .invalidatepage = nfs_invalidate_page,
449 .releasepage = nfs_release_page,
450 .direct_IO = nfs_direct_IO,
451 .launder_page = nfs_launder_page,
454 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct page *page)
456 struct file *filp = vma->vm_file;
457 struct dentry *dentry = filp->f_path.dentry;
458 unsigned pagelen;
459 int ret = -EINVAL;
460 struct address_space *mapping;
462 dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%s/%s(%ld), offset %lld)\n",
463 dentry->d_parent->d_name.name, dentry->d_name.name,
464 filp->f_mapping->host->i_ino,
465 (long long)page_offset(page));
467 lock_page(page);
468 mapping = page->mapping;
469 if (mapping != dentry->d_inode->i_mapping)
470 goto out_unlock;
472 ret = 0;
473 pagelen = nfs_page_length(page);
474 if (pagelen == 0)
475 goto out_unlock;
477 ret = nfs_flush_incompatible(filp, page);
478 if (ret != 0)
479 goto out_unlock;
481 ret = nfs_updatepage(filp, page, 0, pagelen);
482 if (ret == 0)
483 ret = pagelen;
484 out_unlock:
485 unlock_page(page);
486 return ret;
489 static struct vm_operations_struct nfs_file_vm_ops = {
490 .fault = filemap_fault,
491 .page_mkwrite = nfs_vm_page_mkwrite,
494 static int nfs_need_sync_write(struct file *filp, struct inode *inode)
496 struct nfs_open_context *ctx;
498 if (IS_SYNC(inode) || (filp->f_flags & O_SYNC))
499 return 1;
500 ctx = nfs_file_open_context(filp);
501 if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags))
502 return 1;
503 return 0;
506 static ssize_t nfs_file_write(struct kiocb *iocb, const struct iovec *iov,
507 unsigned long nr_segs, loff_t pos)
509 struct dentry * dentry = iocb->ki_filp->f_path.dentry;
510 struct inode * inode = dentry->d_inode;
511 ssize_t result;
512 size_t count = iov_length(iov, nr_segs);
514 if (iocb->ki_filp->f_flags & O_DIRECT)
515 return nfs_file_direct_write(iocb, iov, nr_segs, pos);
517 dprintk("NFS: write(%s/%s, %lu@%Ld)\n",
518 dentry->d_parent->d_name.name, dentry->d_name.name,
519 (unsigned long) count, (long long) pos);
521 result = -EBUSY;
522 if (IS_SWAPFILE(inode))
523 goto out_swapfile;
525 * O_APPEND implies that we must revalidate the file length.
527 if (iocb->ki_filp->f_flags & O_APPEND) {
528 result = nfs_revalidate_file_size(inode, iocb->ki_filp);
529 if (result)
530 goto out;
533 result = count;
534 if (!count)
535 goto out;
537 nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, count);
538 result = generic_file_aio_write(iocb, iov, nr_segs, pos);
539 /* Return error values for O_SYNC and IS_SYNC() */
540 if (result >= 0 && nfs_need_sync_write(iocb->ki_filp, inode)) {
541 int err = nfs_do_fsync(nfs_file_open_context(iocb->ki_filp), inode);
542 if (err < 0)
543 result = err;
545 out:
546 return result;
548 out_swapfile:
549 printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
550 goto out;
553 static int do_getlk(struct file *filp, int cmd, struct file_lock *fl)
555 struct inode *inode = filp->f_mapping->host;
556 int status = 0;
558 lock_kernel();
559 /* Try local locking first */
560 posix_test_lock(filp, fl);
561 if (fl->fl_type != F_UNLCK) {
562 /* found a conflict */
563 goto out;
566 if (nfs_have_delegation(inode, FMODE_READ))
567 goto out_noconflict;
569 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM)
570 goto out_noconflict;
572 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
573 out:
574 unlock_kernel();
575 return status;
576 out_noconflict:
577 fl->fl_type = F_UNLCK;
578 goto out;
581 static int do_vfs_lock(struct file *file, struct file_lock *fl)
583 int res = 0;
584 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
585 case FL_POSIX:
586 res = posix_lock_file_wait(file, fl);
587 break;
588 case FL_FLOCK:
589 res = flock_lock_file_wait(file, fl);
590 break;
591 default:
592 BUG();
594 if (res < 0)
595 dprintk(KERN_WARNING "%s: VFS is out of sync with lock manager"
596 " - error %d!\n",
597 __func__, res);
598 return res;
601 static int do_unlk(struct file *filp, int cmd, struct file_lock *fl)
603 struct inode *inode = filp->f_mapping->host;
604 int status;
607 * Flush all pending writes before doing anything
608 * with locks..
610 nfs_sync_mapping(filp->f_mapping);
612 /* NOTE: special case
613 * If we're signalled while cleaning up locks on process exit, we
614 * still need to complete the unlock.
616 lock_kernel();
617 /* Use local locking if mounted with "-onolock" */
618 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
619 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
620 else
621 status = do_vfs_lock(filp, fl);
622 unlock_kernel();
623 return status;
626 static int do_setlk(struct file *filp, int cmd, struct file_lock *fl)
628 struct inode *inode = filp->f_mapping->host;
629 int status;
632 * Flush all pending writes before doing anything
633 * with locks..
635 status = nfs_sync_mapping(filp->f_mapping);
636 if (status != 0)
637 goto out;
639 lock_kernel();
640 /* Use local locking if mounted with "-onolock" */
641 if (!(NFS_SERVER(inode)->flags & NFS_MOUNT_NONLM))
642 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
643 else
644 status = do_vfs_lock(filp, fl);
645 unlock_kernel();
646 if (status < 0)
647 goto out;
649 * Make sure we clear the cache whenever we try to get the lock.
650 * This makes locking act as a cache coherency point.
652 nfs_sync_mapping(filp->f_mapping);
653 if (!nfs_have_delegation(inode, FMODE_READ))
654 nfs_zap_caches(inode);
655 out:
656 return status;
660 * Lock a (portion of) a file
662 static int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
664 struct inode *inode = filp->f_mapping->host;
665 int ret = -ENOLCK;
667 dprintk("NFS: lock(%s/%s, t=%x, fl=%x, r=%lld:%lld)\n",
668 filp->f_path.dentry->d_parent->d_name.name,
669 filp->f_path.dentry->d_name.name,
670 fl->fl_type, fl->fl_flags,
671 (long long)fl->fl_start, (long long)fl->fl_end);
673 nfs_inc_stats(inode, NFSIOS_VFSLOCK);
675 /* No mandatory locks over NFS */
676 if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
677 goto out_err;
679 if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
680 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
681 if (ret < 0)
682 goto out_err;
685 if (IS_GETLK(cmd))
686 ret = do_getlk(filp, cmd, fl);
687 else if (fl->fl_type == F_UNLCK)
688 ret = do_unlk(filp, cmd, fl);
689 else
690 ret = do_setlk(filp, cmd, fl);
691 out_err:
692 return ret;
696 * Lock a (portion of) a file
698 static int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
700 dprintk("NFS: flock(%s/%s, t=%x, fl=%x)\n",
701 filp->f_path.dentry->d_parent->d_name.name,
702 filp->f_path.dentry->d_name.name,
703 fl->fl_type, fl->fl_flags);
705 if (!(fl->fl_flags & FL_FLOCK))
706 return -ENOLCK;
708 /* We're simulating flock() locks using posix locks on the server */
709 fl->fl_owner = (fl_owner_t)filp;
710 fl->fl_start = 0;
711 fl->fl_end = OFFSET_MAX;
713 if (fl->fl_type == F_UNLCK)
714 return do_unlk(filp, cmd, fl);
715 return do_setlk(filp, cmd, fl);
719 * There is no protocol support for leases, so we have no way to implement
720 * them correctly in the face of opens by other clients.
722 static int nfs_setlease(struct file *file, long arg, struct file_lock **fl)
724 dprintk("NFS: setlease(%s/%s, arg=%ld)\n",
725 file->f_path.dentry->d_parent->d_name.name,
726 file->f_path.dentry->d_name.name, arg);
728 return -EINVAL;